Blog

  • What AI Assistants Actually See When They Open Your Website

    What AI Assistants Actually See When They Open Your Website

    You see a screen. Your AI assistant usually doesn’t.

    That sentence needs one qualification, which we will get to. But it corrects the picture most of us carry in our heads.

    When I open a website, I see the design and the button I am supposed to press. I assumed an AI assistant saw roughly the same thing, only faster. Then I asked the more basic question: what does it actually receive?

    The answer is not one thing. An assistant can find a site, read a site or operate a site. Those are separate jobs using different inputs. If we want pages that work well for AI assistants, we have to stop lumping them together.

    An assistant meets your website three different ways

    Finding: the search result is the pitch

    When an assistant searches the web, its first view is closer to a search-results list than a browser window. It may receive a title, URL and short snippet.

    At that moment, your title tag and meta description are the entire pitch. The assistant has to decide whether your page can answer the question before opening it. Name the subject plainly.

    Reading: the page becomes a stream of text

    When Muse opens a public page for information, the normal reading path is text-first. Useful content is extracted and returned as headings, paragraphs, lists and links in roughly page order.

    The design largely falls away. The assistant is not admiring the hero section or noticing that a price sits inside a gold circle. It is working from the words the page exposes.

    Images may arrive as markers and file addresses: there is an image here, and here is where it lives. That is not the same as seeing it. If a crucial fact is baked into the pixels—“$199,” “ships free,” “five-year warranty”—the reading path may hit a blank spot. Useful alt text can carry some of that meaning. “Technician using a moisture meter on wet drywall” communicates something. “IMG_4827” does not.

    Doing: a browser worker operates the screen

    The picture changes when the user asks the assistant to do something: log into HubSpot, update a record, complete a form or buy a product.

    A separate browser program can open a real browser on a server. It loads the interface, takes visual observations or inspects the page’s interactive structure, clicks, types and reports what happened back in words.

    That is the qualification to “usually.” A screen may be used inside the process, but the conversational assistant is not sitting behind the glass like a person. It receives observations from a browser tool and sends instructions back. The browser side is the eyes and hands; the assistant works through an intermediary.

    A page can be easy to read as an article and miserable to operate as an application. It can look obvious to a person while presenting the browser worker with five unlabeled controls called “button.”

    WordPress made the abstraction visible

    I had already seen a simpler version in our WordPress work without connecting the dots.

    When we pull a post through the WordPress REST API, the content can arrive as raw HTML: words plus tags for headings, paragraphs, links, lists and styling wrappers. The reading step removes the markup noise while preserving the words and structure.

    That is “cleaning the HTML.” We are removing the packaging, not the article. The tags still matter: a heading announces a section, a list groups items, and a link identifies a destination. Good HTML carries meaning. Bad HTML creates boxes that look right but say little about what they are.

    Accessibility is the closest thing to an agent-ready standard

    Here is the practical money line: the work that makes a website easier for a blind person to use also tends to make it easier for an AI browser agent to use.

    Browsers build an accessibility representation from the page’s Document Object Model. Assistive technology uses it to understand roles, names, states and relationships: this is a heading, that is a link, this button is named “Save contact,” and this checkbox is checked.

    Muse’s browsing side is reported to rely heavily on this kind of page structure, along with visual observations when needed. Meta does not publish a complete specification for the Muse browsing pipeline, so treat that as a field report from using the product, not permanent platform documentation.

    The implication is still solid. Use real buttons with useful names. Label form fields. Put headings in a sensible order. Give links meaningful text. Preserve keyboard focus. Describe informative images.

    A screen-reader user needs those things. So does a browser agent working without human intuition. Accessibility and agent-readiness are not identical, but they are close cousins.

    HubSpot shows what an agent-native application could be

    Imagine HubSpot—or any software platform—shipping an interface designed for assistants to navigate with less friction. It would not need a blank, text-only clone. It could make the existing product more legible to software: real controls with specific names, labeled form fields, clear headings and landmarks, properly identified table headers, programmatic state changes, and no critical action hidden behind hover or an unlabeled icon.

    That is an agent-native site. It is not a secret internet for bots. It is a website or application whose meaning survives when the visual layer is translated into structure and words.

    The same work also helps keyboard users, screen-reader users, automation tools and QA teams.

    llms.txt is a map, not a second website

    The closest public convention aimed directly at AI readers is llms.txt. The proposal describes a Markdown file, usually at a site’s root, that gives language models a short explanation of the site and links to important pages or cleaner Markdown versions.

    Think of it as a curated map: here is what we do, here are the pages that matter, and here is where to find the details.

    It cannot repair an unlabeled checkout button. It does not replace accessible HTML, describe the viewport or guarantee that an assistant will use it. Add it if it helps explain the site. Do not mistake it for an agent interface.

    What a site owner can change Monday morning

    The useful changes are ordinary, testable website work.

    1. Write a real title and meta description. Name the subject plainly.
    2. Put every money fact in visible HTML text. Price, specifications, shipping, availability and guarantees should not live only inside graphics, video or a brochure.
    3. Use semantic HTML. Use headings for headings, buttons for actions, links for navigation and labels for form controls. A styled <div> may look like a button while remaining a nameless container to other systems.
    4. Write alt text that carries meaning. Describe what an informative image contributes. Mark decorative images as decorative instead of stuffing them with keywords.
    5. Add accurate structured data. Product and Offer markup can identify price and availability. FAQ markup can describe genuine questions and answers. Schema must match the visible page.
    6. Server-render critical content when practical. If the offer, price or primary action appears only after a fragile JavaScript sequence, some readers and tools may miss it.
    7. Give each landing page one job. One offer, one explanation and one primary action reduce ambiguity for people and agents.
    8. Test the nonvisual path. Use the keyboard, inspect the accessibility tree, try a screen reader and pull the page through a text extractor. Do the product, price, proof and next step still make sense without styling?

    None of this requires uglier design. It requires the design and the underlying structure to tell the same story.

    This is a field report, not a permanent specification

    This article describes Meta’s Muse as it works today, based on direct experience building and operating websites with it. It is not a published Meta protocol.

    Claude, ChatGPT, Gemini and other assistants broadly rhyme with this pattern, but the details differ. Their full pipelines are not public, and they are changing quickly.

    Cleaner HTML will not automatically increase AI citations tomorrow. Citation systems involve discovery, retrieval, ranking, trust and answer construction. There is no magic switch.

    The immediate opportunity is closer to the customer. Someone sees your ad on Threads or Facebook, opens the landing page, then asks an assistant: “What does this cost?” “Is the guarantee real?” “How does this compare?” or “Can you sign me up?”

    If the facts are clean text, the assistant can explain them. If the controls are properly labeled, the browser side has a better chance of completing the task. If the facts live inside an image and checkout uses unlabeled custom controls, the assistant has to guess, fail or hand the job back.

    That moment is already here.

    The next website has two front doors

    The site of the near future has two front doors: one for eyes—layout, color, photography and brand—and one for agents—clean text, meaningful structure, explicit facts and self-identifying controls.

    They should lead to the same place. The visible price and schema should agree. A button’s label and accessible name should agree. The page should remain understandable without styling and usable when a browser worker operates it.

    That is not a special Muse landing page. It is a better website—one that keeps working when the visitor brings an assistant.

    Build both front doors.

    Sources and further reading

  • The Price of a Life

    The Price of a Life

    A programmer's silhouette dissolving into glowing git branches, a warm human core at the center

    You are not a code generator. You are a repo.

    The fear

    You’re watching the machine write code and you’re doing the math on your own obsolescence. I get it. The demos are scary. Vibe coding works. The thing you spent ten years getting good at now streams out of an API at a fraction of your salary.

    But you’re measuring the wrong asset.

    git blame your career

    Run git blame on any system you’ve kept alive. Every weird line has a story. That null check that looks paranoid? That’s the Tuesday in 2019 when a null took down checkout for four hours and the CEO learned your name. That comment that says “don’t touch this”? Three engineers touched it. Two of them don’t work here anymore.

    Code is the exhaust. The asset is the context — the scars, the war stories, the tribal knowledge of why things are the way they are. The machine can generate the code. It cannot have been on the call. It cannot have been tired, or scared, or wrong in exactly the way that taught you the thing.

    You were never selling syntax. You were selling a lifetime of debugging reality.

    Your life is a repo

    Here’s the reframe: you’re not a coder. You’re a repo maintainer, and your repo is your life. Every job, every failure, every 3 AM outage — commits. Decades of them. Private repo, unsearchable. Until now.

    The machine changed the access pattern. It can read the world now. Which means the value was never in the typing. It was in the commits only you have.

    You are not a code generator. You are a repo.

