Tag: The Working Years

  • The Working Years — Episode 2: The Lantern Principle

    The Working Years — Episode 2: The Lantern Principle

    The Working Years is a series drawn from my own archive — posts I wrote years ago, given the room to become the articles they were trying to be.

    The seed

    On July 26, 2025 — the 27th in the archive’s UTC clock — I posted this to my @willtygart account, the one I later deleted:

    “The new gold isn’t answering the questions people ask. It’s illuminating the ones they can’t yet vocalize.”

    It went up at 10:24 PM Pacific, two minutes after a sibling post that linked the Native Data essay: “Being the closest store gets you noticed. Speaking the local language gets you chosen.” Two posts, two minutes apart, one night. The Lantern essay names the Oxxo Principle — being the closest store — as the first layer; no Oxxo essay survives in the archive, but the idea lived on as the tagline of the Native Data post. Two essays got their own subdomains: the Native Data Principle and the Lantern Principle. The Lantern one carried the title “The Lantern Principle: The Final Layer of SEO.”


    What happened afterward

    Within days, the Lantern post had become a full essay. The earliest surviving capture is July 30, 2025 — four days after the post — and it’s already complete: the hero, the comparison, the four-step method, the whole thing. So the thinking wasn’t new on the 26th. The post was the tip of something I’d already worked through.

    Here’s what’s worth noticing about that July burst. The two posts that night were one idea unfolding in layers:

    1. The Oxxo Principle — “Being the closest store gets you noticed.” Proximity. Be there.
    2. The Native Data Principle — “Speaking the local language gets you chosen.” Relevance. Speak like the customer.
    3. The Lantern Principle — the final one. Anticipation. Light the path they can’t describe yet.

    Be there. Speak their language. Then — the hardest one — see the problem they can’t articulate and hand them clarity anyway.

    The subdomains are offline now. The essays survive only in the Wayback Machine. That itself is a small footnote about rented space versus owned space, which is a different episode. The point for this one: the idea outlived its hosting. It turned out to be early, not wrong.

    Because look at what the web became. The AI assistant era made the Lantern Principle the default shape of a good answer. When someone asks a chatbot a question now, the good systems don’t just answer — they anticipate. They offer the next step, the related consideration, the thing the user was really getting at. “What is the underlying problem they are trying to solve?” — that’s not a 2025 content strategy anymore. It’s how the good ones behave now. The principle went from marketing theory to product behavior in about a year.


    The piece itself

    Here’s the essay as it ran, cleaned up from the archive — I haven’t rewritten it, just given it the room it was always asking for.


    For decades, we treated content as a reactive tool. A user asks, we answer. Simple transaction. But that model assumes the user knows what to ask. When they don’t — and they often don’t — they get stuck. They bounce. They stay frustrated.

    Consider the difference:

    Answering the question: “How do I change a tire?” → “Here are the 5 steps to change a tire.”

    Illuminating the path: The person’s unspoken reality is “I’m stranded and stressed.” So the answer becomes: the 5 steps, plus safety precautions, plus a link to 24/7 roadside assistance, plus how to check the spare’s pressure. Nobody searched for those extra things. Everybody needed them.

    The greatest opportunity in content isn’t in the keywords people search for. It’s in the needs they can’t yet articulate. The new gold is found in the dark — in the space between a user’s problem and their ability to ask for a solution.

    This is the Lantern Principle: stop being a dictionary, start being a guide. Our job is no longer just to provide answers but to anticipate needs — to create content that doesn’t just solve the stated problem but illuminates the entire context around it, guiding the user to clarity and confidence.


    How to build a lantern

    The shift is from keyword research to empathy mapping. The operative question changes from “What are people searching for?” to “What is the underlying problem they’re trying to solve?” Four moves:

    1. Answer the unasked question. Someone searching “how to write a resume” is really asking “how do I get a better job?” Serve both. Resume templates and the interview tips and the career planning resources. The stated query is the door; the unasked question is the house.
    2. Provide the next step. Never let content be a dead end. Every article should lead naturally to the next logical step in the user’s journey. A product page links to its user manual. A tutorial links to the advanced technique. If you end at the answer, you’ve ended too early.
    3. Simplify the complex. The ultimate act of empathy is taking a complex, intimidating topic and making it simple — analogies, plain language, visuals that actually explain. This is what builds the trust that makes you the go-to source.
    4. Create foundational resources. Build definitive, comprehensive guides — “digital lanterns” — that cover a topic so thoroughly they become the starting point for anyone exploring it. These serve thousands of unasked questions over time. They compound.

