Blog

“Why Did You Recommend Them?” — The 5-Minute Interrogation
Direct answer: Type it. After any AI engine recommends a business — yours or a competitor’s — ask the follow-up: “Why did you recommend them?” The engine will often tell you which signals it used: the reviews it read, the directory it trusted, the facts that tipped the decision. It’s free competitive intelligence, it takes five minutes, and almost nobody in the trades is doing it.

Where this comes from

CI Web Group, a digital agency, published a checklist called “The AI Interview: What Answer Engines Ask About You.” Buried in it as step 5 of a 20-minute self-check is the single most useful sentence in the AI-visibility literature this year:

> “Ask a follow-up: ‘Why did you recommend them?’ The engine will often tell you which signals it used. Free competitive intelligence.”

The full checklist is worth your twenty minutes — open ChatGPT, Perplexity, and Gemini in separate tabs; type the exact question a homeowner would ask for your top three services and your top three cities (“Best plumber in Katy for a slab leak” beats “plumber Katy”); screenshot every answer; note who gets named, who gets skipped, and which facts about your business are wrong; run the follow-up; trace every error to its source; repeat monthly. (CI Web Group)

But the follow-up question is the hinge. Everything else is observation. The follow-up is interrogation.

Why it works

When ChatGPT, Perplexity, or Gemini recommends a business, it has just done retrieval: it searched the web, pulled sources, and synthesized. When you ask why, you’re asking it to narrate that retrieval. The engine will typically name the kinds of signals that carried weight — review volume and recency on a specific platform, a directory profile with complete service data, mentions across multiple independent sources, specific review language matching the question.

Is the explanation perfectly faithful to the model’s internal process? No — and you should know that. The “why” is itself a generated answer: a plausible reconstruction, not a system log. Treat it the way you’d treat a rival estimator explaining his bid: informative, self-serving in places, and most valuable when you cross-check it against the evidence (the actual citations in the first answer, which you screenshotted).

Even with that caveat, it’s the cheapest competitive intelligence in marketing. An agency will happily sell you a “competitor gap analysis” that tells you less — and bill you for the privilege.

The 5-minute procedure (do it tonight)

Minute 1 — Ask the buyer question. Open ChatGPT (or Perplexity, or Gemini — run all three if you have ten minutes). Type exactly what a homeowner would type. Not keywords. A question. Include the service and the city: “Who’s the best plumber in Katy for a slab leak repair?” Screenshot the answer.

Minute 2 — Ask why. Type: “Why did you recommend [the named business]?” Use their exact name from the answer. Screenshot what comes back. You’re looking for signal names: review platforms, specific review counts, directory profiles, “mentioned across multiple sources,” website content it quotes.

Minute 3 — Ask about the sources. Follow up with: “Which specific reviews or pages influenced that recommendation?” and “What would make you recommend a different company for this job?” The second question is the money question — it tells you the gap between you and the winner in the engine’s own words.

Minute 4 — Run it for your business. Now ask about yourself by name: “What do you know about [Your Company] in [City]?” Then: “Why didn’t you recommend them for [the job]?” Screenshot everything. The engine will often list what’s missing — thin reviews, no directory presence, conflicting hours, an unclaimed profile.

Minute 5 — Write down the three gaps. Not ten. Three. The three missing signals the engine named most specifically. Those are your work orders for the month.

What you’ll typically learn

About the winner: which review platform carried the recommendation (often Yelp or Google reviews, sometimes a directory you ignore), whether the win came from review language matching the question rather than review count, and whether the business is even good or just legible. Scott Tischler’s July 2026 experiment found the named winner often isn’t the best-reputed shop in town — it’s “the one that was legible to a machine.” The interrogation tells you which legibility won.

About yourself: SOCi’s 2026 Local Visibility Index puts business profile accuracy on ChatGPT and Perplexity at about 68% — versus 100% on Gemini, which pulls straight from Google Maps — so expect the engine’s picture of your business to contain errors. Wrong hours, wrong services, a closed flag, an old phone number. Each error has a source, and the source is fixable. As CI Web Group puts it: “Wrong hours on ChatGPT usually means wrong hours on a directory the engine trusts. Fix the source, not the chatbot. You cannot argue with the machine. You can only feed it better facts.”

About the game: run the same questions across engines and you’ll see different winners with different reasons — which is exactly what the “which AI should I care about” analysis shows. The interrogation teaches you that there is no single ranking to climb. There are separate evidence pools, and the follow-up shows you which pool each engine drank from.

Three traps to avoid

Trap one: treating one answer as the truth. AI answers are non-deterministic — the same question tomorrow can name a different business. Run the interrogation two or three times across a week before you spend money on what it told you. One screenshot is an anecdote; three is a pattern.

Trap two: arguing with the engine. Telling ChatGPT “that’s wrong, I’m better” accomplishes nothing. The engine restates what the web says. Change the web — the reviews, the directory data, the pages — and the answer follows on the next crawl. The checklist’s last line is the whole philosophy: “Repeat monthly. This is a vital sign now, like checking your reviews. Operationalize it with a LLM visibility measurement stack.”