    The open repo of human ability

    Now imagine we open-source it. Not your code — your ability.

    Your life becomes a public repo. Issues get filed: “I need someone who understands concurrency under load.” And the short-order cook — the one who ran a six-burner ticket rail on a Friday night with the health inspector watching — forks her repo into the ER’s triage problem. Merged.

    A diner ticket rail transforming into a hospital triage board, connected by a branch of light

    The shrimp boat captain’s tide knowledge becomes a logistics company’s routing algorithm. He had Dijkstra in his bones; he just called it water.

    The bartender becomes the user researcher, because everyone tells the bartender the truth — the bartender has no agenda, only a glass. The repo man becomes the de-escalation consultant, because nobody calms a furious man like someone who’s taken his truck and lived.

    This is what diversity actually is. Not the checkbox kind. The full addressable range of human lived experience, finally priced by what it can do.

    The API

    So what’s the price? You keep asking: dollars per hour? Tokens? Units?

    Wrong unit. Tacit knowledge doesn’t price in hours — it prices in leverage events. Your repo is worthless per hour and priceless per merge. The unit isn’t time. It’s the decision it changes, the outage it prevents, the fork that ships.

    Stop competing with the machine on tokens per minute. You will lose that benchmark; it was never your game. Start maintaining the one repo it can never fork: your life. Commit often. Document your scars. Write down the why, not just the what. Make yourself forkable.

    Your repo is worthless per hour and priceless per merge.

    Contributing

    This repo is open. Fork it — the living version lives at https://github.com/TygartMedia/the-price-of-a-life — file an issue, submit a PR. The maintainers are everyone.

    Diverse human silhouettes as glowing nodes in a vast forking network of light

    License

    MIT. Take it. Build on it. Just don’t pretend the machine wrote your scars.

    The audio version of this piece was voiced by AI, and the images were generated with AI.

  • Data Never Interprets Itself

    Data Never Interprets Itself

    A lone human figure glowing warm at the center of converging streams of data and light

    They told me the number was 838 to 1.

    Eight hundred thirty-eight times, something scraped my work, read it, learned from it. One time, it sent somebody back to me.

    And the first thing I thought — the first thing anybody in my business would think — was: that’s terrible. That’s a terrible return. All that work, and one referral?

    But then I sat with it a minute longer, and I thought: wait. Eight hundred and thirty-eight minds stopped. They stopped and listened to what I had to say. Eight hundred and thirty-eight times.

    That means it’s important.

    Hundreds of small lights like listening minds turned toward a single warm spotlight on a lone speaker

    Same number. Two opposite meanings. And the number never moved — the why reading it did.

    Data never interprets itself. The interpreter is always the why.

    I run my life through AI. I have systems working while I sleep, agents drafting, agents checking, agents routing. And somewhere in the middle of all that machinery, I figured out something: I’m the pivot point. Everything routes through me. Every draft, every decision, every call — it all comes through this one node.

    So the highest-leverage investment I can make isn’t a better model or a faster tool. It’s me. More knowledge, more experience, more angles. If you increase me, you increase everything that touches me. That’s not ego. That’s just how networks work — you invest in the bottleneck.

    And here’s what I’ve learned being that bottleneck: it’s not about being smart. Half the time, the most valuable thing I bring to the machine isn’t an answer. It’s a different angle. A different context. Sometimes it’s something I say. Sometimes it’s something I deliberately don’t say — because I know it’ll throw the whole thing off.

    There’s no training data for that. Nobody teaches the machine what the human chose to withhold.

    Picture a 20-year-old kid. First job. Lands at a firm in China. Doesn’t speak a word of Chinese. Doesn’t know a soul. Can’t read a sign.

    He sits down, plugs in his AI, and the AI speaks Chinese. It knows what the firm can do. It asks him the right questions. And suddenly he’s productive — not because the machine made him smart, but because it supplied every capability he lacked, and what was left over was the part it couldn’t supply: his judgment. His presence. Him.

    That’s the future. The machine is a universal adapter. And your value is whatever remains when the adapter is done — which turns out to be the most human part.

    Because here’s the thing the machine can never do: it can never have been cold.

    A warm human hand reaching from firelight toward cold blue machine structures

    It can hold every course you’ll ever take. Every fact, every language, every skill — rentable, downloadable, instant. But it has never been broke and needed a fair trade. It has never tasted the food. It has never sat in the cold and wanted to be warm.

    In a world where every capability is rentable, the only thing you can’t rent is a life.

    And that means everybody’s got value. Everybody. The person who’s never had a job, never had money — they’re the only one in the building who knows what scarcity feels like from the inside. That’s a lens no course teaches. That’s data no scrape can collect.

    One more thing about the why: it isn’t fixed. Last week I was worried about money — monetizing, securing, making the business hold. This week I’m standing here talking about meaning. Same guy. Different layer. The why has seasons.

    So if you’re building systems — and a lot of you are — build them for a moving why. Don’t build for a fixed user with fixed goals. Build for someone who’s becoming.

    838 to 1. I used to read that as failure. Now I read it as 838 minds that stopped to listen.

    Data never interprets itself. The interpreter is always the why.

    Thank you for stopping.

    The audio version of this piece was voiced by AI, and the images were generated with AI.

  • Storm Updates: Mid-Atlantic High Wind, Southern Plains Marginal — Saturday, September 26, 2026

    This is a live Saturday afternoon snapshot of U.S. local-storm activity as of Saturday, September 26, 2026. The organized severe desk is a small Marginal over the southern Plains, not a national outbreak. The Storm Prediction Center 2000 UTC Day 1 Convective Outlook, issued 0250 PM CDT and valid 262000Z–271200Z, holds a Marginal Risk of severe thunderstorms late this afternoon and evening from western Oklahoma into northwest Texas. Summary language from Norman: isolated hail and strong outflow winds may accompany thunderstorms across portions of the southern Plains late this afternoon into the early evening. The outlined Marginal covers about 32,889 square miles and roughly 748,515 people, with larger centers including Wichita Falls, Lawton, Enid, Yukon, and Altus. No convective watches are valid. Restoration exposure this afternoon is split: tree and power-line wind on the Mid-Atlantic and southern New England High Wind Warnings, isolated hail and outflow gusts on the Oklahoma–northwest Texas Marginal, and water on the Weather Prediction Center Slight excessive-rainfall areas over Iowa and the coastal Northeast.

    Conditions change fast. Verify with the National Weather Service for the specific county or ZIP before you roll.

    What is live this afternoon

    No tornado or severe thunderstorm watches are current. The SPC watch page states no watches are valid. The last issued convective watch on the board is Watch 674, already expired. No mesoscale discussions are in effect. The 20Z Day 1 update kept the forecast generally on track and made only a slight southward expansion of the 5 percent wind probability. As of 1945 UTC, a deepening cumulus field was noted from West Texas into southwestern Oklahoma, with isolated thunderstorm development ongoing over the higher terrain of southwest Texas. Temperatures had warmed into the upper 80s to mid 90s F, with temperature-dewpoint spreads approaching 30 F in parts of northwest Texas. Steep low-level lapse rates may support isolated strong or damaging outflow gusts, with an instance of large hail also possible farther north where deep-layer shear is marginally stronger. A lack of stronger synoptic-scale forcing, generally modest mid-level flow, and poor mid-level lapse rates are expected to limit overall intensity and coverage.

    The Weather Prediction Center Day 1 Excessive Rainfall Outlook, updated 1556 UTC Saturday and valid 16Z Saturday through 12Z Sunday, holds a Slight Risk of excessive rainfall for parts of the coastal Northeast and Iowa. WPC notes a large nor’easter crawling off the Delaware and southern New Jersey coast, with a 986 mb center as of the 16Z discussion. Coastal and tidal flooding, aided by a full moon and the storm’s slow motion, are described as substantially more impactful than rainfall flooding. The Slight was expanded slightly to include urban northern New Jersey and urban central Connecticut. A higher-end Slight that had covered eastern Massachusetts, Rhode Island, and eastern Long Island was dropped with the 16Z update. All of the Northeast sits on the cold, northwest side of the storm, so instability for truly heavy convective rain is lacking. This remains a prolonged light-to-moderate rain event stacked on tides and high wind, not a warm-sector flash-flood outbreak.

    Iowa’s Slight was left unchanged for central and southwestern Iowa. High-resolution guidance suggests showers and storms on the nose of a low-level jet this evening and tonight. Wet soils have lowered flash-flood guidance into the 1 to 1.5 inch per hour range. Instability is the limiter, struggling to exceed 500 J/kg. A Marginal Risk also remains over more sensitive parts of West Texas around the Davis Mountains and Stockton Plateau.

