AI Search, Explained for ContractorsTygart Media

Published September 29, 2026 · Last updated September 29, 2026 · Per-engine citation data is current as of August–September 2026 — this page is re-checked monthly.

Google Quotes My Website but ChatGPT Never Names Me — Do They Even Read the Same Websites?

Direct answer: Mostly not. BaaDigi tracked 1,429 answers to 88 contractor-marketing questions in August 2026 (write-up updated ~September 24, 2026) and found google.com cited 310 times inside Google AI Overviews — and zero times by ChatGPT. YouTube: 214 citations in AI Overviews, 4 in ChatGPT. One contractor-marketing site picked up 46 ChatGPT citations and zero in AI Overviews. And in a four-engine study (arXiv 2609.22655, cited by WildRun), 84.9% of engine pairs shared no cited URLs at all. You are not doing one job called "AI visibility." You are doing three or four separate jobs — and any tool that hands you one blended score is hiding exactly where you're named and where you're invisible.

The per-engine table that ends the argument

BaaDigi, a contractor-marketing agency, put the exact question to itself and printed the answer: "Does ChatGPT use the same sources as Google AI Overviews? Mostly not, in our sample." (BaaDigi, "What Gets a Contractor Named in ChatGPT?")

The numbers underneath that sentence are the whole story:

  • google.com — cited 310 times inside AI Overviews, 0 times by ChatGPT
  • YouTube — cited 214 times in AI Overviews, 4 times by ChatGPT
  • beseencontractors.com (a contractor-marketing site) — 46 ChatGPT citations, 0 in AI Overviews

That table is not a fluke of one weird sample. In an independent study across ChatGPT, Copilot, Google, and Perplexity (arXiv 2609.22655, cited by WildRun), 84.9% of engine pairs shared no cited URLs at all — the measured overlap was 0.0079. Read that number slowly. Being named in one engine is no evidence about the others. Zero evidence.

And the most brutal version of the finding: a marketing firm tested "best AC repair near me in Arlington TX" across four AI platforms, and not a single contractor appeared in more than one tool's recommendations (via piiiico/proof-of-commitment GitHub user-journey.md, cited in the September 29 sweep). Four tools, four completely different contractor lists.

What this feels like on the ground

Here is the owner version of the same finding. You ask Google about your own trade and your own city, and the AI Overview quotes your website by name. It feels like confirmation: the internet knows you, the machines know you, the work is done.

Then a homeowner tells you they asked ChatGPT who to call for a burst water heater, and your name never came up. Not quoted. Not linked. Not even on the list. Meanwhile the competitor across town — the one whose website looks like it was built during the Bush administration — is named twice.

Nothing is broken. No one made an error. You just discovered that the engine quoting you and the engine naming your competitor are not reading the same websites. The Google-flavored answer pulled from Google's shelf, where your site sits nicely. The ChatGPT answer pulled from a different shelf entirely — one built out of Reddit threads, directories, and trade sites where you don't exist yet. Same homeowner, same trade, same city. Different shelves, different winners.

Different shelves, same library — why the citations don't overlap

It helps to stop thinking of "the internet" as one thing the AIs all read. Think of it as a library, and each engine pulls books from different shelves.

Google's AI answers drink from Google's well. Of course google.com shows up 310 times — that's Google's own maps, profiles, and properties being quoted by Google's own answer engine. YouTube is Google's property too, which is exactly what you'd expect from 214 citations on one side and 4 on the other.

ChatGPT's shelf looks nothing like that. Zero google.com citations. Instead it leans on Reddit threads, directories, and niche trade sites — a contractor-marketing site most Google AI Overviews would never touch earned 46 citations from ChatGPT alone.

This is why the owner experience feels so strange. You Google your trade plus your city, the AI Overview quotes your website, and you think "we're doing fine." Then a customer says they asked ChatGPT who to call for a slab leak and your name never came up — and ChatGPT is quoting a Reddit thread and a contractor directory you've never heard of. Both engines answered honestly. They just read different shelves. (For the full engine-by-engine breakdown of what each one actually looks for, see how the engines choose contractors.)

