Why Does AI Get My Business Wrong When People Ask About It?

Because you never wrote it down. Your business is real, and to a machine it’s a rumor.

The story

A client rang me genuinely upset. Someone had asked ChatGPT about her firm and it had confidently described a company that didn’t exist. Wrong services. Wrong specialism. A location she’d left in 2021. It had invented a second office.

Her first instinct was that she’d been defamed by a robot.

So we went looking. And here’s the thing — every single wrong detail had a source. The old location was still on two directory listings and the footer of a PDF price list from 2020 that was still sitting on her own server. The wrong specialism came from a guest article she’d written years ago about work she no longer did. The second office was a registered address that had never been an office at all.

The model hadn’t made anything up. It had done exactly what it was designed to do: gather what exists, weigh it, and produce the most corroborated story. The most corroborated story about her business was four years out of date, because she’d never published a better one.

She’d told me all the correct information in ninety seconds on the phone. Every bit of it was true, current, and completely undiscoverable. It lived in her head.

To a machine, that means it doesn’t exist.

Here’s the direct answer

AI describes your business incorrectly because it can only work with what you’ve published, and most businesses have never written down what they actually do in a form a machine can read. Your real knowledge lives in people’s heads, in PDFs, in inboxes, and in a homepage that says “innovative solutions.” AI-readability means giving machines one clear, structured, current, corroborated source of truth. sosSTEVIE builds that source of truth — the foundation your AI search visibility, your chatbot and your automation all sit on.

Key Takeaways

AI doesn’t lie about you. It reports the most corroborated version of you that exists in public — usually four years old.
Undocumented knowledge is invisible knowledge. If it’s only in your head, it isn’t real to a machine.
Contradictions are worse than gaps. Three different addresses makes you an unreliable entity.
This is the prerequisite. Chatbots hallucinate and AI search ignores you for the same root cause.
Depth beats keywords. What gets cited is genuinely useful material, not optimized thin pages.

Where your knowledge actually is

Run this exercise. It takes four minutes and it’s uncomfortable.

Ask yourself who your best-fit client is, what you do that competitors don’t, what you charge, what you won’t take on, and what happens in week one of an engagement.

You know all of it instantly. Now: where is any of that written down, publicly, in a form a machine can read?

A wall of scattered handwritten notes beside a single structured document, representing undocumented business knowledge

For nearly everyone, the honest answer is nowhere. It’s in:

Your head. Perfectly clear, entirely inaccessible.
A PDF. A capability deck from 2022 that four people have. PDFs are where knowledge goes to die — badly parsed, rarely indexed, never updated, and still the default way businesses write things down.
Your inbox. You’ve explained your process beautifully, in prose, four hundred times. Every one of those explanations is locked in a private thread.
Your homepage, in code. “Innovative solutions for the modern business.” That sentence has been published ten million times and it describes nothing. It’s not that a machine can’t read it — it reads it fine and correctly concludes it contains no information.
Directories from 2019. Old address, old phone, old services, still authoritative, still being counted.

That last one is where the damage is. Your knowledge isn’t just missing — it’s been replaced by a stale version that never got taken down.

Why contradiction is worse than silence

Multiple conflicting business listings outnumbering a single current source of truth

Answer engines resolve entities by corroboration. They read many sources and ask: what does everyone agree on?

If you’re silent, you’re absent. That’s bad.

If you’re contradictory, you’re unreliable. And that’s genuinely worse, because now the machine has to pick a winner between your current site, a 2019 directory, and an old profile — and it doesn’t pick the newest. It picks the most corroborated. Your one accurate homepage loses to five stale listings that all agree with each other.

This is why “we updated our website” doesn’t fix it. You updated one source. You didn’t touch the eleven that outvote it.

The gap between what you know about your business and what a machine can find out about your business is now a revenue gap. It just doesn’t appear in any report, because there’s no analytics line for “was described inaccurately in a conversation you’ll never see.”

Depth beats optimization

Here’s the thing this work taught me, and it came from an unlikely place.

Years ago I built a site about the history of rock and roll. Not for a client. Just something I cared about, and I went properly deep — the labels, the session players, the regional scenes, the arguments about who did what first. No keyword research. No optimization strategy. I wasn’t trying to rank; I was trying to get it right.

It out-performed nearly every client site I was being paid to optimize at the time. For years. And it wasn’t close.

I spent a long time slightly annoyed about that. The lesson I eventually took: search engines were always trying to find the most genuinely useful page, and I’d been spending my professional life trying to trick them into picking a thinner one. The hobby site won because it deserved to.

That was true then. It’s structurally true now. Answer engines are explicitly built to find the most authoritative, complete, corroborated source and quote it. There’s no thin-page strategy that survives that — you can’t optimize your way into being the best source. You have to actually be the best source, which means writing down what you know in more depth than anyone else has bothered to.