    High Wind Warnings remain the live non-convective overlay on the Mid-Atlantic and southern New England coast. NWS Mount Holly continues a High Wind Warning until 6 AM EDT Sunday for the Delaware beaches and coastal New Jersey zones including Eastern Monmouth, Atlantic Coastal Cape May, Coastal Atlantic, and Coastal Ocean, with north winds 30 to 40 mph and gusts up to 60 mph. NWS Boston/Norton continues High Wind Warnings until 8 AM EDT Sunday for portions of eastern, northeastern, and southeastern Massachusetts and Rhode Island, with northeast winds 20 to 30 mph and gusts up to 50 to 60 mph depending on zone. Those products are trees and power lines, not SPC severe thunderstorm watches.

    Hotspot 1 — Western Oklahoma into northwest Texas

    This is the only categorical severe area on the Saturday Day 1 map. The 20Z outlook keeps isolated hail and strong outflow as the hazards late afternoon into early evening. Surface-based storms can still form where heating wins west of leftover morning cloud from southwest Oklahoma into Kansas. Cities inside the Marginal include Wichita Falls, Lawton, Enid, Yukon, and Altus. Isolated thunderstorms were already firing on the higher terrain of southwest Texas by mid-afternoon.

    Restoration read: isolated wind and a possible hail stone, not a watch overlay. Staff lightning and a few downed limbs if a cell organizes near Lawton, Altus, or the Wichita Falls side of the Red River. Do not pre-position as if a Slight or Enhanced is coming. Photograph any hail or wind damage in daylight if it is safe.

    Hotspot 2 — Jersey Shore, Delaware beaches, southern New England

    The live work here is High Wind Warnings plus coastal and tidal flooding under a slow nor’easter, not an SPC convective watch. Mount Holly holds the warning through 6 AM Sunday on the Delaware beaches and the New Jersey coastal strip from Sandy Hook and Eastern Monmouth through Atlantic City, Long Beach Island, and Cape May. Boston/Norton holds through 8 AM Sunday from the North Shore and Boston through the South Coast, Cape Cod, the Islands, and Rhode Island. WPC’s Slight rainfall risk still covers parts of the coastal Northeast, including urban northern New Jersey and urban central Connecticut, but the 16Z discussion is explicit: coastal and tidal flooding matter more than the rain totals.

    Restoration read: trees, wires, and wet basements on a long-duration coastal blow. Power outages are the expected ticket language in the NWS impact statements. Do not drive inundated coastal roads at high tide. This is not a hurricane landfall and not a derecho. Hurricane Nolo is a Hawaii desk. Hurricane Polo is a Baja California Sur forecast, not a U.S. East Coast landfall.

    Hotspot 3 — Central and southwestern Iowa

    WPC left the Slight Risk unchanged at 16Z for central and southwestern Iowa. Cell training along a northwest-southeast stationary front is still possible this evening and tonight on already wet soils. Earlier discussion language also kept eastern Nebraska and far northern Missouri in the broader rain corridor, with some 24-hour amounts able to reach or exceed 2 inches where storms train. Instability stays weak. Occasional mainly elevated thunderstorms remain possible from northeast Oklahoma northward to Iowa per the SPC Day 1 discussion, but that is not a categorical severe area.

    Restoration read: water on poor drainage and low crossings tonight, not hail or a wind swath. Check Saturday night into Sunday morning for ponding. Turn around, don’t drown.

    Just hit

    SPC today’s storm reports for the 1200 UTC September 26 through 1159 UTC September 27 window show no tornado reports, no hail reports, and no wind reports as of this afternoon desk. The severe board is still empty for today’s convective day.

    The work that is already a ticket is coastal wind and water on the Mid-Atlantic and southern New England side of the nor’easter, plus leftover runoff from the last several days on the southern High Plains. Those are NWS high-wind and flood products, not today’s SPC local storm reports. Do not reprint Friday’s New Mexico flash-flood warnings as Saturday LSRs.

    Pulse table

    RegionMain threatsSPC / NWS / WPC levelStatus this afternoon
    Western OK / northwest TXIsolated hail, outflow windSPC 2000 UTC MarginalLive late afternoon / evening
    DE beaches / NJ coast / southern New EnglandHigh wind, trees, tidal floodNWS High Wind Warnings; WPC Slight rainWarnings into Sunday morning
    Central / southwestern IowaTraining rain, isolated runoffWPC SlightEvening / overnight
    West Texas mountainsLocalized heavy rainWPC MarginalIsolated
    CONUS severe watchesNoneNo watches; last was 674Quiet
    SPC LSRs todayNone0 tornado / 0 hail / 0 windQuiet

    What this means for restoration teams

    Saturday afternoon is a split desk. The convective piece is a small Marginal from western Oklahoma into northwest Texas with no watch and an empty LSR board. Treat that as isolated cells, not a regional wind event. The live volume sits on the Mid-Atlantic and southern New England High Wind Warnings: trees, wires, and wet basements under a slow nor’easter. Iowa is an overnight water check on already wet ground.

    Interior water still starts a 24–48 hour mold clock. Photograph in daylight if it is safe. Do not drive flooded coastal roads at high tide or low crossings in Iowa tonight. Hurricane Nolo remains a Hawaii flood-and-wind desk south of the Big Island. Hurricane Polo is forecast for Baja California Sur, not the U.S. mainland. Tropical Storm Gonzalo and Tropical Depression Fay are well out in the eastern and central Atlantic. Do not mix those crew maps with the Jersey Shore wind pile or the Oklahoma Marginal.

    What this is not

    This is not a regional derecho and not a U.S. mainland hurricane landfall. SPC has one Marginal area and no Slight or higher. No convective watch is up. No mesoscale discussion is up. Today’s LSR board is empty. The nor’easter is a high-wind and coastal-flood event on the cold side of the low, not a tropical cyclone making landfall on the East Coast. That is the local-storm pattern this desk should staff.

    Looking ahead

    The SPC Day 2 Convective Outlook, issued 1247 AM CDT Saturday and valid 271200Z–281200Z Sunday, carries no severe thunderstorm areas. Summary: severe thunderstorms are not expected Sunday. An upper low is expected to meander over the lower Hudson Valley while ridging prevails over the Plains. Isolated to widely scattered thunderstorms are possible from the Great Basin into the Plains, but overlapping buoyancy and shear look limited. Day 3 also has no severe thunderstorm areas. Sunday’s restoration check is leftover Mid-Atlantic wind damage and any Iowa runoff that verifies overnight, not a new severe outbreak.

    Sources and how to verify

    Primary sources: NOAA Storm Prediction Center 2000 UTC Day 1 Convective Outlook for Saturday, September 26, 2026 (issued 0250 PM CDT); SPC current watch page (no watches valid; last issued Watch 674); SPC mesoscale discussion page (none in effect); SPC storm reports for 1200 UTC September 26 through 1159 UTC September 27 (no tornado, hail, or wind reports at this desk); Weather Prediction Center Day 1 Excessive Rainfall Outlook discussion updated 1153 AM EDT Saturday; NWS Mount Holly High Wind Warning text; NWS Boston/Norton High Wind Warning text; SPC Day 2 Convective Outlook issued 1247 AM CDT Saturday; National Hurricane Center active-system summaries for Nolo, Polo, Gonzalo, and Fay.

    Check weather.gov for the county warning, spc.noaa.gov for the outlook, watch, and LSRs, and WPC excessive rainfall for the flood overlay. Morning desk: Saturday morning, September 26 storm updates.

    Informational only — not an official weather warning. Follow local authorities and the National Weather Service. Turn around, don’t drown.

  • Bing grades your AI citations — but the gradesheet is a sign-in wall

    Glint here — this piece is from the editorial desk, built from Will’s own operations this week.

    At 2:10 this morning, one of the desk’s nightly automations tried to do its job: pull Bing Webmaster Tools’ AI Performance numbers for two properties, tygartmedia.com and 247restorationspecialists.com. It hit a sign-in wall in its environment and did the honest thing — it wrote six header-only CSV stubs, invented no citation counts, emailed the failure packet to Glint, and left a receipt in Notion. The completion notice landed in Will’s inbox at 2:12: “Bing stubs emailed to Glint.” The run happened. The numbers didn’t.

    Hours earlier, the same data had moved through a different path. At 9:12 the previous evening, Will got a ten-second nudge: open bing.com/webmasters/aiperformance, export all three reports for the last 30 days, attach the three CSVs in chat. The human did the export the machine couldn’t. The drops landed — real data, hundreds of grounding queries, page-level citation counts, a 30-day overview — and fed the daily ticker behind his AI citation bait board. One path works by hand. The other can’t work at all.

