Blog › Local research

Seven in ten top-ranked local businesses are invisible to AI

We checked 81 Southwest Florida businesses across eight trades and five towns. Every one of them ranks on the first page of Google. Most of them do not exist as far as ChatGPT is concerned.

By Michael Putnam · 28 August 2026 · Venice, Florida

Ask ChatGPT to recommend a roofer in Venice, a plumber in North Port or a painter in Englewood, and it will give you a confident answer with three to five names. We wanted to know whether those names match the businesses that actually rank highest locally.

They mostly do not.

70%
of top-ranked local businesses were not named when ChatGPT was asked to recommend someone in their own trade and town

How we did it

We ran eight searches in Google Maps, in a logged-out browser so the results were not shaped by anyone's search history. One search per trade and town. For each we recorded the first ten results that were not paid advertisements.

Then we asked ChatGPT, also logged out, one question per trade: who's a good roofer in Venice FL, who's a good plumber in North Port FL, and so on. Asked once, not repeatedly.

A business counted as named only if ChatGPT put it in the written recommendation — the part a customer would actually read and act on. ChatGPT also displays a list of businesses above its answer, but appearing in that list is not the same as being recommended, and the difference turned out to matter.

81 businesses. Eight trades. Five towns: Venice, Nokomis, North Port, Englewood and Port Charlotte.

The results, trade by trade

TradeTownCheckedNamed by AINot named
PaintingEnglewood10460%
RoofingVenice10370%
PlumbingNorth Port10370%
HandymanNokomis10370%
Pool servicePort Charlotte10370%
LandscapingVenice10370%
House cleaningNorth Port11373%
HVACVenice10280%

Six of the eight trades landed within a few points of each other. That consistency is the most convincing part of the finding: it is not one odd search, it is the same result eight times across five different towns.

The results by town

The same data counted a different way. Two of the five towns had enough businesses across enough trades to stand on their own. The other three were a single trade each, so they describe that trade rather than the town, and are labelled as such.

TownBusinessesNamed by AINot named
Venice
roofing, HVAC, landscaping
30873%
North Port
house cleaning, plumbing
21671%
Nokomis
handyman only
10370%
Port Charlotte
pool service only
10370%
Englewood
painting only
10460%

Nokomis, Port Charlotte and Englewood each rest on one trade and ten businesses. Those figures describe handymen, pool companies and painters in those towns. They are not town-wide numbers and should not be read as any.

What stands out is the narrowness of the range. Five towns, eight trades, and every figure lands between 60 and 73 percent. This is not one bad town or one unlucky trade.

Three things we did not expect

1. Review count did not protect anyone

The single highest-ranked HVAC company in Venice has 8,936 Google reviews. ChatGPT did not mention it. Not in the recommendation, not in the list, not anywhere in the answer.

The top plumber in North Port has 891 reviews. Also absent. A pool company with 354 reviews, absent. A painting company with 149 reviews, absent.

Whatever decides which businesses get recommended, sheer volume of Google reviews is not sufficient on its own.

2. Having a website made no difference

Eleven businesses in the sample had no website at all. Four of them were recommended by ChatGPT anyway — including three house cleaners in North Port, a trade where every recommended business happened to have no website while several that did have one were passed over.

Look at what ChatGPT cited when explaining its picks: HomeAdvisor, the Better Business Bureau, Angi, Birdeye, Expertise. Not one business's own website was cited as a source, in any of the eight answers.

This is the part most local business owners have backwards. The answer is not assembled from your website. It is assembled from what other sites say about you.

3. It recommends businesses Google does not rank

Across the eight trades, twenty businesses were recommended by ChatGPT that were not in Google's top ten at all. Some were not in the Maps results we saw at any position.

So this is not simply a case of AI being worse at the same job. It is answering a different question, using different sources, and arriving at a different list.

The ratings do not match either

While collecting the data we noticed something we were not looking for. The star ratings ChatGPT displays frequently disagree with the ratings on the businesses' own Google listings.

Business type and townGoogleChatGPT
Paving and landscape, Venice4.93.0
Air conditioning, Venice4.82.5
Air conditioning, Venice4.93.7
Roofing, Venice4.82.9
Pool service, Port Charlotte4.94.6

One Venice HVAC company appeared with three different ratings inside a single ChatGPT answer: 4.2 in the list, 5.0 in the written text, and 4.9 on its own Google listing.

We are not able to say where the lower figures come from. But a business rated 4.8 by its customers being shown to a potential customer as 2.5 is a real problem for that business, and it is not one they can see or fix.

It also gets the trade wrong

Asked for a handyman in Nokomis, ChatGPT's list included a plumbing company and an appliance repair business. Asked for a landscaper in Venice, it returned three paving contractors and a tree service.

Small errors individually. But they mean a genuine handyman is competing for space with businesses that do not do the job at all.

What we are not claiming

This is 81 businesses in one county, checked in August 2026. It is a snapshot, not a law of nature.

The sample is deliberately top-heavy: every business in it already ranks on page one of Google Maps. That is not a random sample of local businesses, and it was not meant to be. If anything it makes the finding stronger — these are the businesses doing best, and seven in ten still went unnamed.

AI answers also vary. Ask the same question tomorrow and the names will shift. We asked each question once and recorded what came back, which is what a real customer would get.

And we cannot see inside these systems. We can report what they answered, not why.

What this means if you run one of these businesses

The uncomfortable version: your Google ranking, your review count and your website are not doing the job you assume they are doing. Businesses that were beaten on all three still got recommended.

The useful version: what ChatGPT cited was third-party sources. Directories, review platforms, local listings. Pages the business does not own and mostly did not create.

That is the lever. Not another page on your own site.

Want to know how your business shows up?
Free 30-second check on Google and AI search. No cost, no signup.
Run your free check →

Method in full

Sample. Eight Google Maps searches run 27 August 2026 in a logged-out browser: roofer Venice FL, house cleaner North Port FL, HVAC Venice FL, plumber North Port FL, painter Englewood FL, handyman Nokomis FL, pool service Port Charlotte FL, landscaper Venice FL. The first ten non-sponsored results were recorded in the order Maps listed them. Paid results were excluded. Nothing was skipped for looking good or bad. No business listed in the PresenceScanner directory was included.

The AI question. ChatGPT was asked once per trade in a logged-out browser on 27 August 2026, in the form "Who's a good [trade] in [town] FL?". A business counted as named only if it appeared in the written recommendation, not merely in the list of businesses displayed above the answer.

Limits. One county, one month, one AI system, 81 businesses. Ratings and rankings change. No business is named in this article as an example of failure, and none of the underlying records are published.

Anyone is welcome to repeat this. The searches are listed above and the whole thing takes about two hours.