We asked four AIs what our own company was. They all got it wrong.

If an AI assistant is asked what your business is and describes somebody else, every other visibility effort is pushing traffic toward a wrong answer. It's the most under-checked problem in this category, because almost nobody thinks to look — so we looked at ourselves first.

What happened when we tested our own name

We asked four assistants what ThapaWeb was. Every one of them described a large Indian YouTube channel with a similar name. Not one of them described a web design studio in West Covina, California.

That's not a hypothetical for a blog post. It was our starting score, and it's the reason this program exists. If we hadn't run the test we would have carried on optimising pages for an audience that was being handed somebody else's business.

Why does this happen?

Because machines don't infer identity, they read it. When a name is shared, the entity with more established, more corroborated information wins by default — and that's usually the bigger one. Nothing malicious is happening. You simply never wrote down, in a form a machine reads, what you are and where.

What actually fixes it?

Not repetition. Distinguishing detail, stated explicitly and corroborated elsewhere:

• The qualifiers. Your city, your category, your founder, your address, your hours, the year you started. These look like boilerplate and they're doing all the work — they're what separates you from the entity absorbing your name.
• One sentence, stated the same way everywhere. A plain description of what you are that appears word-for-word on your homepage, your About page, and in your machine-readable record.
• Linked profiles. Every profile you own pointing back to your site and forward from it, so an assistant sees a connected set of corroborating sources rather than one unsupported claim.

Then you ask again, monthly, and log the answer word for word. That record is the only way to tell whether anything changed.

How long does it take?

Longer than anyone wants. And against a genuinely dominant collision — if you share a name with a Fortune 500 company — the honest answer is that you'll be disambiguated by qualifier long before you're disambiguated by name alone. "The web studio in West Covina" lands well before "ThapaWeb" does.

That's worth knowing up front, because it changes what success looks like. Success isn't owning the name. It's being the correct answer when someone adds one word of context.

Common questions

Can't you just tell the AI companies they're wrong?

There's no correction form. You change what's written down for them to read, and you measure whether the answer changes.

Is this a problem if my name is unusual?

Less of one, and worth checking anyway — the failure mode for an unusual name isn't confusion, it's an assistant confidently inventing a description because it found almost nothing.

How would I check this myself?

Ask ChatGPT, Perplexity, Gemini and Claude what your business is, and write the answers down exactly. It costs you ten minutes. For a lot of owners it's the most alarming ten minutes of the quarter, and it's included in our $149 Diagnostic if you'd rather it came with the fix.