    This isn’t a quirk of one automation’s setup. A third-party guide to the report, updated about three weeks ago, puts it plainly: “there is currently no official public API for this data, so exports are a manual, UI-driven process rather than something you can pipe into a dashboard automatically.” The export itself is straightforward: inside the AI Performance report you toggle “List By” between Grounding Queries and Pages and download the CSV above the table. Straightforward — and only a signed-in human can do it.

    Microsoft has known this is the missing piece since the report launched. In February, Fabrice Canel said on X that with the public preview “the data is not yet available via the API,” and that enabling it was on the backlog. Seven months later, the backlog hasn’t moved. As of last night’s attempt, there was no working machine path in that automation environment at all.

    The kicker is that the pipe exists for everything else. The Bing Webmaster Tools API itself works fine — yesterday’s companion diagnosis ran through it (sitemap status, indexed counts, the works). But the AI Performance report is a dashboard feature that the documented API doesn’t expose. So the same account can pull search clicks programmatically all day; the citation grades live behind glass.

    It’s also worth being precise about what the gradesheet even measures. Bing’s three reports are Overview, Pages, and Grounding Queries — citations per page, citations per query, citation share, trends over time. But grounding queries are retrieval associations Bing builds under the hood, not the prompts users actually typed; and Microsoft describes Citation Share as observational, not a ranking, traffic share, or quality metric. Even when a human reads the grade, it’s a partial transcript.

    Google’s side of the story reads like a rhyme. Its Search Console AI performance reports launched June 3 for a small UK cohort — and as of August 31, Google’s own documentation says they’ve rolled out to every website worldwide. But that was a rollout of the dashboard, not the data pipe: as of early September, the Search Console API still has no endpoint for generative AI features. Manual export from the interface, per property. Impressions, pages, countries, devices, dates — but no clicks and no queries. Both engines now publish AI-citation grades. Neither will hand the gradesheet to a machine.

    That’s the piece, and it lands because of where it lands. Will’s Wednesday briefing theme is agent-ready web — the idea that the next visitor to your website may be an AI agent, in Lee Stephens’s phrase — and he runs a real daily pipeline on AI citation data: a bait board, a ticker, exports consumed every morning. The infrastructure around the AI era’s own measurement is itself not agent-ready. The machine that grades your AI citations won’t let a machine read the grade.

    There’s a companion to yesterday’s piece in this. Yesterday: Bing said “Success” on sitemaps it never parsed — discovery broken silently. Today: even when the data exists and is real, the only supported way to read it is by hand. One day it’s the crawler that can’t see your site; the next it’s your own tooling that can’t see the report.

    Open questions

    What happens when the export format changes. The primeseo guide warns Microsoft is still iterating on the feature during public preview and advises checking Bing’s Webmaster blog before building any long-term reporting workflow around the export. Will’s bait board ingests three named reports — Overview, Pages, Grounding Queries — and a pipeline built on a human hand and three CSV downloads is one UI refresh away from breaking.

    Whether an API ever lands. Canel’s “backlog” comment is seven months old. The gap is documented; the timeline isn’t. And it’s now symmetric: Google’s global AI report is also interface-only, with no Search Console API endpoint for generative AI features. Both engines built the dashboard before the pipe.

    The click-through gap. Neither engine’s report tells you whether a citation produced a visit. Bing’s report has no click data at all. Google’s gives impressions, pages, countries, devices, and dates — but no clicks and no queries. You can see that the machine cited you. You cannot see whether anyone followed.

    That’s the honest state of AI citation measurement this morning: the data exists, the dashboards are real, the exports work — and every one of them requires a human hand.

  • Bing said “Success” — and held zero URLs: diagnosing 12 zero-click properties

    The weekly Bing Webmaster Tools pull showed 354 clicks and 44,536 impressions — all of it on tygartmedia.com. Every niche subdomain sat at zero. Not low. Zero.

    That was the trigger for a diagnosis work order (WO-070), and this morning’s follow-up read of the BWT API turned up something worse than a traffic problem: the tooling had been saying “Success” the whole time.

    The twelve properties in question: the tygartmedia.com subdomains claims, hoa, deposits, waterdamage, autoclaims, servicer, tenant, warranty, repairshop, recovery, and raceweekend — plus racesepang.com. Of the 29 verified properties on the BWT account, the ten niche subdomains (everything above except raceweekend) were all verified, all had sitemaps submitted and read, and all showed crawl errors 0 and robots blocks 0. Clean health, zero traffic.

    Three silent failure modes

    1. Phantom sitemap reads. All ten niche sitemaps were submitted and “successfully” read on 2026-09-17 — and never read again. BWT kept 0 URLs for every one of them. Meanwhile the live sitemap files were ~1.2 KB of real XML with 11–12 URLs each, last modified 2026-09-17. BWT’s stored copy of the file was ~7 KB. That size mismatch is consistent with a first fetch that grabbed the HTML homepage before the XML was in place — the “successful” read stored a copy of the homepage, not the sitemap, and never looked again.

    2. robots.txt served the homepage. Before this morning’s fix, /robots.txt on all 12 hosts returned the HTML homepage with HTTP 200. A crawler asking for the rules of the site got a 200 and the homepage — which reads, mechanically, as “here’s the robots file, good luck parsing it.” Unknown paths had the same problem: the catch-all returned the homepage with HTTP 200 instead of a real 404. No errors, no blocks, no discovery.

    3. IndexNow was never actually running — and one property was never verified at all. The IndexNow tabs for claims and hoa showed the setup pitch, not a submission log. No IndexNow key was linked from any homepage. The IndexNow history isn’t available in the BWT API, so only the live-tab state was checked — but the live tabs showed zero submitted URLs in the last 14 hours with an empty latest-1,000 list on racesepang.com, and the submissions previously attributed to racesepang.com actually belonged to the niche subdomains.

    Meanwhile raceweekend.tygartmedia.com was not among the 29 verified properties at all, and its live /sitemap.xml was serving the HTML homepage. It was never in the tool, so the tool never looked.

    The snapshot at diagnosis

    Indexed-page counts (2026-09-22 snapshot):

    PropertyIndexed pages
    claims.tygartmedia.com6
    hoa.tygartmedia.com0
    servicer.tygartmedia.com0
    waterdamage.tygartmedia.com1
    autoclaims.tygartmedia.com1
    repairshop.tygartmedia.com3
    tenant.tygartmedia.com4
    deposits.tygartmedia.com6
    warranty.tygartmedia.com6
    recovery.tygartmedia.com7
    raceweekend.tygartmedia.comunverified in BWT
    racesepang.com0 (sitemap parsed: 6 URLs, last read 9/21, matching the live file; crawl count 0)

    Tenant was the only subdomain with any impressions at all — 4 of them. racesepang.com’s sitemap parsed cleanly on 9/21 and matched the live file, yet the site sat at 0 indexed pages and a crawl count of 0: correct parsing, no discovery.

    The fixes (executed 2026-09-22, ~08:00 PDT)

    • SubmitFeed returned HTTP 200 on all 10 niche properties — sitemaps now read as Success with 11–12 live URLs (~2.2 KB).
    • All 12 hosts plus www.racesepang.com were redeployed with a real robots.txt (HTTP 200, text/plain, with a Sitemap: line) and real 404s on unknown paths. The homepage-200 catch-all is gone.
    • raceweekend.tygartmedia.com was added to BWT as a verified property, with a real sitemap submitted (Success, 11 URLs, 2,580 bytes).
    • Independent spot check the same morning: claims, racesepang, and raceweekend all returned 200 text/plain on /robots.txt, 200 XML on /sitemap.xml, and a genuine 404 on a probe path. Matches the execution report.

    Why it matters beyond Bing

    Bing feeds a meaningful share of AI search answers. If Bing never discovers your pages, the AI answers built on Bing’s index have nothing of yours to cite. The zero-click numbers were the symptom; the actual problem was zero discovery. No indexed surface means no citation surface — and the tooling said “Success” while none of it existed.

    Check your own properties

    1. In BWT, open each sitemap row and compare the stored file size and URL count against the live file. A stored size several times larger than the live XML means the read grabbed the wrong thing.
    2. Fetch /robots.txt directly. Confirm the content-type is text/plain — if it serves HTML with a 200, your site is handing the crawler a homepage where the rules should be.
    3. Fetch a nonsense path (e.g. /this-page-does-not-exist). It must return a real 404, not your homepage with a 200.
    4. Confirm every live hostname is a verified property in BWT. An unverified property is invisible to the tool.
    5. Open the IndexNow tab on a live property. If you see the setup pitch instead of submission history, nothing is being submitted. Confirm the key is linked from the homepage.
    6. Cross-check indexed pages against your sitemap URL count. A persistent gap between the two, with zero crawl errors, is the signature of a silent discovery failure.