    The goal: a web that feels less like a vast, cold library and more like a network of helpful guides, each holding a lantern. An internet that doesn’t wait for the perfect query but proactively offers clarity.

    The future of content isn’t about being found. It’s about shedding light.


    Why this one aged well

    I want to be honest about what this principle got right and what it didn’t.

    What it got right: anticipation is the real moat. Everything I wrote about SEO in 2025 assumed the user arrives with a query and the job is to match it. The Lantern Principle was me noticing, before I had the language for it, that the highest-value content serves the need behind the query — and that AI systems would eventually do this natively. When an assistant now reads between the lines of a question and offers what you actually needed, that’s the Lantern Principle running as software.

    What I understated: how hard anticipation is to fake. A lantern only works if you genuinely understand the person in the dark. The four moves above are easy to write and hard to do — because “answer the unasked question” requires you to actually know people, not just their search terms. Every content farm can optimize for keywords. Very few can hold the lantern, because it requires the one thing that doesn’t scale: empathy for a specific human being stuck on a specific problem.

    That’s also why this principle pairs with the one from Episode 1. In the 23-model experiment, I learned that scope — who the reader is, what they actually need — matters more than the prompt. The Lantern Principle is scope taken to its conclusion: know the reader so well you can answer what they haven’t asked.


    The restoration version

    I run a niche agency for restoration contractors, so let me ground this where it hurts.

    A homeowner never searches “I need emergency water mitigation with proper psychrometric documentation for my insurance claim.” They search “water under sink” or they call because the kitchen smells weird. They are stranded and stressed, and they cannot vocalize the real need: someone who will stop the damage, document it so insurance pays, and explain what’s happening in plain language.

    The restoration company that answers the stated query — “yes, we do water damage” — is the dictionary. The one that anticipates — shows up, explains the process before the adjuster calls, hands the homeowner a clear next step at every stage — is the lantern. Guess which one gets the review, the referral, and the adjuster’s trust.

    That’s not a marketing insight. It’s an operating insight. The companies winning in restoration right now are the ones whose process illuminates the path, not just their website.


    Related from the series

    • Episode 1: We Ran 23 AI Models on the Same Article. The Prompt Was Never the Point. — scope beats the prompt; knowing the reader beats optimizing the query.
    • The Native Data Principle — speaking the local language gets you chosen. The middle layer of the trilogy.
    • The Oxxo Principle — being the closest store gets you noticed. The first layer, named in both essays; no Oxxo essay survives in the archive.
    • Stop Renting Space. Start Owning Your Content. — on why the essay subdomains being offline is a footnote, not a funeral.
  • We Ran 23 AI Models on the Same Article. The Prompt Was Never the Point.

    We Ran 23 AI Models on the Same Article. The Prompt Was Never the Point.

    From the archive

    July 4, 2025 — I posted this on X:

    We tested 23 AI models on the same structured article scope — not just a prompt, but a data-backed framework with embeds, search mapping, and internal RAG.

    Each model got the same Google Doc. Same structure. Same instructions.

    July 5, 2025 — the follow-up, with the writeup:

    We gave the exact same prompt to 23 different AI models. Same input. 23 different outputs. 23 different personalities.

    The takeaway? It’s not just how you prompt — it’s who you’re talking to.

    Fourteen months later, the model names in the results table are history. The finding got more true, not less. Here’s the whole thing, with room to breathe.


    What happened since

    When I ran this test, Claude Sonnet 4 was the new hotness and Grok 3 was still in beta. The industry has turned over since then — the names below belong to mid-2025. Every specific ranking is a fossil, dated July 2025, preserved as-is.

    But the doctrine this experiment produced is now the operating system for everything I build:

    • Model selection beats prompt engineering. The gap between the best and worst model on the same scope was bigger than any prompt trick I knew.
    • Models have personalities. Not metaphorically — operationally. The same way you’d hand a delicate contents job to a different tech than a Category 3 demo.
    • Match the model to the job. Emails, SOPs, longform, code — different work, different worker.

    If you’ve read anything I’ve written about AI costs in 2026, you’ve seen this doctrine wearing different clothes. Model choice is cost choice. The most expensive mistake in AI isn’t a bad prompt — it’s the wrong model doing the wrong job at the wrong price.