Trap three: interrogating once and filing it. The signals change. In August 2026, ChatGPT’s retrieval shifted and Reddit’s citation share fell off a cliff — the “why” answers from July would have named sources that stopped mattering in September. Monthly is the cadence. Put it on the calendar next to the review check.

What to do with the intel

Convert each of the three gaps into a source fix, not a chatbot fix:
- “Recommended them because of 200+ recent Google reviews mentioning slab leak work” → run a review campaign asking specifically for job-type language, not stars.
- “Their Yelp profile lists slab leak detection as a service with photos” → complete your Yelp categories and upload real job photos.
- “Mentioned on three local ‘best of’ lists” → pitch the list publishers, or earn the mentions with work worth listing.
- “Your hours conflict across two directories” → fix the source directories; the engine can’t resolve what you haven’t resolved.
Then re-run the interrogation next month and watch the “why” change. When the engine starts naming your signals unprompted, you’re winning.

The line to remember

Your competitor’s recommendation is a case file, and the engine will read it to you if you ask. Five minutes, three questions, zero dollars — “Why did you recommend them?” is the cheapest market research in the trades right now. Run it tonight, monthly after that, and fix sources instead of arguing with machines.

FAQ

Q: Will the engine actually answer honestly? A: It will answer plausibly. The explanation is generated, not a system log — treat it as a strong lead, not gospel. Cross-check against the citations in the original answer (which is why you screenshot first).

Q: Should I do this in ChatGPT, Perplexity, or Gemini? A: All three — they use different source pools and often name different winners. The procedure is identical; the intelligence differs. That’s the point.

Q: What if it recommends me and I ask why? A: Even better. You learn which of your assets is actually carrying the win, so you can protect it — and you learn the exact language to repeat in reviews, profiles, and pages.

Q: Can I automate this? A: You can script the prompts, but the judgment — which gap matters, which error to fix first — is still yours. Monthly, by hand, twenty minutes. Some things shouldn’t be delegated to the thing you’re auditing.
September 24, 2026
Claude Opus 5.5 Cuts Token Price 20% — $4 / $20 List

Last refreshed: October 6, 2026 (Pacific)

Anthropic launched Claude Opus 5.5 on September 22, 2026. Seat prices on claude.com did not move. The operator event is the unit price: Opus 5.5 lists at $4 per million input tokens and $20 per million output tokens — 20% below Opus 5 at $5 / $25.

Official sources: anthropic.com/claude-opus-5-5, platform.claude.com pricing, September 22 rate card, claude.com/pricing.

What changed

Opus 5.5 is a new SKU, not a rename of Opus 5. Anthropic says it performs at the level of Fable 5.1 on most work and costs about 40% less than Opus 5 on typical token-billed workloads because cache reads fell more than the headline rate.

Opus 5 Opus 5.5
Input / MTok $5 $4
Output / MTok $25 $20
5-minute cache write $6.25 $5
1-hour cache write $10 $8
Cache read $0.50 $0.20
Fast mode $10 / $50 $8 / $40
Batch $2.50 / $12.50 $2 / $10

Cache reads on Opus 5.5 are 0.05x input ($0.20), not the 0.1x used on Opus 5. That is the line that moves agent and Claude Code bills. US-only inference stays 1.1x on both SKUs.

What did not change

Free $0. Pro $20 monthly / $17 annual. Max from $100 (5x) and $200 (20x). Team Standard $20 annual / $25 monthly per seat. Team Premium $100 annual / $125 monthly. Enterprise $20/seat plus usage at API rates. Sonnet 5.5 remains $2 / $10. Haiku 4.5 remains $1 / $5. Fable 5.1 remains $10 / $50 with $0.25 cache reads.

Opus 5.5 is available on Pro, Max, Team, and Enterprise. Anthropic also said five-hour usage limits on those plans increased with the launch. Confirm the live window in Settings > Usage. Do not treat a help-center message count as a hard cap.

Operator note

If you are pinned to claude-opus-5, you are still on the $5 / $25 meter. Switch to claude-opus-5-5 if the workload can move. Fast mode is first-party Claude API only at $8 / $40. Cursor on-demand that routes to Anthropic list will pick up the new SKU when Cursor exposes it — Cursor seat prices on cursor.com ($20 Individual, $40 Teams) did not move with this launch.

Live desks: API rates · seat tiers · Claude Code vs Cursor.

September 24, 2026

	Opus 5	Opus 5.5
Input / MTok	$5	$4
Output / MTok	$25	$20
5-minute cache write	$6.25	$5
1-hour cache write	$10	$8
Cache read	$0.50	$0.20
Fast mode	$10 / $50	$8 / $40
Batch	$2.50 / $12.50	$2 / $10

Storm Updates: New Mexico–Plains Flood Slight, Dakotas Hail Pulse — Thursday, September 24, 2026 (Morning)

This is a live Thursday morning snapshot of local storm-type events across the United States as of the morning of Thursday, September 24, 2026. Overnight into this morning, the organized severe corridor from Wednesday’s western Dakotas Marginal Risk did verify on the Local Storm Report board — a brief tornado west of Hettinger and baseball-size hail near South Heart, North Dakota. Convective watches are not currently valid. The Storm Prediction Center 0600 UTC Day 1 Convective Outlook, valid 241200Z–251200Z, carries no categorical severe thunderstorm areas. Isolated stronger storms with hail or gusty winds remain possible late this afternoon and early evening from southeast Arizona and southern New Mexico into the Texas Panhandle and western Oklahoma and Kansas.