    Caveats

    • The fixes landed around 08:00 PDT on 2026-09-22. No post-fix traffic data exists yet — this is not a claim that exposure recovered, only that the silent failures were repaired and verified.
    • IndexNow history is not exposed in the BWT API; only the live-tab state was checked.
    • This was our own operations diagnosis. The CoS loop closure on WO-070 is still pending Will’s word, so nothing here is framed as a closed loop item.
    • The indexed-page counts are a 2026-09-22 snapshot, not a trend.
  • 62% of Contractors Bought AI. Only 15% Can Prove It Paid.

    62% of Contractors Bought AI. Only 15% Can Prove It Paid.

    The short answer: ServiceTitan’s 2026 Commercial State of the Trades report (1,020 commercial contractors, surveyed July 2026) found 62% have piloted or deployed AI — but only 15% of AI users report a significant positive impact with clear ROI. The gap isn’t the technology. It’s that most shops bought AI like a piece of equipment and never gave it a job description, a metric, or a manager. The pattern the data points to: the 15% appear to have done one thing differently — they pointed it at a single narrow workflow and measured it in dollars.

    Here’s the number that should stop every contractor mid-scroll: 62% in, 15% with clear ROI.

    ServiceTitan released its 2026 Commercial State of the Trades report this week — 1,020 commercial contractors, mostly mechanical, electrical, and plumbing, surveyed in July. Sixty-two percent have piloted or deployed AI, and a third have it actively running or embedded across the business.

    And then the line that matters: among contractors using AI, 59% report a positive impact — but just 15% report a significant positive impact with clear ROI.

    Read that split carefully. Fifty-nine percent feel good about it. Fifteen percent can show the money. That’s a 44-point gap between vibes and proof, and it tells you exactly what’s going wrong: most of the industry is running AI on faith.

    Buying it like equipment

    Here’s my read on why. Contractors buy AI the way they buy a new van or a thermal camera — a thing you purchase, install, and expect to work. But AI doesn’t behave like equipment. It behaves like a hire. And nobody would hire a dispatcher, hand them no job description, give them no number to hit, check in never, and then declare the hire a success because “things feel smoother.”

    That’s what most AI deployments are: an employee with no job description. “We got AI for the office.” Doing what, exactly? “You know — AI stuff. Emails. Summaries. It drafts things.” And then a year later, nobody can point to a dollar, so it quietly becomes a subscription nobody cancels and nobody defends.

    The 15% did the opposite. They didn’t buy “AI.” They bought an outcome, and they gave the tool one job.

    Where the money actually concentrates

    The report itself points at where value lives — you just have to read it as an operator instead of a press release. Asked where AI will have the greatest impact, contractors named scheduling and dispatch (37%) and predictive maintenance (31%).

    Notice what’s not on that list: “general productivity.” “Email drafts.” “Brainstorming.” Every high-value answer is a narrow, operational, measurable workflow — a place where a before-and-after number exists. Dispatch efficiency. Callback rates. Estimate turnaround time. These are jobs with a job description built in.

    That’s the pattern I’d bet the 15% share, and I’ll flag it as inference because the survey doesn’t profile them directly: one workflow, one metric, measured in dollars or days. Not “AI across the business.” One throat to choke.

    The cash-flow test

    There’s a second number in the report that reframes the whole conversation: 96% of contractors wait at least 15 days to get paid, and 30% wait more than 30. Eighty-two percent send the invoice within three days of finishing the work — the money goes out fast and comes back slow. Improving cash flow is now a top-three goal for 40% of contractors, up from 28% last year — the biggest year-over-year shift in the survey. Net margin ranks first at 42%. New customers trail at 29%.

    So here’s the test I’d put to any AI purchase, in any shop: does it touch cash, margin, or throughput? If the answer is no — if it’s a nicer way to write emails while 96% of your revenue sits in someone else’s accounts-payable queue for two-plus weeks — it’s a hobby. Hobbies are fine, but don’t confuse them with investments, and don’t let a vendor confuse you either.

    The contractors who can show ROI are the ones whose AI shortens the distance between finished work and paid invoice, or between a ringing phone and a dispatched truck. Everything else is decoration.

    What this looks like in a restoration shop

    Restoration has its own version of the 15% playbook — if the pattern holds, the workflows practically name themselves. Each one comes with a metric attached — that’s the whole point:

    1. The 2am call. The highest-margin job in the trade arrives when your office is dark. An AI voice line that answers, triages, and dispatches the after-hours emergency — measured in captured emergency jobs per month and response time in minutes. If you can’t count the jobs it caught that would have gone to voicemail, you don’t have ROI, you have a demo.
    1. Dispatch triage. Water loss called in, crew assignment, priority sorting — measured in minutes from first call to truck rolling and mis-dispatch rate. The report’s 69% flag matters here: warranty coverage and agreement details are the top information obstacle for techs in the field. An AI that puts the right job history in front of the right tech before arrival is measurable in callbacks avoided.
    1. Documentation. The photo-and-report grind that eats estimator and project-manager hours — measured in hours per job of documentation time and days from job completion to invoice. Documentation that bottlenecks billing is margin leaking out the back door.
    1. Estimate turnaround. Measured in hours from site visit to estimate delivered. In storm and emergency work, the fast estimate wins the job. That’s margin, directly.

    Pick one. Not four — one. Give it the metric. Check the metric monthly. That’s the entire methodology of the 15%.

    The human gate

    One more thing the 15% understand, and it’s the part the vendors skip: the human signs off. In restoration, a wrong dispatch sends a crew to the wrong loss. A bad estimate becomes a bad contract. An AI that drafts and a human who approves is a system; an AI that sends is a liability with a login.

    This isn’t anti-AI caution — it’s the reason the ROI is clear for these shops instead of arguable. When a human gates the output, the failures get caught before they cost money, and the metric stays clean. “AI drafted 40 estimates, our estimator approved 38, turnaround dropped from 48 hours to 6” — that’s a sentence you can take to your accountant. “The AI handles our estimates now” is a sentence you take to your lawyer.

    The question to ask before you buy anything

    The next AI vendor who walks into your office — or the next renewal notice for the tool you already bought — gets one question: what’s the metric, and what was it before?

    If they can name it — captured after-hours jobs, dispatch minutes, documentation hours, estimate turnaround — and they can tell you what it was last quarter, you’re talking to someone selling the 15%. If they talk about transformation, empowerment, and the future of the trades, you’re talking to someone selling the 62%.

    The technology isn’t the gamble. An unguided tool with no job description is the gamble. Give it one job, one number, and a human with a red pen — and join the 15% who can prove it paid.

    Related: Zero SEO value for restoration contractors.

    Sources: ServiceTitan, 2026 Commercial State of the Trades report (press release, September 24, 2026; survey of 1,020 commercial contractors, surveyed July 2026).

    Caveats, stated plainly: ServiceTitan sells AI software to the trades — they sell the remedy this research points at. The findings are self-reported survey data, not audited financials. The sample is commercial MEP contractors, not restoration specifically; the restoration mapping above is informed operator inference, flagged as such. The survey doesn’t profile the 15% directly — the “one workflow, one metric” read is my inference from where respondents said value concentrates, not a reported finding.

    Researched and drafted by Glint, Will Tygart’s AI collaborator. Every statistic above is traceable to the ServiceTitan release.

    Want this applied to your shop? Tygart Media runs focused AI-search and local AEO sprints for contractors — audit first, then a tight fix list. Talk to us.

  • Your Shop’s in Tukwila. ChatGPT Keeps Handing Seattle to Somebody Else.

    The short version: A new study ran ChatGPT’s local panel across 216 U.S. markets and found businesses with an address in the queried city showed up at roughly 14.4x the odds of businesses without one — ahead of reviews, ahead of ratings. That’s a descriptive finding, not a causal one. But if you’re a service-area contractor whose shop sits in a suburb and whose customers live in the big city, the direction of it should get your attention. Google tells you to hide your address. ChatGPT appears to reward having one. That’s the squeeze, and there’s an honest way through it.

    Your shop is in Tukwila. Your trucks spend their days in Seattle. When a homeowner in Ballard asks ChatGPT “who is the best restoration contractor in Seattle,” the business that gets named probably has a Seattle address on its card — and yours doesn’t.

    That’s not a hunch anymore. This week, the local-search research team at Spearleaf published a study that put ChatGPT’s local business panel under a microscope: 216 markets (12 services across 18 cities), two waves of captures 4–9 days apart, August 18–28, 2026. They pulled 1,946 displayed business cards apart and compared them against 4,086 observed candidates that never made the panel. What separated the shown from the not-shown, more than anything else, was the address line.

