The Company AI Remembers Isn't the Company You Are

A founder recently asked ChatGPT the same question five times in a row: “What is my company?” He got two different answers. One described his business today. One described the business he ran three years ago.

That’s not a glitch. It’s how these systems work.

An AI model learns your identity from more than your homepage. Wikipedia votes. Software directories vote. Review sites vote. Your own help center votes, even the page nobody has opened since 2023.

Change your positioning, and your homepage updates in an afternoon. The rest of the web doesn’t get the memo. Old category pages keep describing the old you. Abandoned support articles keep answering “how are you different” with an answer from two products ago.

Call it the inherited answer. It’s what the web learned about you before you changed, and it doesn’t disappear just because you rewrote your hero copy.

One company traced this problem after noticing its API documentation still called it something it hadn’t been in over a year. A help-center post, untouched since launch, was still telling readers the old story. Worse: an AI model had cited that exact page as a source.

The fix wasn’t complicated. It was tedious. Find every page still voting for the old identity. Fix what you control first: your site, your docs, your directory listings. Then go after the sources you don’t control, one by one, with evidence a model can retrieve and trust.

Branded questions clear up fastest, because the answers lean on pages you own. Category questions, the ones where someone asks what kind of company does what you do, take longer. Those answers lean on the wider web’s opinion of you, built over years, by people who stopped paying attention the moment they hit publish.

Here’s the uncomfortable part. You can’t run this check once and call it done. Models retrain. Pages get cited, then forgotten, then cited again. The old version of you can resurface in an answer six months after you thought you’d buried it.

Most teams don’t have the bandwidth to hunt down every stale directory listing and outdated review profile by hand. That’s the actual work behind AI visibility, and it’s mostly invisible until a customer asks the wrong question and gets the wrong answer.