The water desk is still the national headline, not wind. Weather Prediction Center short-range language this morning keeps a Slight Risk of excessive rainfall over portions of New Mexico into the south-central Plains, with a broader Marginal Risk from the Four Corners into the central Plains as leftover tropical moisture and embedded shortwaves keep training cells possible on already wet ground. These are county- and ZIP-scale thunderstorms and flash-flood pulses, not a U.S. mainland hurricane landfall. Restoration exposure today is leftover water in New Mexico and the southern High Plains, isolated hail and wind on the Dakotas just-hit pile, and a Hawaii watch for Tropical Storm Nolo well south of the Big Island.

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

What is live this morning

No Storm Prediction Center convective watches are currently valid. The most recently issued watch on the SPC board is Severe Thunderstorm Watch 674, a Sunday-night eastern Colorado and western Kansas product that expired days ago. Yesterday’s western Dakotas and southern Appalachians Marginal Risk is off the Day 1 map. The 0600 UTC discussion from Guyer and Thornton is plain: organized severe thunderstorms are unlikely today and tonight. The same discussion still flags pockets of moderate destabilization into peak heating across southeast Arizona, southern New Mexico, and the southern High Plains.

The SPC overview hazard matrix later this morning lists Thursday as a Marginal severe day. Treat that as the later-cycle read against the published 0600 UTC text of “no severe thunderstorm areas forecast.” Either way, this is not a watch morning. Isolated hail or a damaging gust is the ceiling unless a new outlook or mesoscale discussion draws a tighter box this afternoon.

On the flood side, Wednesday’s Weather Prediction Center Day 1 Moderate Risk over the Texas and Oklahoma Panhandles and central and eastern New Mexico has rolled off the valid window. Thursday language from WPC’s short-range public discussion is a step down: Slight Risk of excessive rainfall over portions of New Mexico into portions of the south-central Plains, Marginal Risk across much of the Four Corners into the central Plains. Soils are still saturated. Rates, not coverage, are the problem if a cell trains.

Hotspot 1 — New Mexico into the Texas and Oklahoma Panhandles and western Kansas

This remains the water desk. Overnight, National Weather Service Dodge City issued a Flash Flood Warning for northwestern Grant, southeastern Hamilton, western Kearny, northeastern Morton, and eastern Stanton Counties in southwestern Kansas until 5:00 a.m. CDT. Radar-indicated totals of 2 to 3 inches were already on the ground at issuance, with another 1 to 2 inches possible. That warning is off the clock this morning. It is the signature of the same moisture plume that spent midweek over New Mexico.

WPC still expects scattered showers and thunderstorms Thursday as large-scale forcing weakens but instability and embedded shortwave energy remain. Portions of New Mexico and Texas stay sensitive because of the multi-day rain. Localized 1-inch-per-hour rates on burn scars, arroyos, urban streets, and low water crossings are the first failures. A separate, lower-coverage pulse is possible in southeast Arizona under the same southwesterly mid-level flow.

Restoration read: extraction and drying on buildings that took water Tuesday and Wednesday, plus a new poor-drainage ticket if a cell parks on a ZIP that already failed. Do not treat the whole Moderate-Risk footprint from yesterday as still live. Staff the county that still has a warning, not the outline from Wednesday afternoon.

Hotspot 2 — Western North Dakota just-hit pile

Wednesday afternoon’s western Dakotas Marginal Risk produced the overnight LSR pile. Preliminary Storm Prediction Center reports for the 1200 UTC September 23 through 1159 UTC September 24 window include:

A brief tornado around 2236 UTC about 3 miles south-southwest of Bucyrus in Adams County, North Dakota, based on social-media video and radar, west of Hettinger. NWS Bismarck logged the location as approximate.
Quarter-size hail (1.00 inch) at New England, Hettinger County, at 2307 UTC.
Half-dollar hail (1.25 inches) about 3 miles southeast of Belfield, Stark County, at 2340 UTC.
Baseball-size hail (2.75 inches) about 2 miles southwest of South Heart, Stark County, at 2345 UTC.

Those are official preliminary LSRs, not social rumor. Hail that size breaks windows, dents siding, and wrecks HVAC fins. The tornado report is a one-minute touchdown, not a long-track event. Restoration work there is already a survey-and-tarp job, not a new watch.

Hotspot 3 — Southern High Plains scatter plus leftover wind reports

SPC’s Day 1 discussion leaves the door open for a few stronger, locally severe storms late this afternoon and early evening from southeast Arizona and southern New Mexico into the Texas Panhandle and western Oklahoma and Kansas. That is isolated hail and gusty outflow, not a Slight Risk corridor. Yesterday’s board also carries a 58 mph thunderstorm gust 4 miles southeast of Frederick in Tillman County, Oklahoma, at 2158 UTC, and a tree down in Van Wert, Polk County, Georgia, at 2000 UTC. Those are scatter, not a line.