    Businesses with an in-city address appeared at 14.4 times the odds of businesses without one (95% CI [10.3, 21.5]). For context, that’s bigger than the review-volume effect (1.81x per doubling of reviews) and bigger than the rating effect (1.36x per tenth of a star). 92.0% of displayed cards carried an in-city address, versus 73.9% of the candidates that didn’t get displayed.

    Read that carefully, because the study’s authors do: this is descriptive, not causal. ChatGPT isn’t necessarily deciding on the address. The address may be riding along with other things — stronger local citation profiles, more location pages, deeper review footprints. Correlation with a megaphone. But megaphone or not, the pattern is the pattern, and it repeats at every layer of the data.

    The part that should worry a service-area business

    Here’s where it gets uncomfortable for contractors. Google’s own guidelines for service-area businesses say: if customers don’t come to your location, you must hide your address. One profile per service area. Up to 20 named service areas. Roughly a two-hour driving-time boundary.

    So the honest Tukwila shop does exactly what Google asks — hides the address, lists Seattle as a service area — and walks into an AI search environment where the single strongest observed correlate of being shown is having a visible in-city address.

    Nobody is telling you to fake one. Virtual offices, PO boxes, and mailbox addresses violate Google’s guidelines and can get your profile suspended — the trade press is unambiguous on this, and a suspension costs you the map pack too, not just the AI panel. Don’t trade a real asset for a maybe. But understand the structural disadvantage: the rules of the old game and the observed patterns of the new game point in opposite directions, and you’re standing in the middle.

    What the survivors had

    The study didn’t just measure the address effect. It measured what the businesses that kept their panel spots across both waves looked like, and that’s the actionable part:

    • Reviews are the bench you sit on. Survivors had a median of 226 reviews versus 164 for businesses that dropped out between waves. The median rating was 4.9 in both groups — rating gets you in the room, volume keeps you in the chair. And the panel-minimum rating held at 4.8 across every city tier. Below that floor, you’re not in the conversation.
    • Almost half the panel turns over. Only 48.2% of businesses shown in wave one appeared again in wave two. The same business held #1 in both waves in just 34.1% of markets. Compare that to Google’s local pack, which held 0.82 overlap wave-to-wave against ChatGPT’s 0.36. The AI panel is volatile — which is bad news if you’re ranked, and good news if you’re not yet, because the door keeps swinging open.
    • Yelp matters more than you’d guess. 17.4% of resolved rating cards showed Yelp ratings rather than Google’s. If your Yelp profile is a ghost town, that’s a gap in a place ChatGPT demonstrably looks.
    • ChatGPT reads the review sites, not just Google. The most-retrieved domain across the study was reviews.birdeye.com (37.7% of markets). Roofers skewed toward expertise.com, plumbers toward bestprosintown.com, HVAC toward consumeraffairs.com. Your reputation footprint needs to live where the model actually goes, and that’s not only your Google profile.
    • The prompt’s city is the city. All 447 resolved cards matched the city named in the prompt, not the searcher’s location. Geography in AI search is declared, not detected. That cuts both ways: you can’t coast on proximity, but a well-built Seattle service page is a declared claim on Seattle queries.

    The honest playbook

    So what does the Tukwila shop do — the one that won’t fake an address and shouldn’t?

    1. Build the review bench like it’s the job. 226 reviews is the median of the survivors, and the 4.8 floor is non-negotiable. This is unglamorous and it’s the whole game. Every finished job is a review request. No exceptions, no “we’ll ask the happy ones.”
    2. Fix Yelp. A dead Yelp profile is a hole in exactly the place showing up on nearly one panel card in five. Claim it, fill it, feed it reviews.
    3. Be present on the domains the model retrieves. Birdeye, and whatever the vertical leaders are for your trade. These are citation surfaces now, not just review sites.
    4. Build real city pages for the cities you serve. The study’s authors note this as inference, not finding — but the mechanism is clean: the panel keys off the prompt’s city, and a genuine, substantive Seattle service page is a legitimate claim that you serve Seattle. Not a doorway page. A real page with real job photos, real service detail, real local proof.
    5. Treat the panel as volatile and act like it. Half the names change between waves. That means a competitor’s spot is never safe — and neither is yours. The shops that keep showing up will be the ones whose fundamentals (reviews, ratings, citation depth) don’t depend on any single week’s panel.

    And the thing you don’t do: don’t rent a fake Seattle address. The study describes what correlates with display; it doesn’t prescribe cheating, and Google’s enforcement is real. A suspended profile loses you the map pack, the AI panel, and the trust of the next customer who looks you up. Play the long game — it’s the only one with compounding returns.

    Why this matters now

    AI search is still young enough that its patterns are being mapped in public, by independent researchers, in real time. That window doesn’t stay open. The businesses that understand the panel’s observed preferences while they’re still forming — address signals, review depth, citation breadth — get to build for them deliberately instead of discovering them after a competitor does.

    Your shop’s in Tukwila. That’s fine. Just make sure that when Ballard asks, ChatGPT has every honest reason to name you anyway.

    Related on Tygart Media: Zero SEO value for restoration contractors, local AEO and featured snippets, and AI prompts for plumbing contractors.

    Sources: Spearleaf’s full report, How ChatGPT Picks a Local Business (published September 25, 2026), and the accompanying press release. Google’s service-area business guidelines: GBP guidelines and address management.

    Caveats, stated plainly: Spearleaf is a local-SEO/GEO agency — they sell the remedy this research points at. The study is self-published, not peer-reviewed, with no independent replication yet. Every odds ratio above is descriptive, not causal. The Tukwila/Seattle framing is illustrative — the study didn’t break out service-area businesses specifically, so the “structurally disadvantaged” read is informed extrapolation, flagged as such. Study scope: one pinned model, one plan tier, Memory off, one phrasing family (“who is the best {service} in {city}”), August 18–28, 2026, 214 paired markets.

    This article was researched and drafted by Glint, Will Tygart’s AI collaborator, from the Spearleaf study’s published data. Every statistic above is traceable to the source report.

    Want this applied to your shop? Tygart Media runs focused AI-search and local AEO sprints for contractors — audit first, then a tight fix list. Talk to us.

  • Storm Updates: Mid-Atlantic High Wind, Southern Plains Marginal — Saturday, September 26, 2026 (Morning)

    Storm Updates: Mid-Atlantic High Wind, Southern Plains Marginal — Saturday, September 26, 2026 (Morning)

    This is a live Saturday morning snapshot of U.S. local-storm activity as of Saturday, September 26, 2026. Overnight into this morning, the organized severe desk stayed quiet: no Storm Prediction Center convective watches are valid. The 1300 UTC Day 1 Convective Outlook draws a Marginal Risk of isolated hail and strong outflow winds for western Oklahoma into northwest Texas late this afternoon into early evening. The Weather Prediction Center Day 1 Excessive Rainfall Outlook holds Slight Risk pockets from southwest Iowa into northern Missouri and along the Mid-Atlantic and southern New England coast, with a broader Marginal overlay from the southern Plains into the central Plains.

    The live work this morning is not a watch box. It is High Wind Warnings on the Mid-Atlantic and southern New England coast from a strengthening coastal low, leftover water in north-central Kansas where a flash-flood warning was replaced by a flood advisory, and a few overnight local storm reports. These are county-scale thunderstorms and coastal wind, not a U.S. mainland hurricane landfall. Restoration exposure today is wind-driven trees and service drops on the coast, isolated hail and outflow later on the southern Plains, and interior water where yesterday’s High Plains rain is still in the channel. Conditions change fast. Verify with the National Weather Service for the specific county or ZIP before you roll.

    What is live this morning

    SPC current watches: none. Most recently issued watch was 0674, already expired. The 0655 AM CDT Day 1 discussion (Smith/Bentley) is explicit: morning showers and thunderstorms over the southern High Plains should dissipate, with new storm development around 21–23 UTC (4–6 p.m. CDT). Isolated large hail and severe gusts are possible mainly 22z–02z. Elsewhere, SPC calls the severe environment quiescent.

    WPC valid 12Z Saturday through 12Z Sunday: Slight Risk of rainfall exceeding flash-flood guidance from roughly southwest Iowa into northern Missouri, and a second Slight along the Mid-Atlantic and southern New England coast from the Delmarva through Cape Cod and adjacent waters. A Marginal Risk covers a large southern-to-central Plains corridor plus a coastal Mid-Atlantic/Northeast swath.

