I audited one local brand across four AI search engines. It exposed orphaned pages, inaccurate data, and a content library the business didn’t know was invisible. Here’s the method — and the spreadsheet!

I ran a local business through four AI search engines to see how they’d describe it.

The surprising part wasn’t what the AI got wrong. It was where the answers came from.

We spend a lot of energy debating whether AI search will replace Google. Meanwhile, ChatGPT, Gemini, and Perplexity are already describing our clients’ businesses to potential customers every day — pulling from sources most of us never audit. So I decided to actually look.

In traditional SEO we have comfortable metrics: rank position, the map pack, SERP features. None of them apply cleanly to AI. An AI answer doesn’t have a “position 3.” It has citations — sources it chose to trust and pull from.

So I needed a different instrument. I built a simple spreadsheet I’ve been calling the AI Visibility Tracker. The structure is deliberately basic: one row for every search term × engine (ChatGPT, Gemini, Perplexity, Google AI Overviews). For each row I logged four things:

  • Status — is the business Cited (named and linked as a source), Mentioned (named, but the credit goes elsewhere), or Absent?
  • Accuracy — is what the AI says factually correct, checked against the site and the Google Business Profile?
  • Sentiment — neutral, positive, or negative framing?
  • Source — and this is the one everyone skips — what is the AI actually pulling from to build this answer?

Then I ran the client’s five core service queries across all four engines and filled it in, screenshotting everything (AI answers change by the hour — treat every result as a dated snapshot).

Four lessons fell out of it.

Lesson 1: AI barely used their website

The most important column turned out to be Source. Across nearly every query, the engines built their answers from the Google Business Profile and third-party reviews — not the company’s own website. And each engine had a fingerprint:

  • Gemini leaned on the GBP — hours, categories, attributes — and was consistently accurate because that data is structured.
  • Perplexity leaned on reviews and GBP attributes, and even quoted a specific parent review.
  • ChatGPT was pulling Yelp reviews — which, for local businesses, are usually thinner and lower-quality than the GBP reviews most owners ignore.

If AI is describing your client using Yelp and a half-finished GBP, that’s not a content problem. That’s a data-source problem you can only see if you look.

Lesson 2: It exposed a content library the business didn’t know was invisible

Here’s the finding that reframed the whole audit. The practice had roughly two dozen in-depth service and specialty pages — dedicated content on anxiety, EMDR, play therapy, and more. Genuinely good, specific writing.

Every one of them was orphaned: no internal links pointing to them, not in the site navigation, no structured data. They existed only in the sitemap.

The result? AI could read the content — it accurately surfaced specific techniques straight off those buried pages — but it never cited the business as the source. Three separate times, an engine described a specialized service correctly, and three separate times the page it came from was invisible in the site’s own architecture.

That’s the opposite of the usual SEO story. The problem wasn’t missing content. It was excellent content, disconnected and unmarked — an asset the business already owned but couldn’t get credit for.

Lesson 3: Where structured data was missing, AI improvised

The accuracy column surfaced a clear pattern. Where the business had authoritative, structured data, AI was correct. Where it didn’t, AI filled the gap with plausible-but-unverified details:

  • One engine claimed the office had elevator access — not stated anywhere on the site.
  • One listed specific hours that contradicted the website (the site said “hours vary”; the engines pulled the real hours from the GBP).
  • One named a service the site didn’t clearly claim.

None of these were catastrophic. But for a business in a trust-sensitive industry, “AI is inventing details about you” is a real risk — and it traces directly back to the absence of a single, structured, authoritative source.

Lesson 4: It doubled as a local SEO and GBP audit

To verify what the AI was saying, I had to verify the GBP itself — the hours, the primary category, the attributes (accessibility, parking, LGBTQ+ friendly), the review profile. The AI audit forced a local-SEO audit. One exercise, three findings: GEO visibility, technical SEO issues, and GBP accuracy.

The bigger lesson: GEO isn’t a separate discipline

The headline for me wasn’t any single result. It’s that one simple tracker surfaced orphaned pages, content-gap opportunities, GBP inaccuracies, and citation sources — all at once.

GEO, technical SEO, and local SEO aren’t three separate practices. They’re the same fundamentals viewed through a new lens. And I’ll say this plainly, because the space is full of hype: this isn’t about stuffing schema and expecting AI to rank you. It’s about being the most discoverable, consistent, and authoritative source for your own information — across your site, your GBP, and the third-party platforms AI actually trusts. Schema is one supporting signal in that system, not a magic lever.

How to run this yourself

You don’t need my spreadsheet to start — you need the discipline. Here’s the method:

  1. Pick 5 money queries a real customer would type (service + city works well for local).
  2. Set your search location to the client’s city and run logged-out / incognito to reduce personalization.
  3. Run each query in ChatGPT, Gemini, Perplexity, and Google (watch for the AI Overview).
  4. For each, log Status, Accuracy, Sentiment, and Source — and screenshot it.
  5. Verify every specific claim against the website and the GBP. The claims AI gets wrong, and the sources it pulls from, are your findings.
  6. Note what’s Cited vs. Mentioned. “Present but not the source” is a fixable gap, not a win.

Do that for one brand and I promise you’ll find something you didn’t expect — an orphaned page, a wrong hour, a Yelp review outranking a better GBP one.

I’m sharing the AI Visibility Tracker template so you can run this on your own brand or a client’s. It’s a starting point — make it yours.

If this was useful, follow along; I’m documenting more of how I audit AI search for local businesses. Want the template? Here’s a downloadable link to the AI Visibility Tracker template!

What’s the strangest thing AI has said about a business you manage? I’d genuinely like to know.