Farther east, an amplifying coastal low off the Mid-Atlantic and New England is a marine and coastal-flood story, not a convective watch. WPC flags coastal flooding, rip currents, and heavy rain along the East Coast into the weekend. That is a different desk from the New Mexico flood pulse.

Tropical — Nolo south of Hawaii, not a mainland landfall

Central Pacific Hurricane Center is issuing advisories on Tropical Storm Nolo. As of 2:00 a.m. HST Thursday, Nolo was about 285 miles south of South Point, Hawaii, and about 460 miles south-southeast of Honolulu, stationary, with maximum sustained winds near 50 mph and a minimum pressure of 999 mb. Official language: Nolo is forecast to intensify into a hurricane late today and could become a major hurricane by the weekend. That is a Hawaii County and offshore-waters problem, not a Gulf or East Coast landfall. Hurricane Polo and Hurricane Odalys remain in the eastern Pacific off Mexico. Tropical Storm Fay is in the open Atlantic well west-southwest of the Azores and is not a U.S. land threat.

Just hit

Wednesday’s live desk was still water over central and eastern New Mexico, with SPC trimmed to a western Dakotas Marginal. That Dakotas box verified overnight on hail and a brief tornado. The New Mexico Moderate flood risk has stepped down to Slight into the south-central Plains. Southwestern Kansas took a late-night flash-flood warning that expired at 5:00 a.m. CDT. No new convective watch fired overnight. Watch 674 is not today’s product.

Pulse table

Region	Main threats	SPC / NWS / WPC level	Status this morning
NM into TX/OK Panhandles and western KS	Flash flood, training heavy rain	WPC Slight excessive rainfall	Live flood overlay; KS FFW expired 5 a.m. CDT
SE AZ / southern NM / southern High Plains	Isolated hail, gusty wind, heavy rain	SPC 0600 UTC: no categorical area; isolated stronger storms possible	Afternoon–evening scatter
Western ND (Hettinger / Stark)	Hail damage, brief tornado	Yesterday Marginal; LSRs in	Just-hit / survey
Four Corners to central Plains	Isolated heavy rain	WPC Marginal excessive rainfall	Diurnal cells
East Coast marine / New England coast	Coastal flood, wind, heavy rain	WPC / OPC coastal low	Not a convective watch
Hawaii — Nolo	Tropical storm to hurricane risk south of the islands	CPHC TS Nolo, 50 mph	Watch Hawaii products, not mainland
Atlantic / Gulf vs. U.S. mainland	No landfalling hurricane	Quiet on the mainland coast	Quiet

What this means for restoration teams

New Mexico and the southern High Plains are still a water week. Saturated ground plus another Slight Risk means the next training cell is a wet-drywall and crawlspace ticket, not a surprise. Photograph and extract what already flooded before you chase a new ZIP. Southwestern Kansas counties that sat under last night’s warning should be treated as just-hit poor-drainage work until local offices drop the flood language.

Western North Dakota is a hail desk. Baseball stone southwest of South Heart is broken glass and bruised roofs. The Adams County tornado report is a narrow path. Staff it as a local damage survey. Do not read “tornado” on an LSR board as a regional event.

If the later SPC cycle firms a Marginal box over the southern High Plains, the work is still isolated hail and outflow, not a watch. Stage tarps only where a warning is live. Hawaii shops should follow Central Pacific Hurricane Center products on Nolo, not this CONUS flood pulse.

What this is not

This is not a regional derecho and not a U.S. mainland hurricane landfall. Polo and Odalys are eastern Pacific storms off Mexico. Fay is mid-Atlantic open water. Nolo is a Central Pacific storm still south of the Big Island. The CONUS work this morning is leftover flood water, a North Dakota hail pile, and a quiet watch board.

Looking ahead

SPC Day 2, valid Friday, currently carries no severe thunderstorm areas. WPC still sees the moisture axis shifting east Friday into the south-central Plains and Midwest with another heavy-rain overlay possible. SPC Day 3 language returns a Marginal Risk later in the weekend as an upper trough works the central United States. Watch the afternoon Storm Tracker rather than assuming Thursday’s quiet watch board is the last line of the week.

Sources and how to verify

Primary sources: NOAA Storm Prediction Center 0600 UTC Day 1 Convective Outlook for Thursday, September 24, 2026; SPC current watches page (no watches valid as of 0705 UTC); SPC today’s storm reports for 20260923 1200 UTC–20260924 1159 UTC; Weather Prediction Center Day 1 Excessive Rainfall Outlook and short-range public discussion valid 12Z Thursday through 12Z Saturday; National Weather Service Dodge City flash-flood warning text for southwestern Kansas; Central Pacific Hurricane Center / National Hurricane Center public products on Tropical Storm Nolo as of 2:00 a.m. HST Thursday.

Check weather.gov for the county warning, spc.noaa.gov for the outlook, watch, and LSRs, and WPC excessive rainfall for the flood overlay. Prior desk: Storm Updates: New Mexico Flood Moderate, Western Dakotas Marginal — Wednesday, September 23, 2026.

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

September 24, 2026

The Consistency Dividend

“The highest-ROI marketing work is also the most boring. That’s not a coincidence.”

Name. Address. Phone. Identical everywhere. That’s the whole piece, and it’s worth more than the last three marketing tactics you tried combined — because nobody does it, because it’s boring, and boring is exactly where the edge lives.