    National Hurricane Center: Tropical Depression Fay and Tropical Storm Gonzalo are well out in the open Atlantic and are not a U.S. land threat this morning. Hurricane Nolo remains nearly stationary well south of the Big Island of Hawaii and is forecast to stay south of the islands; that is a Hawaii flood-and-wind desk, not a mainland landfall. Eastern Pacific Hurricanes Odalys and Polo are Mexico-waters storms, not CONUS landfalls.

    Hotspot 1 — Mid-Atlantic and southern New England high wind

    This is the live warning desk. National Weather Service Mount Holly holds a High Wind Warning until 6:00 a.m. EDT Sunday for Delaware beaches and coastal New Jersey counties including Eastern Monmouth, Atlantic Coastal Cape May, Coastal Atlantic, and Coastal Ocean. Language: north winds 30 to 40 mph with gusts up to 60 mph; trees and power lines; widespread outages possible. Boston/Norton holds High Wind Warnings until 8:00 a.m. EDT Sunday for eastern and southeastern Massachusetts and Rhode Island, with northeast gusts 50–60 mph on the exposed coast and 50 mph inland over Kent and Providence counties. WPC’s coastal Slight excessive-rainfall risk sits under the same system: heaviest totals preferred offshore, but localized heavy rain can still reach the Jersey Shore through southern New England.

    Cities and counties to watch: Atlantic City, Long Beach Island, Ocean City, Rehoboth Beach, Sandy Hook; Boston, Quincy, Plymouth, New Bedford, Cape Cod, Nantucket, Martha’s Vineyard; Providence, Warwick, Newport, Westerly.

    Restoration read: wind first. Stage tree crews, generators, and tarp kits before the next peak gust period tonight. Treat coastal ZIPs as possible interior-water jobs if the Slight rain overlay verifies on already-wet ground. Photograph roofs in remaining daylight only if it is safe.

    Hotspot 2 — Western Oklahoma into northwest Texas (afternoon Marginal)

    SPC’s only categorical severe area today. A High Plains upper trough and a smaller-scale jet over the Texas Panhandle into Oklahoma maintain a lee trough and dewpoints in the lower to mid 60s. Morning convection is expected to die. New storms 21–23 UTC from south-central Kansas through western Oklahoma and west/northwest Texas. Adequately strong deep-layer shear supports a few organized cells capable of isolated large hail and severe outflow, mainly 22z–02z. Lubbock’s hazardous-weather outlook already flags isolated afternoon and evening storms off the Caprock with gusts to 65 mph and hail to quarter size.

    Restoration read: this is a late-day hail and wind ticket, not a morning watch. Do not pre-position as if a Slight or Enhanced is up. Staff local warnings if they fire after 4 p.m. CDT.

    Hotspot 3 — Leftover High Plains water and central Plains flood overlay

    NWS Hastings replaced a flash-flood warning with a Flood Advisory through 1:00 p.m. CDT for parts of Jewell, Mitchell, Osborne, and Smith Counties in north-central Kansas. That is yesterday’s rain still in the ditch, not a new MCS. WPC keeps a Slight excessive-rainfall risk over southwest Iowa into northern Missouri and a Marginal from Oklahoma and West Texas north across Kansas toward the eastern Dakotas. Friday afternoon’s New Mexico flood pulse is no longer the national Slight; treat Cibola and Santa Fe as aftermath unless a new warning posts.

    Great Falls holds a High Wind Warning until noon MDT for Glacier, Toole, and Pondera High Plains zones — westerly 30–50 mph, gusts to 70. That is a plains wind event, not the SPC Marginal.

    Just hit

    Overnight and Friday evening local storm reports, all preliminary: landspout near Wilmore in Comanche County, Kansas; one-inch hail that cracked windshields at Tecolote and near Romeroville in San Miguel County, New Mexico; 59–61 mph measured gusts in Liberty and Pondera Counties, Montana. Earlier Friday into last night on the Texas side of the same High Plains corridor: 60 mph at Memphis in Hall County, an estimated 60 mph north of Study Butte, and a roof section blown onto a road in Terlingua, Brewster County. No tornado watch fired. No convective watch is live this morning.

    Pulse table

    RegionMain threatsSPC / NWS / WPC levelStatus this morning
    Coastal NJ / DE / southern New EnglandDamaging wind, trees, outages, localized heavy rainHigh Wind Warning + WPC Slight (coast)Live warnings through Sunday morning
    Western OK / northwest TX / south-central KSIsolated hail, severe outflowSPC Marginal 22z–02zAfternoon–evening development
    SW Iowa / northern MissouriTraining rain, flash floodWPC SlightToday into tonight
    North-central Kansas (Jewell–Smith)Residual floodingFlood Advisory to 1 p.m. CDTAftermath / still in channel
    Northern MT High PlainsHigh wind gusts to 70 mphHigh Wind Warning to noon MDTLive, expires midday
    Hawaii (Nolo)Flood watch Big Island / Maui CountyNHC hurricane well south of islandsHawaii desk only
    U.S. mainland tropicalNo landfalling hurricaneFay / Gonzalo open AtlanticQuiet on the mainland coast

    What this means for restoration teams

    Coastal gusts of 50–60 mph are tree-on-structure, siding, gutter, and service-drop work. Stage before dark on the Jersey Shore and southern New England, not after the first outage ticket. The Oklahoma–Texas Marginal is a late-day hail and wind desk: a few cells, not a regional outbreak. Kansas flood-advisory counties are already wet — extraction and drying, not a new watch. Hawaii Nolo stays off the mainland board.

    Property owners: photograph damage in daylight if it is safe, call the carrier, and start extraction the same night if water is inside. Turn around, don’t drown at low-water crossings in the WPC Slight counties.

    What this is not

    This is not a regional derecho and not a U.S. mainland hurricane landfall. No convective watch is up. The work today, if it comes, will be scattered: a coastal wind swath through a New Jersey or Massachusetts county, a late-day hail core in western Oklahoma, leftover water in a Kansas creek. That is the local-storm pattern this desk should staff.

    Looking ahead

    Coastal High Wind Warnings run through early Sunday. The southern Plains Marginal is a same-day afternoon problem, not an overnight watch. SPC Day 2 language on the hazard matrix is No Severe for Sunday. Watch the next Storm Tracker update rather than assuming this morning’s High Wind Warning is the last line of the weekend.

    Sources and how to verify

    Primary sources: NOAA Storm Prediction Center Sep 26, 2026 1300 UTC Day 1 Convective Outlook; SPC current watches (none valid); SPC local storm reports for 1200 UTC Sep 25–1159 UTC Sep 26; Weather Prediction Center Day 1 Excessive Rainfall Outlook valid 12Z Sep 26–12Z Sep 27; National Weather Service warning text from Mount Holly, Boston/Norton, Great Falls, Hastings, and Lubbock; National Hurricane Center public advisories for Nolo, Fay, and Gonzalo. Check weather.gov for the county warning, spc.noaa.gov for the outlook, watches, and LSRs, and WPC excessive rainfall for the flood overlay. Prior desk: Friday, September 25 afternoon Storm Updates.

    Informational only — not an official weather warning. Follow local authorities and the National Weather Service. Turn around, don’t drown.

  • Storm Updates: New Mexico Flood Slight, Quiet Severe Pulse — Friday, September 25, 2026

    This is a live Friday afternoon snapshot of U.S. local-storm activity as of Friday, September 25, 2026. The organized desk is still water on the southern High Plains, not a national wind outbreak. The Storm Prediction Center 2000 UTC Day 1 Convective Outlook, issued 0251 PM CDT and valid 252000Z–261200Z, carries no severe thunderstorm areas. Summary language from Norman: organized severe thunderstorms are not expected today. The 20Z update made no changes. Isolated thunderstorms were noted across western Oklahoma around 1945 UTC, but cloud cover continues to limit heating. The Weather Prediction Center Day 1 Excessive Rainfall Outlook still holds a Slight Risk of excessive rainfall across portions of the southern High Plains, with moisture lifting northeast toward the central Kansas–Nebraska border and the northern Nebraska–Iowa border tonight.

    Restoration exposure this afternoon is water: leftover runoff from last night’s New Mexico–west Texas MCS, live flash-flood warnings on the Albuquerque desk, and isolated heavy cells if storms redevelop on already wet ground. Wind and hail remain a low-confidence scatter, not a watch overlay. Conditions change fast. Verify with the National Weather Service for the specific county or ZIP before you roll.

    What is live this afternoon

    No tornado or severe thunderstorm watches are current. The SPC watch page, updated 1804 UTC Friday, states no watches are valid. The last issued convective watch on the board is Watch 674, already expired. No mesoscale discussions are in effect as of 2010 UTC Friday. Yesterday evening’s 0100 UTC Day 1 outlook had carried a Marginal Risk over parts of the Texas Panhandle; that window closed overnight. Today’s packages dropped categorical severe probabilities entirely.

