ChatGPT Recommends Just 1.2% of Locations: What the New AI Visibility Data Means

Muhammad Shahid·AI-Powered Digital Marketing Consultant

Google Ads Certified

Meta Blueprint Certified

Google Analytics Certified

Updated August 2026

AI assistants recommend a small fraction of the businesses Google ranks. In new research covering around 350,000 locations across 2,751 multi-location brands, ChatGPT recommended just 1.2 percent of those locations, Perplexity 7.4 percent, and Gemini 11 percent, while the same brands appeared in Google's local 3-pack 35.9 percent of the time. The researchers put AI visibility at three to 30 times harder to achieve than ranking in traditional local search. If your phone has gone quieter while your rankings held, this is the mechanism worth understanding.

The numbers come from SOCi's Local Visibility Index, covered by Search Engine Land. It is enterprise data from multi-location brands, not corner shops, which is worth holding in mind. But the direction it shows is the same one I see in the accounts I run for single-location service businesses, and the practical lesson transfers cleanly.

What did the research actually measure?

It measured how often a business gets named by an AI assistant compared to how often it appears in Google's local results, across the same set of locations. The study analysed roughly 350,000 locations across 2,751 brands against about 120 performance metrics, then asked the assistants the kind of questions a customer would ask and recorded which businesses came back.

Read the Gemini row twice. It recommends nine times more locations than ChatGPT and states their details with full accuracy, because it is grounded in Google Maps. That is the clearest signal in the whole dataset: the assistant wired to a maintained business database is the one that gets businesses right.

Why the gap is so wide

A results page can afford to be generous. It lists three map results, then ten links, then more on page two, and lets the searcher sort it out. An assistant answers in a sentence or two, which means it names two or three businesses and drops everyone else. As SOCi's CMO framed it, AI “isn't replacing search” so much as “reducing choice”. The shortlist got shorter, and nothing about your ranking guarantees you a place on it.

Why do businesses that rank well still get skipped?

Because the two systems read different inputs. Google's local results run on relevance, distance, and prominence, the three factors Google states publicly. An assistant is doing something else: assembling an answer from Google Maps, Yelp, Facebook, review sites, and your own website at once. Disagreement between those sources is a reason to leave you out. The research found only about 45 percent overlap in retail between brands visible in traditional search and those visible in AI results, so more than half of what wins one does not win the other.

In the accounts I take over, the pattern is almost always the same three things. Hours that differ between the Google profile and the website footer. A Facebook page with an old address nobody has touched since a move. A service the business genuinely offers that appears nowhere except a photo caption. None of those stop a Google ranking. All of them give an assistant a reason to pick the competitor whose story is consistent.

How does an assistant decide which businesses to name?

It reads several sources at once and looks for agreement between them, which is a meaningfully different process from ranking a page. The research found AI engines synthesising signals from Google Maps, Yelp, Facebook, and brand websites rather than leaning on one platform. That gives you a practical checklist of what to fix, because each source is a place your business can contradict itself:

• Google Maps and the Business Profile. The strongest single input, and the one Gemini reads directly. This is why its accuracy came back at 100 percent.
• Review platforms. Volume, recency, average rating, and the language in the reviews themselves, which is what an assistant paraphrases when it says a business is “known for” something.
• Social profiles. Usually the most neglected and most often wrong, especially after a move or a phone change.
• Your own website. The only source you fully control, and the one that should state your services and areas in plain, quotable sentences.

The word the report keeps returning to is selectivity, not ranking. Here is how SOCi's CMO put the shift:

“AI isn't replacing search, it's reducing choice.”SOCi, via Localogy

That is the whole story in six words. Nobody removed your listing. The number of names the customer sees simply dropped from ten to three, and the criteria for making that shortlist are stricter and less visible than the ones you have spent years optimising for.

The accuracy problem nobody is watching

The recommendation rate got the headlines, but the accuracy figure is the one I would act on first. Business profile information was about 68 percent accurate on ChatGPT and Perplexity. Roughly one detail in three is wrong, and the categories that go wrong are the ones that cost money:

• Opening hours that send someone to a closed door on a Sunday.
• Phone numbers from a previous provider or an old tracking line.
• Service areas that exclude suburbs you actively cover.
• Services attributed to you that you stopped offering years ago.

