A shop that ranks on Google can still be missing from ChatGPT. The model is not reading the local 3-pack and copying it. It looks for a business it can treat as a named entity, with matching listings, reviews that clear a cutoff, and pages it can parse. If those signals are thin or messy, the answer names Angi, a franchise, or nobody at all.
That is the usual case, not the exception. Omni Eclipse queried ChatGPT for 1,700 businesses across 32 industries between January and March 2026. Only 202 showed up. That is an 11.9% hit rate, which means 88% of the sample never made the answer. The US slice was worse: 20 of 283 businesses, or 7.1%. US roofing was 0 of 19.
What "not in AI answers" actually means
There are a few different failures, and they feel the same from the truck.
The first is total absence. Someone asks "best plumber in Hackensack" and the reply lists platforms, or gives generic advice about licenses and reviews, and never says a shop name.
The second is a wrong name. The model invents a company, merges two shops with similar names, or cites a closed location.
The third is a platform hijack. Metricus ran 200-plus homeowner-intent prompts in 2026 and reported Angi in more than 90% of answers. Independent contractors landed under 1%. HomeAdvisor, Thumbtack, and big-box installer programs filled most of the rest.
SOCi's 2026 Local Visibility Index, built on about 350,000 locations, found ChatGPT recommended 1.2% of businesses. The Google local pack still surfaced 35.9%. Showing up on the map is a different contest from getting named in a chatbot.
Why Google rank does not put a shop in the answer
Trades owners treat page one as the finish line. For AI answers, it is not.
Omni Eclipse looked at 356 businesses that already sat on Google page 1. Only 82 of them also appeared in ChatGPT. The other 274 had the ranking and still missed the recommendation. Google Ads did not help: 121 of 126 advertisers in that set were invisible in ChatGPT.
That matches what FirstCall has already covered in Do AI Chatbots Use Google Rankings?. Ahrefs measured about 12% overlap between URLs cited by assistants and Google's top 10 for the same prompt. A #1 Maps listing can coexist with a blank ChatGPT reply.
Yext's citation study (6.8 million citations across ChatGPT, Gemini, and Perplexity) found 86% of citations came from sources a brand already controls: websites and listings first. First-party sites produced 44% of citations. Listings produced 42%. Reddit-style forums were about 2% once location and intent were applied. For a plumber, that means the website, Google Business Profile, Bing Places, and consistent NAP data do more work than a forum thread ever will.
The five reasons a trades shop stays invisible
Most of the misses stack. One broken signal is enough. Three broken signals make the shop easy to skip.
The entity is fuzzy. Name, address, and phone do not match across the site, GBP, Yelp, and the license lookup. AI has to resolve "who is this business" before it can recommend it. Two DBAs, an old suite number, and a tracking phone on the site are enough to split the record.
The site is a brochure. "Quality service since 1998" and "licensed and insured" do not give a model a fact it can quote. Homeowners ask for panel upgrades, hydro-jetting, or a 3-ton changeout in a specific town. If those jobs are not on the site in plain language, with service area and a way to book, the model has nothing extractable. Schema helps here. LocalBusiness plus the right subtype (Plumber, HVACBusiness, Electrician, RoofingContractor) is machine-readable. A lot of trades sites still ship none of it.
Reviews sit under the cutoff. Scatterbranch's 2026 research found businesses near 3.4 stars with weak response rates dropped out of AI recommendations entirely. Volume matters too. A franchise with thousands of reviews across a brand wins "near me" prompts even when the local truck is average.
Directories out-publish the shop. Angi and Thumbtack mint millions of pages. The average independent site gets a few hundred visits a month and a handful of third-party mentions. Metricus put that gap in the 10,000x range. The model has seen the platform far more often than it has seen the shop.
The shop never gets checked. Recommendations move. FirstCall's grounding drift piece covers why a name can appear on Tuesday and vanish on Friday. A one-time ChatGPT test in the office Wi-Fi is not a measurement system.
How to test it this week
Open a private window. Use the city a homeowner would type, not the county the shop markets. Run the same prompt in ChatGPT, Gemini, and Perplexity:
"Who should I call for [trade] in [city], [state]?"
Write down every name, every platform, and every source the model cites. Then run two tighter prompts: one for an emergency job, one for a planned job (water heater, furnace, panel). If the shop is absent on all six, the problem is entity and content, not a fluke prompt.
Check the obvious files next. Does the homepage state the legal name, street address, phone, hours, and towns served in text, not only in an image? Does Google Business Profile match that text character for character? Is there LocalBusiness JSON-LD in the page source? Are service pages real URLs, or one "services" blob?
5WPR's 2026 HVAC and Plumbing AI Visibility Index found 87% of independents had zero citation share on "service near me" style prompts. Franchises still took 25-30% of those prompts while making up under 5% of contractors. The independent path is not "join Neighborly." It is clean listings, schema, trade-press or local-news mentions, and pages that answer the job.
What to fix before buying another ad
Paid search does not transfer. Omni Eclipse's advertiser split already showed that. Spend the next two weeks on signals the models actually read.
- One legal name, one address, one phone, copied into GBP, Bing Places, the website footer, and the schema block.
- LocalBusiness schema with the correct trade type, plus opening hours and geo coordinates.
- Service pages for the five jobs that pay the bills, written as answers: what the job is, who it is for, which towns, what happens on site.
- Review responses on GBP. Star rating above the 3.4 danger zone Scatterbranch flagged. Ask for reviews after completed jobs, not as a blast.
- A monthly AI check with the same prompts, logged in a spreadsheet. Track who got named, not vanity traffic.
FirstCall Visibility's free audit walks the 12-point version of that list. Run it if the private-window test came back empty. The point is not a new ranking report. The point is whether ChatGPT can name the shop when a homeowner in the service area asks.
FAQ
Why is my business not in AI answers?
ChatGPT and similar tools name businesses they can identify as a clear local entity with matching NAP data, reviews above their cutoff, and pages they can parse. Ranking on Google is not enough. Omni Eclipse's 2026 study of 1,700 businesses found 88% never appeared in ChatGPT recommendations. US roofing scored 0 of 19.
If a shop ranks in Google, why wouldn't ChatGPT mention it?
Google and ChatGPT score different signals. Omni Eclipse found that only 23% of Google page-1 businesses in its sample also showed up in ChatGPT. Ads do not close the gap: 121 of 126 advertisers in that dataset were still invisible in ChatGPT.
Who does AI name instead of local contractors?
Often Angi, HomeAdvisor, Thumbtack, big-box installer networks, or a national franchise. Metricus tested 200-plus homeowner prompts in 2026 and reported independent contractors named in under 1% of answers, while Angi showed up in more than 90%.
What should a plumber, HVAC shop, or electrician fix first?
Make the shop a clean entity: identical name, address, and phone on the website, Google Business Profile, and Bing Places. Add LocalBusiness schema with the right trade type. Keep star rating and review volume above the cutoff used in local AI tests. Publish answer-shaped service pages for the jobs people actually ask about.
How can a trades shop check if it is in AI answers?
Ask ChatGPT, Gemini, and Perplexity the same homeowner prompt in a private window, using the city and trade. Record who gets named and which sources get cited. Repeat monthly, because recommendations move. FirstCall Visibility also runs a free 12-point AI visibility audit.