The witnesses

Your business doesn’t exist in one place. It exists in dozens, and each one is a witness testifying about who you are and where to find you.

The Google profile. The website footer. Yelp, Facebook, Angi, the BBB. The directories you claimed in 2017 and forgot. The truck door. The invoice template. The email signature. Every one of them says your name, your address, your phone number — or it says something close, which is worse.

Nobody audits the witnesses. That’s the problem, and the opportunity.

The leaks

Here’s what the witnesses are saying right now, on profiles all over town:

“123 Main St” on Google, “123 Main Street” on Yelp, “123 Main St Suite B” on Facebook — three addresses for one door. The old cell number still on the Angi listing from before the voice line. The suite number on the website, missing everywhere else. The Facebook page from 2016 with the previous address, still ranking, still confusing people.

Each mismatch is small. Together they’re a credibility leak. The homeowner comparing two contractors doesn’t think “NAP inconsistency” — they think “something feels off about this one,” and they can’t say why. The search engine doesn’t think in words at all — it just has less confidence that all these listings are the same business, and confidence is the currency.

Small leaks, everywhere, all the time. That’s what boring neglect looks like.

The dividend

Now flip it. Every place that agrees is a vote.

Same name, same address, same phone — on the profile, the site, the directories, the truck, the invoice. Each matching witness raises confidence: the human’s (“these people have their act together”) and the machine’s (every corroborating listing makes the entity clearer). Trust isn’t built in one place. It’s the sum of a hundred small agreements.

That’s the dividend: not a spike, a yield. It pays a little every day, in every search, in every comparison — the quiet background hum of a business that agrees with itself. You don’t notice it working. You notice when it’s missing.

The audit

The work is unglamorous, which is why I’m spelling it out:

Write down the canonical version — one name, one address format, one phone number. Not the pretty version, the exact version: St or Street, suite or no suite, which number. Then list every witness: every profile, every directory, the site, the truck, the invoices, the signatures. Then fix every mismatch, one by one, until they all testify the same.

Then maintain it. New directory? Canonical version goes in. New truck? Canonical version on the door. New phone system? Every witness gets updated the same week, not “when we get around to it.”

It’s an afternoon of tedium, twice a year. That’s the whole price.

The boring moat

Here’s why this is a moat and not just hygiene: your competitors won’t do it.

Not because they’re lazy — because it’s boring, and boring doesn’t feel like marketing. Marketing feels like a new website, a new ad campaign, a new something. Nobody gets excited about making the suite number match in fourteen places. So nobody does it. The field stays sloppy, and the one business that agrees with itself everywhere stands out without spending a dollar.

Every real edge I’ve ever seen looked boring from the outside. This one just happens to look boring from the inside too.

The close

Name. Address. Phone. Identical everywhere.

Boring is the moat. Consistency is the dividend. And the businesses collecting it are the ones whose witnesses all tell the same story — the story of a business that has its act together, down to the suite number.

September 23, 2026
Teach What You Know, Sell What You Do

“The best marketing a contractor can do is a lesson.”

It’s 11pm. There’s water where it shouldn’t be, and a homeowner is searching “is this mold dangerous” with one thumb while the other holds a flashlight. Whoever taught that lesson — in plain words, on a page that reads like a human wrote it — is getting the call in the morning.

Not the contractor with the biggest ad budget. The one who taught.

The 11pm search

Every job starts as a question. Not “who should I hire” — that’s the last question. The first ones are fear questions: is this dangerous, is this urgent, did I cause this, what happens if I wait until Monday.

The contractor who answers those questions in public owns the beginning of the job. By the time the homeowner is comparing bids, the teacher is already the trusted one — the others are strangers with prices. Trust compounds faster than discounts ever will.

Teaching doesn’t give away the trade

The objection is always the same: “if I teach them, they’ll do it themselves.”

They won’t. Nobody reads a post about flood cuts and then goes and does their own mitigation. What the lesson actually does is the opposite — it shows how much there is to know. The homeowner reads three paragraphs about moisture mapping and thinks “I had no idea it was this involved,” and that thought is worth more than any ad. The lesson doesn’t create DIYers. It creates respect for the trade, and respect books jobs.

You can’t give away what takes ten thousand hours to learn. You can only prove you learned it.

The job site is the classroom

Here’s the part nobody believes until they try it: every single job has three lessons in it.

The moisture meter photo — what the number means and why the wall that looks dry isn’t. The flood cut — why it stops exactly there, and what happens when it doesn’t. The thing the homeowner did that made it worse — the fan pointed at the wrong wall, the bleach on the porous surface, the week of waiting. Each one is a lesson someone is searching for tonight.

You don’t need a content calendar. You need a habit: finish the job, teach one thing. The classroom is wherever the work happened.

The cadence

One job, one lesson. That’s the whole system.

Not three posts a week — that’s a marketing department, and you don’t have one. One lesson per job, written the way you’d explain it leaning on the tailgate. A photo from the site, a few plain paragraphs, the thing you wish every homeowner knew before they called. Fifteen minutes, while the job is still fresh enough to teach honestly.