You will not see any of this in an analytics dashboard. There is no report for “customers who asked an assistant and got told the wrong closing time”. The only way to find it is to go and ask. I now run a five-minute check monthly for clients: ask ChatGPT, Gemini, and Perplexity the three questions a real customer would ask, then read what comes back about the business, including the details nobody thought to verify.

Does this mean local SEO is finished?

No, and the data argues the opposite. Gemini is grounded in Google Maps and reported full accuracy, which means a complete, correct Google Business Profile is doing double duty: it wins the map pack and it feeds the assistant with the largest recommendation rate of the three. Ratings mattered too. Locations ChatGPT recommended averaged around 4.3 stars, and brands with weak review data often failed to appear at all. The work has not changed. What changed is that doing it badly now costs you twice.

Why service trades may fare better than the headline suggests

The 45 percent overlap figure was measured in retail, and retail is the hardest case. A shopper asking for a clothing store has thousands of interchangeable options, so an assistant filters aggressively. Someone asking for an emergency electrician who works Sundays in one suburb has a candidate pool of maybe eight businesses, and half of them will not publish their hours clearly. The filter is the same, but the competitive set is far smaller, which is why I read this data as an opportunity for trades rather than a threat. A 1.2 percent recommendation rate across 350,000 mostly interchangeable locations does not mean your odds in your suburb are 1.2 percent. It means the assistant is looking for a reason to pick someone, and most businesses are not giving it one.

What should a local owner do this month?

Nothing exotic, and nothing that needs new software. This is the order I would work in:

1. Ask the assistants yourself. Open ChatGPT, Gemini, and Perplexity and ask for the best business in your category and suburb. Note whether you appear, and whether the details they give are right.
2. Make your details identical everywhere. Name, address, phone, hours, and services matching exactly across your Google Business Profile, Yelp, Facebook, and your own site. Identical, not similar.
3. Complete the Google Business Profile properly. Correct primary category, every service listed with its own description, current photos. This is the input Gemini reads directly.
4. Get reviews arriving steadily and reply to all of them. The 4.3 average among recommended locations is a floor to respect, not a target to game.
5. Write answer-first service pages. Put the plain answer in the first two sentences so an assistant can lift a clean line. Vague brochure copy gives it nothing to quote.
6. Re-check monthly. Assistants change their answers as their sources change. A single audit is a snapshot, not a system.

Steps two and three are where most of the movement comes from, and both are free. That is the honest headline under the scary one. The businesses being recommended are not doing anything clever. They are the ones whose information agrees with itself everywhere an assistant looks.

The five-minute monthly check

Since there is no dashboard for this, the check has to be manual, and it is short enough to do on a phone between jobs. Ask each assistant the same four things, in a fresh chat so previous context does not skew the answer:

• “Who is the best [your trade] in [your suburb]?” and see whether you are named at all.
• “What are [your business name]'s opening hours and phone number?” and check every digit.
• “Does [your business name] do [a service you actually offer]?” and see whether it knows.
• “Compare [you] and [your closest competitor]” and read what it says you are worse at.

That last one is uncomfortable and the most useful. An assistant will summarise your weaknesses from your reviews in a way no customer ever tells you to your face. I have had two clients change an operational habit, not a marketing tactic, off the back of what a chatbot said about their response times.

The bigger pattern

This sits alongside a second trend from the same period: AI Overviews now appear on around 68 percent of local business queries, against 39 percent for traditional local packs, per Whitespark data reported by Search Engine Journal. So an AI answer sits above the results on most local searches, and a separate set of assistants names a shortlist that most businesses never make. Both point the same way. Being findable is no longer the same as being chosen, which is the thing I keep writing about in Map Pack vs AI Search and Google AI Overviews and SEO.

One caution on the data. This is a multi-location enterprise sample, and a single-location business in a suburb faces a different competitive set. I would not treat 1.2 percent as your odds. I would treat it as evidence that the shortlist is short and the inputs decide who is on it. If you want that checked properly for your own business rather than guessed at, that is what the audit covers, and the ongoing side of it sits under AEO and GEO.

The questions below are the ones owners have asked me since this data landed.