For most of this year, if you asked ChatGPT who Chris DuBois was, it told you about a music publisher in Nashville. Different person, same name, and a perfectly confident answer with no hedging in it. Everything I published, every podcast episode I recorded, every page I wrote about agency positioning was landing next to an identity that wasn’t mine.
I found out because I run the same five prompts against my own brand every month and write down what comes back. Most agency owners find out a different way, which is a prospect repeating something odd on a call, or nothing at all.
When an AI assistant describes your agency wrong, it’s usually because the model can’t tell which entity you are, and the fix is to give it a stronger public record to attach you to rather than to correct the answer directly.
Quick Take
- You can’t edit the answer. There’s no dashboard, no support ticket, and no correction form. The only lever is what exists publicly about you.
- Confusion is more common than being invisible, and it’s worse, because the model sounds equally certain either way.
- My own case resolved without the technical fix I’d planned. The model started sourcing me from a podcast directory page instead, which says something uncomfortable about what these systems actually weight.
- Check monthly with a fixed set of prompts, always in an incognito window. Logged in, the model knows you and will flatter you back. An identity problem you find in month one is a content problem, and one you find in month nine has been costing you deals the whole time.
In this article
- Why AI describes companies wrong
- How to check what AI says about you
- What fixed mine
- What people try that doesn’t work
- How often to re-check
Why does AI describe my agency wrong?
Because the model is guessing which entity you are, and it doesn’t tell you it’s guessing. Your name is attached to a thin public record, someone else’s name is attached to a thicker one, and the answer gets built from whichever source the retrieval step reached first. Nothing in the output signals the uncertainty.
This lands in two ways, and the second is the more expensive one. Being absent from an answer is a visibility problem you’ll eventually notice. Being present as someone else is a credibility problem you won’t, because the answer looks complete and plausible to the person reading it. A buyer asks about your agency, gets a confident paragraph describing a different company’s specialty, and decides you’re not the fit without ever saying so. You never hear about that call.
An agency owner I work with ran into a version of this from the other direction. His product’s brand name overlapped with a well-known television property, and searching it returned that show rather than his company. His words when it turned around were “It was like, dang, we really are buried.” A month later the same search returned his site first, then his own AI overview, then the app his team built, then their Reddit posts. The name hadn’t changed. What changed was how much public material existed with that name attached to his actual thing.
How do I check what AI says about my agency?
Write five prompts, run them once a month, and record the answers word for word. That’s the method, and the discipline is in keeping the prompts identical rather than in making them clever.
The five I use cover the ground where a buyer would meet me. One asks what my company is and who runs it, which is the identity check. Two ask for recommendations in my category, phrased the way a buyer would phrase them rather than the way I’d describe myself. One asks about the specific problem I solve. One asks for alternatives to a competitor.
Record who gets named, in what order, and which pages the answer cites. The citations matter more than the mentions, because they tell you which specific URL the model considered worth quoting, and that’s the page to build on.
Two things make this useless if you skip them, and the second one is where most people go wrong.
Change the prompt wording between runs and you’ve got no comparable data, just two unrelated snapshots.
Then run every one of these in an incognito window, logged out of everything. This matters more than it sounds. An assistant you have been talking to for months has built up a picture of who you are and what you are working on, and it is built to give you the answer it thinks you want. So when you ask it who the best providers in your category are, it has every reason to hand your own company back to you near the top. That is not the market talking, it is your chat history talking, and it will tell you the problem is solved when it isn’t. Incognito strips all of that out and shows you what a stranger sees, which is the only version worth writing down.
What fixed mine
Not the thing I’d planned. I had a technical fix written up and scheduled, adding structured data to my site that would state plainly who I am, what I run, and how I connect to my company. It sat in the plan for a month and never shipped.
The confusion resolved anyway. When I re-ran the identity prompt in August, the answer described my company accurately from my own site and named me correctly as founder, and the source it credited was a podcast episode page on a podcast directory. Not my site’s structured data, which didn’t exist. A third-party page where my name, my company, and my subject matter all appeared together in one place a crawler could reach.
I want to be careful about how much weight I put on one result. It’s one reading, and I’m holding it open another month before I call it stable. But the direction matches what the broader research keeps finding, which is that off-site brand footprint correlates with AI visibility far more strongly than the classic on-site authority signals do. The signal that fixed me was somebody else’s page describing me correctly.
That reframes the work. Being described accurately is less about tuning your website and more about how many credible places on the internet connect your name to your category.
What people try that doesn’t work
Adding schema markup and expecting it to change what gets cited. Structured data earns its place for classic rich results and it helps a crawler understand your entity, so it’s worth having. But no major model provider has confirmed that they read JSON-LD when deciding what to cite, and Google’s own guidance for generative AI says no special markup is needed. Ship it because it’s good hygiene, not because it’s the lever.
Writing a correction page addressed to the model. Nothing reads it as a correction, and you’ve spent a page of your site talking about a problem your buyers don’t have.
Publishing more of the same content faster. If the identity is ambiguous, volume makes both entities more prominent rather than separating them.
And optimizing tactics in general deserve more skepticism than they get. C-SEO Bench tested fifteen recommended conversational-SEO methods and found most of them ineffective and frequently negative, with the thing that did work being genuine relevance. That’s a hard finding for anyone hoping there’s a switch.
How often should I re-check?
Monthly is enough, and less than quarterly is not. These answers move faster than search rankings do. In one thirty-day window I watched a single competitor’s page displace four of the five brands named in an answer about my category, which is a pace that makes an annual audit meaningless.
Monthly also gives you the thing that’s hard to get otherwise, which is a record. When an answer changes, you want to know what it said before, because that’s the only way to tell an improvement from normal drift.
If you’re an agency owner who has never once checked what an assistant says about your company, that’s the fifteen minutes I’d spend this week. And if it comes back wrong, that’s an ordinary problem with an ordinary fix, not a crisis. We work on this kind of thing inside the Dynamic Agency Community.
FAQ
Can I contact OpenAI or Google to correct information about my company?
Not in any practical sense. There’s no correction form and no support path that reaches the answer you’re seeing. The only real lever is changing what exists publicly, so the model has better material to build from next time it retrieves.
How long does it take for AI to update what it says about my business?
Faster than most people expect. Crawl frequency has increased sharply and fresh pages get picked up within days rather than months, so a genuine change in your public footprint can show up in a matter of weeks. My own identity problem resolved somewhere inside a thirty-day window.
Does having a Wikipedia page fix this?
It would help, and most agencies won’t qualify for one, so it isn’t a plan. What’s available to everyone is the same underlying thing at smaller scale: third-party pages where your name, your company and your category appear together. Podcast episode pages, conference speaker bios, directory listings and interviews all do this work. Mine got fixed by a podcast page.
What if a competitor is being recommended instead of me?
That’s a different problem from being described wrong, and it’s usually about not existing in the answer at all rather than being misidentified. Start by checking whether you’re even in the retrieved set for that question, then look at which page the cited competitor won on. In my category, most cited pages turned out to be published by vendors inside it rather than by independent reviewers.