What We Measured When We Asked AI About Us

The headline number is not good for us. We’re publishing it anyway, because a contractor deciding where to spend a marketing budget this quarter deserves the actual data, not a confident-sounding guess.

What we did

On one day (September 22, 2026), we ran a fixed bank of 30 non-branded, high-intent prompts — the kind a homeowner or a business owner actually types, never our name or a competitor’s — across four categories of home-services and local-marketing queries. We ran all 30 through standard web search, then ran a fixed 12-prompt subset through ChatGPT search, Google AI Mode, Google’s SERP (AI Overview, the local pack, and organic results), and Bing’s SERP (AI summary, Places cards, and organic). We got 2 of the 30 prompts into Perplexity before hitting its logged-out wall. Every session was logged out, in a fresh automation browser, US locale requested. No accounts, no CAPTCHAs, nothing bypassed. That’s six surfaces: web search, ChatGPT search, Google AI Mode, Google’s AI Overview/local pack, Bing’s AI summary, and Perplexity.

For every observation we recorded whether we were mentioned, whether we were cited with a linkable source, where in the answer, and — critically — the domain and page type of every single source the engine cited, whether or not it involved us. That produced 92 total observations and a full citation map, not just a yes/no scorecard.

What we saw about ourselves

Zero. Across all 92 observations, on every surface, we did not appear once. Web search: 0 of 30. ChatGPT search: 0 of 12 mentioned, 0 of 12 cited. Google AI Mode: 0 of 12. Google’s AI Overview showed up on only 1 of the 12 prompts it was checked against, and we weren’t in it. The local pack appeared on 8 of the 12 prompts checked; we weren’t in any of them. Bing’s AI summary appeared on 3 of 12; we weren’t in any. Perplexity: 0 of 2 before the wall. Zero of 92.

That’s not a metaphor for “needs improvement.” It’s a literal count, on one specific day, for one specific prompt set. We’re a real, operating firm with a real website, and the answer engines in our own market didn’t have a reason to mention us. Whatever else this data says, it says that plainly.

What we saw about the market

The more useful findings are about who does get cited, and how.

Agency-owned pages dominated ChatGPT’s citations. Across the 12 ChatGPT search observations, 27 unique domains were cited, and 36 of the 43 total citations were pages the marketing firms themselves had published — homepages and city/service landing pages that match the query almost word for word. This is a single-cycle, directional finding, and it’s a correlation, not proof that publishing such a page is what caused the citation. But the pattern is consistent enough, and cheap enough to test, that it belongs on any contractor’s list regardless of the mechanism.

One directory showed up everywhere. Across the five source-tallied surfaces we checked (ChatGPT, Google, Bing, Perplexity, and the web-search results pool), a single third-party directory — Semrush’s Agency Partners listing — appeared among the cited or ranked sources on all five, 18 times total in this one-day run. No other source came close to that spread. It’s free to join.

Google Business Profile chips led Google AI Mode’s citations. In the 12 AI Mode observations, Business Profile chips were the single most frequent cited source type — more than any directory, listicle, or review site. AI Mode’s underlying retrieval likely draws on more than Business Profile data alone, and we can’t isolate the mechanism from a one-day sample. But the correlation is hard to miss: it lines up with the local pack, which is directly built on Google’s local-business ecosystem, and which is where the review-count gap was starkest (below).

Source-type mix varied by engine, and none of it was exotic. Google AI Mode’s 12 observations leaned on roughly 14 agency-owned pages, 4 directory listings (Semrush appeared twice, plus one each from two other directories), 2 listicle-style pages, and 2 review-platform mentions, plus the Business Profile chips above. Bing’s AI summary, on the 3 of 12 prompts where one appeared, leaned on roughly 8 agency-owned pages plus repeated hits from three B2B directories and Bing Places cards. In no case did an engine cite anything unusual — directories, agency sites, and Google/Bing’s own local data, in different ratios per surface.

The local pack ran on reviews, and we don’t have any. Across the 8 of 12 prompts that returned a Google local pack, the businesses shown carried Google review counts ranging from 19 to 254. Our own Business Profile has 1 review. We’re not claiming review count is the only local-pack input — we didn’t test that in isolation — but it’s the most visible gap between us and everyone who showed up.

What this does NOT prove

We want to be as clear about the edges of this data as we are about the findings, so here’s the limitations list in full, unedited:

• This is a single-day, single-cycle sample with no replication yet. AI answers are stochastic — the same prompt can return a different answer on a different run — and we captured one observation per prompt per surface.
• The ChatGPT sample was 12 of our 30 prompts; the Perplexity sample was 2 prompts before we hit a logged-out wall; Copilot wasn’t observable at all behind its sign-in requirement.
• Google’s AI Overview appeared on only 1 of 12 prompts checked, and Google AI Mode’s citation links weren’t extractable in our capture method — we recorded domains and titles, not clickable source URLs.
• Every session ran logged out in an automation browser with a requested US locale; we did not control for IP-based location or any personalization a logged-in, geolocated user would see.
• Our source universe was bounded: 32 external sources checked directly, no backlink index, no rank-tracking tool — web-search operators and direct fetches only. Nothing here is exhaustive, so we don’t use the word “zero” for anything we didn’t check completely, and we don’t use it here either — “zero” appears above only for our own literal appearance count, which we did check completely across all 92 observations.
• None of this is causal. Where we say a page type, a directory, or a review count showed up alongside citations, that’s an observed correlation from one cycle, not a tested mechanism. We do not claim any specific page “runs on” Google Business Profile data or that any single tactic “causes” a citation.

We’ll re-run the same 30-prompt bank on the same surfaces on a regular cadence and will only claim movement after at least two consecutive runs agree.

What a Central Florida contractor should do this quarter

None of this requires waiting on us, and none of it requires guessing.

1. Claim and fully complete a profile on the sources this run actually cited — Semrush Agency Partners (the one directory that showed up on every surface we checked), plus Clutch, GoodFirms, DesignRush, and Yelp, all of which have free tiers. If you’re in Bing’s AI summary audience at all, a completed Bing Places listing matters too — Bing Places cards were one of the cited source types there.
2. Get real reviews on your Google Business Profile. In this run, every business that showed up in a local pack carried somewhere between 19 and 254 Google reviews. You don’t need 254. You need more than the handful most contractors have, from actual customers, this quarter.
3. Build pages that answer the exact question a buyer types, not a general “our services” page. The most common citation pattern we saw was a marketing firm’s own city-and-service page matching the query almost verbatim. We can’t prove that page type causes the citation — but it’s the cheapest, most legitimate thing on this list to test, and it’s good for human readers regardless of what any AI engine does with it.
4. Publish a phone number and name a real person as founder or owner on your site. These are basic, verifiable facts that AI answers tend to paraphrase when they describe a business. If your site doesn’t have both, in plain text, that’s a same-week fix.

We’re publishing our own zero alongside this list because a marketing firm that only shows you winners isn’t showing you the method. Test the same four things on your own business, and you’ll have a better basis for the next marketing dollar than either of us had a month ago.