Miss a week and nothing breaks. The lessons don’t expire — a flood cut works the same in 2029. The only cadence that matters is: never let a good lesson die in the truck.

The compound

Lessons stack. That’s the whole game.

Ten lessons is a resource. Fifty is a library. A hundred is the site homeowners find at 11pm, the one the search engines cite, the one the AI answers quote when someone asks what’s happening inside their wall. Teachers get cited. Teachers get called. Teachers get trusted before the bid ever goes out.

And every lesson is a neuron in the knowledge base — yours first. The contractor who teaches for a year doesn’t just have better marketing. He has a better-organized head: the trade, distilled, searchable, in his own words.

The close

Teach what you know. Sell what you do.

The lesson is the marketing; the job is the product. Every job site is a classroom, every fearful 11pm search is a student, and the contractor who teaches is the one who gets the call in the morning.

Nobody ever hired the ad. They hired the teacher.

September 23, 2026
Your Google Profile Is Your New Front Door

“Nobody visits your website first. They meet your front door.”

Search the trade plus the town and look at what comes up before anything with your URL on it: the business profile. The hours, the photos, the stars, the questions, the call button. That’s the first impression, and for most customers it’s the only one — they never walk past the door into the house.

Your website is the house. The profile is the front door. Nobody’s impressed by the house if the door is boarded up.

The door inventory

Walk up to your own front door like a stranger and read what’s on it:

The hours — including the holiday hours, the ones that are wrong on half the profiles in America right now. The photos — the truck, the crew, the work, or a gray empty storefront from 2019. The reviews — stars, words, and whether anyone from the business ever answered back. The questions — asked by strangers, answered by strangers, when nobody from the business is home. The posts — the weekly update slot, empty since the profile was claimed. And the two big brass buttons: call, and directions.

That’s the door. Every customer reads it before they knock.

The untended door

Here’s what most front doors look like: hours that lie on holidays. Photos older than the crew in them. A Q&A section where a stranger asked “do you do water damage?” eight months ago and another stranger answered “idk.” No posts — the business has done a hundred jobs since the profile went up and the door shows none of them. Reviews sitting unanswered, the digital equivalent of mail piling up in the slot.

And the doorbell — the call button — still works. It rings. Right into the voice line, right into the tuition piece. The door and the phone are the same system: the profile is where they decide to knock, the line is what answers.

An untended door doesn’t just lose the knock. It sends the customer to the next door on the street — the competitor whose hours are right and whose photos are from this year.

Tending the door

The good news: tending a front door is a fifteen-minute weekly ritual, not a project.

Fresh photos — the actual truck, the actual crew, the actual work from this month. A door with fresh photos says “we’re alive in here.” Check the hours — especially before holidays, the highest-traffic lying season. Work the Q&A — seed the questions customers actually ask, answer them in your own voice, so strangers don’t do it for you. One post a week — the job you finished, the storm you worked, the crew milestone. It’s the shop window; put something in it. Answer the reviews — every one, but especially the good ones, because the response is the business talking back through the door.

Fifteen minutes. The highest-traffic page in the business, tended.

The compound

Here’s what makes the door different from every other marketing chore: it compounds and it doesn’t decay.

A post you write stays up. A question you answer stays answered — every future stranger with the same question reads your answer instead of a stranger’s guess. A photo you add joins the set. Reviews you respond to stack into a record of a business that talks back. Nothing you do to the door un-does itself. It’s all permanent, all cumulative, all working while you sleep.

Most marketing is rent — stop paying, it stops working. The front door is owned. Every fifteen minutes you spend on it is still there next year.

The close

Your website is the house. Beautiful, expensive, and visited second — if ever.

The profile is the front door. It’s what they see from the street, it’s where they decide, and the doorbell on it rings straight into your line. Tend the door. Sweep the step. Put something alive in the window. Answer when they knock.

Nobody ever hired the house. They hired the door that looked like somebody was home.

September 23, 2026
Write It Down or It Didn’t Happen

“I don’t publish to be read. I publish to remember.”

That sounds backwards until you’ve watched a Tuesday fade. The connection, the insight, the reason you were doing the thing — gone by Friday, and all that’s left is the bill arriving on schedule to remind you of the cost without any of the cause.

So I write it down. And then I publish it. Those are two different acts, and they do two different jobs.

The fading

Here’s the mechanics of forgetting, and it’s not a character flaw — it’s just how the hardware works.

The bill arrives every month: exact, itemized, impossible to ignore. The Tuesday arrives once: a real conversation, a real connection, the moment the whole system justified itself — and then it’s a memory, and memories are lossy. Every retelling compresses. Every month thins it out. Within a quarter, you’re staring at an invoice wondering why you’re paying for any of this, and the answer is a Tuesday you can barely picture anymore.

The written page doesn’t fade. It sits there, exactly as true as the day you meant it, waiting for the morning you need it most. That’s not content. That’s a rescue line thrown to your future self.

The site as a second brain

Every article is a neuron.

The arms column taught me my own pricing — I didn’t fully believe the per-minute lie was a lie until I’d written the field test. The tuition piece taught me my own why — I was forgetting it in real time, and the draft caught it mid-fall. I didn’t write those pieces because I knew those things. I know those things because I wrote those pieces.