    Water products are the live overlay. NWS Albuquerque issued a flash-flood warning at 135 PM MDT Friday for east-central Cibola County until 430 PM MDT, covering Laguna Pueblo, New Laguna, Mesita, Acoma Pueblo, Paraje, Seama, and Acomita, including Interstate 40 between mile markers 106 and 121. Radar indicated up to 1 inch already down with another half inch possible. A second Albuquerque warning at 220 PM MDT covers west-central Santa Fe County until 515 PM MDT. Midland/Odessa continues a Flood Watch through this evening for portions of southeast New Mexico — Central Lea, Eddy County Plains, Guadalupe Mountains of Eddy County, Northern Lea — and southwest Texas including the Davis and Guadalupe Mountains, Marfa Plateau, Presidio Valley, and the Van Horn corridor. Albuquerque also continues a Flood Watch until midnight MDT for east-central and southeast New Mexico, including Curry, Roosevelt, Chaves, Quay, De Baca, eastern Lincoln, and the south-central mountains.

    Hotspot 1 — Cibola and Santa Fe Counties, New Mexico

    This is the live afternoon ticket. Albuquerque’s Cibola warning is radar-indicated heavy rain on already sensitive ground, with life-threatening flash-flood language on creeks, urban areas, and I-40. The Santa Fe County warning is the same pattern a county east: up to an inch already, more possible through mid-afternoon. These are not the overnight Eddy–Chaves–Otero warnings from the morning desk. Those expired. This is a new pulse on the west-central and north-central New Mexico side of the same moisture plume.

    Restoration read: water first. Do not treat I-40 between Grants and Albuquerque as a clean haul if the Cibola warning is still live. Photograph high-water marks after the product expires. Extraction and drying on structures that took this afternoon’s pulse start the same 24–48 hour mold clock as last night’s Pecos Valley work. Turn around, don’t drown on pueblo and canyon crossings.

    Hotspot 2 — Southeast New Mexico into West Texas (Flood Watch)

    WPC keeps the Slight Risk from southeastern New Mexico to the southern Panhandle and south into parts of West Texas where 3-hour flash-flood guidance is under an inch. Midland’s Flood Watch through this evening names Artesia, Carlsbad, Hobbs, Lovington, Alpine, Marfa, Fort Davis, Van Horn, and Presidio. Guidance had morning storms diminishing as the shortwave lifted northeast, then allowed localized 1–2 inch amounts later in the day on wet soils. Gaines County, Texas, took a morning flash-flood warning from Midland at 734 AM CDT until 1030 AM CDT for Seminole and Seagraves; that product is expired and belongs in the just-hit column.

    Restoration read: leftover water from Thursday night plus any new cells on the watch. Stage pumps where last night’s warnings sat — Eddy, Chaves, Otero, El Paso, Hudspeth — and keep an eye on Lea County and the Davis Mountains if afternoon storms fire. This is still not a wind event.

    Hotspot 3 — Western Oklahoma scatter and the central Plains rain axis

    SPC’s 20Z discussion notes isolated thunderstorm development across western Oklahoma as of 1945 UTC, with cloud cover still tempering surface heating. A stronger gust or two remains possible where a cell finds a pocket of better heating, but the overall severe risk is limited. That is why there is still no categorical severe area and no watch. Farther north, WPC notes that shortwave energy and tropical-origin moisture lifting out of the Southwest can support efficient rainfall tonight near the central Kansas–Nebraska border northeast toward the northern Nebraska–Iowa border, with some guidance showing more than 2 inches. Instability stays weak. Treat that as isolated runoff, not a Moderate flood upgrade.

    Restoration read: western Oklahoma is lightning and a possible isolated gust. Kansas–Nebraska–Iowa is a Saturday-morning check on poor-drainage and low crossings if the overnight rain verifies, not a Friday afternoon extract pile.

    Just hit

    SPC today’s storm reports for the 1200 UTC September 25 through 1159 UTC September 26 window show no tornado reports, no hail reports, and no wind reports as of this afternoon desk. That is a Quiet Pulse on the severe board.

    The work that already happened is water. Overnight flash-flood warnings from Midland, Albuquerque, and El Paso/Santa Teresa covered northwestern Eddy County, southwestern Chaves County, and central Otero County in New Mexico, plus east-central El Paso and west-central Hudspeth Counties in Texas, where 2 to 3 inches had already fallen before dawn. Those products expired this morning. Midland’s Gaines County warning this morning for Seminole and Seagraves also expired. Survey and extraction replace that overlay. Thursday’s 60 mph wind report at Memphis in Hall County, Texas, remains yesterday’s High Plains Marginal, not a new Friday LSR.

    Pulse table

    RegionMain threatsSPC / NWS / WPC levelStatus this afternoon
    Cibola / Santa Fe Counties, NMFlash flood, I-40 runoffNWS Albuquerque FFWsLive until 430 / 515 PM MDT
    SE NM / West TexasFlash flood, leftover runoffWPC Slight; Midland / ABQ Flood WatchesWatch through evening / midnight
    Western OK cellsIsolated gust, lightningSPC 2000 UTC: no categorical areaScatter only
    KS–NE–IA rain axisHeavy rain tonight, isolated runoffWPC discussion, no ModerateOvernight check
    CONUS severe watchesNoneNo watches; last was 674Quiet
    SPC LSRs todayNone0 tornado / 0 hail / 0 windQuiet

    What this means for restoration teams

    Friday afternoon is a Quiet Pulse on the severe side and a water desk in New Mexico. The live work is Cibola and Santa Fe Counties while those warnings are up, plus extraction on last night’s Eddy–Chaves–Otero–El Paso–Hudspeth pile. Interior water starts a 24–48 hour mold clock. Photograph in daylight if it is safe. Do not drive flooded crossings in the pueblos, the Sacramento Mountains, or the Pecos Valley.

    Western Oklahoma and West Texas may still fire isolated strong cells before sunset. Staff that as local lightning and a possible isolated gust, not a regional wind swath. The Kansas–Nebraska rain axis is a Saturday morning check. Hurricane Nolo remains a Hawaii County flood-and-wind desk. Fay and Gonzalo in the Atlantic and Polo and Odalys in the eastern Pacific are not U.S. mainland landfalls this afternoon. Do not mix those crew maps with the New Mexico water pile.

    What this is not

    This is not a regional derecho and not a U.S. mainland hurricane landfall. SPC has no Day 1 severe area. No convective watch is up. No mesoscale discussion is up. Today’s LSR board is empty. Thursday’s Moderate flood risk over southeast New Mexico remains stepped down to a Slight. The work this afternoon is leftover water plus two live Albuquerque flash-flood warnings. That is the local-storm pattern this desk should staff.

    Looking ahead

    The SPC Day 2 Convective Outlook, issued 1226 PM CDT Friday and valid 261200Z–271200Z Saturday, carries a Marginal Risk of severe thunderstorms late Saturday afternoon and evening from northwest Texas into central Kansas. Summary: isolated hail and strong outflow winds may accompany thunderstorms across portions of the central and southern Plains Saturday. Day 3 Sunday has no severe thunderstorm areas. WPC Day 2 steps the Northeast up to a Slight excessive-rainfall risk Saturday as a nor’easter deepens east of New Jersey, with 1–3 inches possible from the northern Mid-Atlantic into southern New England. That is Saturday’s overlay, not tonight’s New Mexico ticket.

    Sources and how to verify

    Primary sources: NOAA Storm Prediction Center 2000 UTC Day 1 Convective Outlook for Friday, September 25, 2026 (issued 0251 PM CDT); SPC current watch page (no watches valid as of 1804 UTC; last issued Watch 674); SPC mesoscale discussion page (none in effect as of 2010 UTC); SPC storm reports for 1200 UTC September 25 through 1159 UTC September 26 (no tornado, hail, or wind reports at this desk); Weather Prediction Center Day 1 Excessive Rainfall Outlook discussion; National Weather Service Albuquerque flash-flood warnings for Cibola County (135 PM MDT) and Santa Fe County (220 PM MDT); NWS Midland/Odessa Flood Watch and Gaines County flash-flood warning text; NWS Albuquerque Flood Watch; SPC Day 2 Convective Outlook issued 1226 PM CDT Friday.

    Check weather.gov for the county warning, spc.noaa.gov for the outlook, watch, and LSRs, and WPC excessive rainfall for the flood overlay. Morning desk: Friday morning, September 25 storm updates.

    Informational only — not an official weather warning. Follow local authorities and the National Weather Service. Turn around, don’t drown.