Which is inconvenient, because it’s real work. It’s also the most durable competitive position available, precisely because most people won’t do it.

So this service makes no ranking guarantees, and you should distrust anyone who does. What it does is make you the most complete, most consistent, most readable source of truth about what you do. That’s the thing the systems are looking for. Whether they pick you in any given week isn’t something I control, and anyone claiming otherwise is describing a lever that doesn’t exist.

The Source of Truth method

1. Extract. Get it out of your head. Structured interviews — what you do, for whom, how, what it costs, what you refuse, what happens when. This is most of the work and the part clients are most surprised by. You’re not learning anything new. You’re saying out loud, for the first time, things you’ve known for a decade.

2. Structure. Turn prose into architecture. Services, entities, relationships, definitions, boundaries. Machine-readable formats: proper HTML with real headings, schema markup, tables, question-and-answer pairs. Not a PDF. Never a PDF.

3. Publish. It has to be public and crawlable. Knowledge in a client portal is knowledge no machine can vote on. This is where most of it lands on your site as real pages — which is also, conveniently, exactly what AI search visibility needs.

4. Corroborate. Hunt down every stale source and fix or kill it. Directories, old profiles, that PDF still on your server, the guest post about work you don’t do. Unglamorous archaeology, and the step that actually changes what AI says about you — because you’re not adding a vote, you’re removing the ones voting against you.

5. Maintain. It goes stale the moment your business changes. A quarterly review, or you’re back here in two years.

Ship /llms.txt. A plain-language file at your root saying who you are, what you do, and where the canonical answers live. It’s the note you leave for the machine describing you when you’re not in the room, and it takes an afternoon.

What this unlocks

This page is unglamorous and it’s the foundation of three other things you probably want:

Your chatbot stops hallucinating. Bots don’t invent from nowhere — they invent when the knowledge base is thin. Every agent build I do starts here, because an agent is only as good as what it’s allowed to read.

AI search starts naming you. You can’t be cited if there’s nothing citable. Visibility is downstream of this.

Automation becomes possible. You can’t automate a process nobody has written down. Half of every automation project is discovering the process only exists as folklore.

People try to buy those three things directly. It’s like buying a roof before the walls. It’s also why so many chatbot projects quietly fail — the bot wasn’t the problem.

Frequently Asked Questions

Why does ChatGPT describe my business incorrectly? Because it’s reporting the most corroborated version of you that exists publicly, and that version is usually years out of date. Every wrong detail almost always has a real source — an old directory listing, a stale PDF, an outdated profile. The model isn’t inventing things, it’s aggregating what’s out there. If the accurate version only lives in your head, it can’t be counted.

What is llms.txt and do I actually need one? It’s a plain-text file at your website root telling AI systems who you are, what you do, and where your authoritative answers live — a summary written for machines rather than people. It’s cheap and takes an afternoon. It’s not a magic switch and adoption varies across AI systems, but it costs almost nothing and removes any excuse for being described badly.

Isn’t this just schema markup with a new name? Schema is one component, not the whole job. Schema tells a machine what type of thing a page describes. AI-readability is about whether the knowledge exists at all, whether it’s structured, whether it’s public, and whether every source agrees. You can have perfect schema on a page that says nothing useful — plenty of sites do, and they’re invisible for exactly that reason.

Can you guarantee this will get me ranked or cited? No, and I’d distrust anyone who says otherwise — there’s no submission form or ranking dial for AI systems. What this does is make you the most complete, consistent and readable source of truth about what you do, which is demonstrably what these systems are built to find and quote. That’s an honest edge rather than a guarantee, and in my experience genuinely deep material has always outlasted optimized thin material.

How long does this take and what do you need from me? The extraction phase needs real time from you — typically several hours of structured interview, because the knowledge genuinely is in your head and there’s no shortcut around getting it out. Structuring and publishing takes a few weeks depending on scope. The corroboration cleanup is the slow part, since it depends on how many stale sources are out there and how cooperative each one is about being updated.

Write it down

My client wasn’t defamed by a robot. She’d just never published a better version of herself than the one from 2021, so the machine used the one it could find.

Everything she needed was true, current, and locked in her head where nothing could read it.

Find out what AI currently says about you — $997 Diagnostic →

About Stevie

I spent years knowing exactly what my business did and never once writing it down properly. Then I lost my site, and with it every trace of what I’d published, and I discovered how little of my business had ever existed anywhere but in my own head.

I rebuilt from a blank page. It taught me that undocumented knowledge isn’t knowledge — it’s a rumor with good intentions.

sosSTEVIE — for when your digital presence is broken, invisible, or about to be. (754) 302-4631

Why Doesn't ChatGPT Recommend My Business?