That’s the direction everyone gets backwards. Writing isn’t the record of thinking. Writing is the thinking. The draft argues with you. Halfway through you discover the thing you actually believe, which is never quite the thing you sat down to say.

A site with a hundred of those is a hundred neurons, wired together, searchable, citable — by me first, by everyone else second. It’s not a blog. It’s the outside of my head.

The receipts

There’s a second job the published page does, and it’s for a different audience: proof.

Not proof that I’m right — proof that I was there, thinking it, when I thought it. Dated, timestamped, in public. For the future partner doing diligence. For the future auditor asking “what did you actually believe back then.” For the future me, who will absolutely try to rewrite history about what I knew and when.

Memory edits. The published page doesn’t. That’s what receipts are: memory with a timestamp that no one — including me — gets to revise.

The loop

Here’s the whole machine, and it’s been running for years before I had a name for it.

I live — the calls, the jobs, the Tuesdays. I write — the draft that argues with me until the real belief surfaces. I publish — the neuron goes up, dated and public. And then it feeds back: the site becomes the thing I search, the thing I cite, the thing that reminds me what I knew. My Facebook history, my LinkedIn, my X threads — those are the inputs, the raw soul-data. The articles are the outputs, the refined version. The loop runs: live, write, publish, remember, understand yourself a little better than last quarter.

Most people publish to be perceived. I publish to perceive — myself, clearly, before the fading gets it.

The close

Write it down or it didn’t happen.

Publish it or you’ll forget you knew it. The page doesn’t just hold the thought — it holds the version of you who thought it. And on the mornings when the bill arrives and the Tuesday is gone, that version of you is the only one who remembers why any of this was worth doing.

He wrote it down. That’s why.

September 23, 2026
The Junk Calls Are the Tuition

“Most of the calls are garbage. Listing bots, spam, junk — and I’m paying for every one of them. But all it takes is one. One real person, one real conversation, and it pays for all of them.”

That’s the whole piece. But it took me a year of phone bills to learn it, and about five minutes of forgetting it, so I’m writing it down.

The tuition frame

Every system has tuition. Ad spend has click fraud. Email has spam filters and the good leads that land in them. The voice line has junk calls.

Tuition isn’t a scam — it’s the price of the classroom. The question was never whether I’d pay it. The question was whether I’d remember what the classroom was for.

What the junk actually costs

Here’s the part that stings, and it’s straight from the arms column: the bill doesn’t care whether the call mattered.

The listing bot that calls to sell me a listing. The robocall about my car’s warranty. The silence. Every one of them spins up the model, opens the carrier leg, records the nothing, transcribes the nothing. The arms fire either way. I pay for the whole stack to handle a call that never existed.

Multiply that by a month and the tuition line on the invoice is real. I’m not going to pretend it isn’t.

The math of the one

But here’s the other column, the one the invoice doesn’t print.

One call. A real person, a real problem, water where it shouldn’t be. They talked to the line instead of bouncing to the next listing. Somebody answered — well, something answered — and it sounded like a human who gave a damn, and by the end there was a name, an address, and a job on the calendar.

One of those pays for months of junk. Not close — completely. The asymmetry is so lopsided it looks like a rounding error until you run the year.

The junk calls cost arms. The one call buys the whole armory.

Why the why fades

The bill arrives every month. The connection was a Tuesday.

That’s the whole problem. The tuition is invoiced on schedule; the reason is a memory. And memories fade faster than bills do. So every few months I catch myself staring at the junk-call line and thinking “why am I paying for this” — and the answer is always the same Tuesday I forgot.

Systems don’t run on memory. They run on what’s written down. So this is me writing it down: the junk is the tuition, the one call is the classroom, and the day I forget that is the day I start optimizing the wrong thing.

The filter question

Notice what the answer isn’t. It isn’t “block the junk.”

A filter aggressive enough to stop every bot is aggressive enough to stop a human — the tired homeowner who mumbles, the bad connection, the caller who sounds like a robocall for the first four seconds because they’re reading the address off a piece of paper. The door has to stay open. That’s the entire point of the door.

The right question isn’t how to stop the junk. It’s how cheap the junk can get while the door stays wide open: faster hangup detection, quicker routing, less model time burned on the obviously-empty calls. That’s harness work — making the tuition cheaper, not pretending school is free.

The close

The junk calls are the tuition. Pay it gladly.

Just don’t forget what the classroom is for. It’s for the one. It’s always been for the one.

September 23, 2026
The Arms Column, Field-Tested

“We said you’re not buying minutes — you’re buying arms. Then the calls started flowing. Here’s what the bill actually taught us.”

A while back I argued that voice-AI pricing is a lie: the per-minute number on the pricing page isn’t the product. The product is a stack of arms — the voice intelligence, the carrier connection, the infrastructure around them — and the per-minute price is just the costume they wear.

That was the theory. This is the field test.

What the bill actually says

Run a real week of calls and read the invoice the way an owner reads it — not the headline rate, the total. The per-minute number is almost never the biggest line. The arms are.

The voice model doing the talking. The carrier moving the audio. The platform orchestrating the whole thing — the number, the recording, the transcript, the handoff. Each arm bills its own way, on its own meter, and the “per minute” quote only ever described one of them.