What dies here: the single "AI visibility" score

This is the question that kills the blended number. BaaDigi's own warning line, verbatim: "a blended 'AI visibility' number will hide that. Measure ChatGPT on its own."

Read that as a buying rule. If a vendor report tells you your "AI visibility score" is 72, that number is averaging your Google visibility with your ChatGPT invisibility — and the fix each engine needs is a different job. Google's AI quoting your site tells you nothing about why ChatGPT is naming the competitor, because ChatGPT was never reading your site in the first place.

So the first question to ask any tool or agency selling you AI visibility is the simplest one: show me the engines separately. If they can't — or won't — the number they're selling you is the blended fiction this page disproves.

What to actually do this week

The fix is a per-engine scorecard, and it's cheap to build yourself.

  1. Write down five buyer questions. The kind a homeowner actually types: "who fixes a slab leak in Fort Worth," "emergency water heater replacement Tacoma," "fire damage restoration company near me." Keep them plain and buyer-voiced.
  2. Run each one in four places: ChatGPT, Google's AI Mode or AI Overviews, Copilot, and Perplexity. Use a fresh session each time — logged-out or temporary chats — so your own history doesn't contaminate the answer.
  3. Log one row per engine. For each engine, write down: which businesses were named, and which domains got cited. Four engines, four rows. Never one blended number.
  4. Read the asymmetries. That's where the work list comes from. If Google's AI quotes your website but ChatGPT cites Angi, BBB, and a Reddit thread instead, your ChatGPT fix is not "improve my website" — it's directories, review volume, and getting talked about on ChatGPT's shelf. The divergence is the diagnosis.
  5. Repeat monthly. These shelves shift — retrieval changes happen without announcement, and what an engine reads this month may not be what it reads next month. The per-engine table is a repeatable format, not a one-time audit. (Recommendations don't even stick week to week inside a single engine — see why a recommendation can vanish in days.)

The line to remember

Google's AI quotes your website. ChatGPT names whoever gets talked about on ChatGPT's shelf of the internet. They don't read the same websites — so don't take one scorecard to four different games. Measure each engine on its own, and fix what each engine actually reads.

FAQ

Q: If Google's AI quotes my site, will ChatGPT eventually pick it up?
A: There is no evidence of that — the BaaDigi sample shows the traffic running the other way. A site can be heavily cited in one engine and at zero in the other, in both directions. Treat each engine as starting from scratch until your own monthly test proves otherwise.

Q: So do I need to do four times the work?
A: No. The work is one pile — real reviews with specifics, consistent directory listings, answer-shaped pages on your own site. The measurement is four separate scorecards. Build the assets once; check each engine's shelf separately to find out which ones are picking them up and which ones aren't.

Q: Should I pay for an AI visibility tracking tool?
A: Only if it reports per engine. A single blended score is exactly the fiction the BaaDigi data disproves — ask the vendor to show you ChatGPT and Google AI Overviews as separate lines before you pay for anything. If they can't, the tool is selling you the average of your visibility and your invisibility. (See how to check your own AI search visibility for the free version of the test.)

Q: Which engine do I fix first?
A: The one your buyers actually use — and you won't know which one is costing you until you run the four-row test. Practical order: start with the engine where you're most invisible, because that's where the gap between you and the named competitors is pure upside. A contractor who's quoted by Google's AI but unnamed by ChatGPT has a bigger ChatGPT problem than a Google problem — the Google side is already working. Fix the empty shelf first.

Q: Is this divergence permanent, or do the engines converge over time?
A: The shelves shift — retrieval changes land without announcement, and what an engine reads this month can differ next month. That's an argument for the monthly per-engine check, not against it. Convergence would be good news; until your own tests show it, run each engine separately.