    The experiment

    The setup mattered more than the scores. This wasn’t “ask 23 chatbots a question and see who sounds smartest.” Every model got:

    1. The same Google Doc — a structured article scope for “Kitchen Fire Cleanup: Expert Restoration and Prevention Tips,” a real topic from my industry, not a toy prompt.
    2. The same structure and instructions — no per-model tuning, no optimizing for quirks. Deliberately unfair in the fairest possible way.
    3. A data-backed framework — embeds, search mapping, and internal RAG behind the scope, so the test measured how models work with structure, not how they freestyle.

    Why a restoration topic? Because generic benchmarks test generic thinking. I wanted to know which models could handle domain work — regulated language, technical accuracy, a reader who’s trusting you with their home. That’s a harder test than poetry.

    Custom agents ran the harness. The models just had to do the job.


    The results (July 2025 — preserved)

    Same input, 23 different answers — and size wasn’t the differentiator.

    The 5 that stood out

    ModelScoreWhy it won
    Claude Sonnet 4 (Extended)5.0Calm, professional, field-ready tone
    GPT-4.15.0Sharp structure, publishable polish
    Claude 4 Opus (regular)5.0Great flow and clarity
    Claude Sonnet 45.0Excellent default performance
    o4 Mini High Effort5.0Lightweight but surprisingly strong

    These were the ones I’d have trusted, back in July 2025, with full-length articles, client emails, and education pieces. Note the last one: a mini model scored a perfect 5. Size wasn’t the differentiator. Fit was.

    Most improved, second round

    ModelBefore → After
    Claude 3.54.2 → 4.75
    o3 Mini High Effort4.5 → 4.8
    LLaMA 4 Maverick3.5 → 4.6

    Second-round testing with better scope alignment lifted every one of them. The lesson I keep coming back to: even machines do better when they’re understood. The fix usually isn’t a better model — it’s a better briefing.

    The full field

    ModelFinal score
    Claude 3.7 Extended4.9
    Gemini 2.5 Flash4.9
    Gemini 2.5 Pro4.9
    Grok 3 Mini High Effort4.9
    GPT-4o4.85
    GPT o34.85
    Qwen 2.5 72B4.85
    DeepSeek V34.85
    Qwen3 235B4.85
    Grok 3 Beta4.8
    Auto (router)4.8
    GPT-4.1 Mini4.6
    4o Mini4.6
    Mistral Large 24.5
    LLaMA 4 Scout4.1

    Two things worth noticing in the middle of the pack. First, the open models (DeepSeek V3, Qwen) hung with the frontier labs on structured domain work — the gap was narrower than the marketing suggested. Second, the auto-router scored 4.8, within spitting distance of the best hand-picked models. The machines were already learning to choose among themselves.


    The takeaway, then and now

    Then (July 2025): “It’s not just how you prompt — it’s who you’re talking to.” Learn the personalities the way you learn your field crew. Some need bullet points. Some need a whiteboard. Some just need to be trusted to go build.

    Now (September 2026): That sentence became a cost doctrine. Every model has a price per token and a personality per task, and the expensive failure mode is mismatch — a frontier model writing a two-line email, a mini model drafting your scope of work. The experiment’s real output wasn’t a ranking. It was a routing table.

    Here’s the version I’d hand a contractor today:

    1. Run your own version of this test. Not 23 models — three. Your best guess, the cheap one, and the weird one. Same job, same brief. You’ll learn more in an afternoon than in a month of prompt tweaking.
    2. Write down the personalities. Which model do you trust with numbers? With tone? With structure? That’s your routing table. Tape it to the wall.
    3. Re-run it when the generations turn. The names change; the method doesn’t. This is maintenance, not a one-time project.

    And if you ever wonder whether AI is broken or you’re just doing it wrong — remember the test. Same input, 23 different answers.

    It’s not about perfection. It’s about alignment.


    Series notes

    Episode 1 of The Working Years — ideas pulled from the 2022–2025 X archive (@willtygart, account since deleted), given room to breathe. The original posts are quoted verbatim from the archive export. The July 2025 beehiiv writeup “We Ran 23 AI Models on the Same Article Scope” carries the full original tables.

    Next in the series: The Lantern Principle — “the new gold isn’t answering the questions people ask.”