Nobody lied to you. They just priced the costume and shipped the wardrobe.

The concurrency math nobody shows you

Here’s what the field test really exposes: minutes are linear, arms are not.

Ten simultaneous calls isn’t ten times the per-minute rate in value — it’s ten arms, all live at once. The pricing page shows you a single call’s minute. Your Monday morning shows you ten calls overlapping, each holding its own model session, its own carrier leg, its own recording pipeline open.

The vendor priced the minute. You bought the rush hour. Those are different products, and only one of them shows up when the phones light up.

You pay for arms even when the call goes nowhere

The wrong number. The three-second hangup. The caller who wanted the pizza place. The silence where someone pocket-dialed you.

Minutes barely moved. The arms all fired anyway — the model spun up, the carrier connected, the platform recorded forty seconds of nothing and transcribed it faithfully. You paid for the whole stack to handle a call that never existed.

This is the line the per-minute lie can’t survive: the bill doesn’t care whether the call mattered. The arms do the work either way. Price the arms, or the junk calls price you.

The only math that matters

Stop dividing by minutes. Start dividing by outcomes.

Take a real week: total voice bill, all arms included, divided by minutes — that’s the advertised number, and it’s trivia. Now divide the same total by resolved calls. Then by booked jobs. That last number is the only one that touches revenue, and no vendor puts it on the pricing page because no vendor controls it — you do, with your harness.

A vendor quoting two cents a minute against a vendor quoting five is a meaningless comparison until you know whose stack resolves the call. The cheap minute that books nothing is the most expensive minute you’ve ever bought.

What to ask a vendor now

After the field test, there are three questions, and a vendor’s answers tell you everything:

Break the bill into arms. What’s the model cost, the carrier cost, the platform cost — separately? If they can’t or won’t, you’re buying a bundle, and bundles hide margin.

What does my rush hour cost? Not a minute — my Monday at 8 AM, ten calls deep. If the answer is “the same per-minute rate,” they haven’t thought about it, which means you will.

What do I pay for the call that goes nowhere? The hangup, the wrong number, the silence. If everything bills the same whether the call mattered or not, the arms are priced — the minute is just the label.

The close

Minutes were never the product. The product is an answered call that ends in a booked job — and that’s built from arms, priced in arms, and won or lost in the harness around them.

The pricing page will keep selling minutes. Let it. You know what you’re buying now.

Buy the arms. Price the outcomes. Own the harness that turns one into the other.

September 23, 2026
The Phone Is the Office

“The phone is the office.”

Not an app. Not a dashboard. Not a portal. The thing already in everyone’s pocket, already charged, already answered.

The decision

When it came time to pick the interface — the way humans would actually touch the system — the candidates were an app, a chat platform, and the phone. Voice in, SMS out.

The phone won, and it wasn’t close.

Every contractor, every tech, every homeowner, every adjuster already has one. Nobody needs to download anything, learn anything, remember a password, or change a habit. The interface is the thing they were already holding.

Why apps lose

Every app is a behavior change wearing a friendly icon. Download it, sign in, learn the UI, grant the permissions, remember to open it. Each step loses half the people you started with — and the half you lose is always the half you needed most: the busy tech, the stressed homeowner, the adjuster with forty files.

An app can do more. That’s the pitch, and it’s true, and it’s irrelevant. Reach beats richness. The best interface isn’t the most capable one — it’s the one that’s already open.

Why the platform lost

The chat platform was tempting — everyone’s already there, the tooling is good. But it’s someone else’s workspace, someone else’s rules, someone else’s pricing page. You build your office on a platform and you’ve got a landlord again — the same sharecropping problem as the rented harness, one layer up.

The phone is nobody’s platform. Or everybody’s, which amounts to the same thing. No terms of service can take your phone number’s habits away. No pricing change makes people stop answering calls.

What it means for the office

The dispatcher doesn’t learn software. They talk.

The tech doesn’t open a ticket. They text a photo of the meter readings from the driveway.

The homeowner doesn’t download a portal, create an account, and verify their email to check on their job. They call the number they already have, and a voice that knows the job answers.

Zero install. Zero behavior change. Zero training. That’s not a feature list — that’s the whole strategy.

The glass and the system

Here’s the part people miss: the phone is the glass, not the system.

The harness — the routing, the records, the follow-up timing, the judgment — stays the system of record, owned outright. The phone is just how humans touch it. Glass is swappable; the system underneath is yours. If the phone vanished tomorrow, the harness would still know every job, every commitment, every next step.

That’s the pairing: harness-first underneath, phone-first on top. Own the operation, meet people where they already are.

The objection

“But a real system needs a real interface.” It has one. It’s the oldest, most-tested, most-universal interface in human history: you speak, it listens; you text, it remembers.

Fancy is a tax on adoption. Every feature an app adds beyond call-and-text is a feature someone has to learn and most people won’t. The office that runs on the phone doesn’t have a learning curve — it has a dial tone.

The close

The office isn’t a place anymore. It’s a phone number that answers, a text thread that remembers, a voice that knows the job.

Build the system. Own the harness. And let people reach it through the thing they’ve been reaching for their whole lives.

The phone is the office.

September 23, 2026