Local Businesses Miss the AI Answer 84% of the Time. Here Is Where

When ChatGPT names three electricians in a town and yours is not one of them, the loss did not happen on your website. It happened before the answer ever reached your page. Across 9,454 question-by-business observations we measured, 84.3% of those checks came back with the business nowhere in the sources the answer used.

Key takeaways

Across 9,454 question-by-business observations, 84.3% of checks found the business was not among the sources the AI answer used, so roughly 94% of measured failure happens before selection.
The strongest predictor of clearing that stage is Google organic presence for the exact question, with an odds ratio of 12.74 and a McFadden R-squared of 0.24.
Nine on-page technical signals were tested against the outcome, and none survived a Bonferroni correction. Page tweaks were not what moved the needle.
This is an association, not a task. We cannot tell you that fixing Google presence causes an AI citation, and anyone who says they can is guessing.
ChatGPT leans on Foursquare Places and now Yelp, Gemini is grounded in Google Maps, Perplexity leans on Yelp and TripAdvisor, and Claude leans on Brave Search. Different engines, different reading lists.
BrightLocal's February 11, 2026 survey (n=1,002) found consumers using AI to find local businesses jumped from 6% to 45% year over year, with ChatGPT at 31%.

What the 84% number actually means

We ran 9,454 question-by-business checks across the four main AI engines. Each check asked a real buyer-style question ("best electrician in [town]", "who fixes panel upgrades near me", "24 hour electrician for a burning smell") and recorded which businesses the answer named in its sources.

In 84.3% of those checks, the business we were watching was not in the sources list. Not ranked last. Not mentioned briefly. Simply not there.

That is the retrieval gate. Before the AI writes a single word, it pulls a short list of sources it will read from. If your business is not on that list, nothing on your page matters, because nothing on your page is being read. Everything downstream (how well your services are described, what your reviews say, how the page is laid out) only matters for the 15.7% who made it in.

A careful note on wording, because it matters. We do not say "AI did not find you." We see the bibliography the engine returned. We do not see what it retrieved and threw out. The correct phrase, and the one we use in every report, is: "not among the sources this answer used." That distinction is the difference between measuring something and guessing at it.

What actually predicts making it into the sources list

We tested a lot of things. On the page: word count, page speed, mobile layout, image alt text, structured data blocks, freshness, internal linking depth, header structure, and a couple more. Nine technical signals in all. After applying a Bonferroni correction (the statistical guardrail against finding fake patterns when you test many things at once), none of them survived. That does not mean none of them ever matter. It means none of them explained the retrieval gate in our data.

One thing did. Whether the business ranked in Google's organic results for the exact question the AI was answering. The odds ratio was 12.74. In plain English, businesses that showed up in Google organic for a query were far more likely to be among the sources the AI answer used for that same query. The model's McFadden R-squared was 0.24, which is a strong result for this kind of measurement.

Here is the trap, and this is where almost every AI visibility pitch you have seen gets dishonest. We measured an association. We did not run a controlled experiment where we improved a business's Google ranking and then watched what happened to their AI presence. We cannot tell you that fixing one causes the other. It is entirely possible that the same underlying things (a real business, a website that has been indexed for years, a Google Business Profile with real reviews, consistent name and address across directories) drive both. If you buy a tool or a course that promises "do X and AI will cite you," ask them where the causal evidence is. There is not any yet, at least not that anyone has published.

Saying this out loud is uncomfortable. It is also the more useful position, because it tells you where to actually look. The retrieval gate is not a page-level problem. It is a presence problem: are you visible in the places the engine already checks for your town.

Where each AI engine actually looks for local businesses

This is the part almost nobody writes correctly, and it is why "one AI SEO checklist for all engines" is nonsense.

ChatGPT leans on Foursquare Places, thanks to a partnership OpenAI announced on December 16, 2024 that covers over 100 million points of interest in 200+ countries. Since July 23, 2026, ChatGPT also pulls Yelp reviews, ratings and photos directly, with reservations and waitlists added on August 10, 2026. When ChatGPT falls back for local, it falls back to Bing Places.
Gemini is grounded in Google Maps and Google Search. That means your Google Business Profile is doing double duty here.
Perplexity leans on Yelp, TripAdvisor and industry directories. If you are a med spa, that means Zocdoc-shaped sources matter. If you are a roofer, that means directories like HomeAdvisor and Angi.
Claude leans on Brave Search, which weights independent sites and directories differently than Google does.

You do not need to memorise this. You do need to understand the shape: each engine has a small handful of places it treats as the reading list for "who does electrical work in this town." If your name is not present, consistent, and recently active on those sources, you are not on the list.

The size of the shift is real. BrightLocal's Local Consumer Review Survey, published February 11, 2026 with 1,002 US adults, found the share of consumers using AI to find local businesses rose from 6% to 45% year over year. ChatGPT alone accounted for 31%. A Foundation and AirOps analysis of 28 million AI responses in Q4 2025 found Yelp took 512,680 citations, 3.4x the next platform, and that 91 to 97% of AI citations came from category queries ("best electrician in Newark") rather than brand searches. Category queries are exactly the questions where the retrieval gate decides who gets to be in the answer.

What this means for a family electrical contractor

Consider a family-run electrical shop in a mid-sized town. Two working brothers, one office manager, thirty years in the trade, real reviews, well-liked. When a homeowner asks ChatGPT "who does panel upgrades in [their town]," the shop is not among the sources the answer uses. The AI names two competitors, one of which is a national franchise that only opened locally last year.

The instinct is to blame the website. Add more content. Write blog posts about GFCI outlets. Get a new theme.

The measurement says that is the wrong lever. The website is not being read, because the shop is not on the reading list. The relevant questions are different:

Does the Google Business Profile show up in Maps for the exact services offered, in the exact towns served?
Is the name, address, and phone number identical across Google, Bing Places, Yelp, and the industry directories a homeowner or an AI would check?
Are recent reviews (last 60 to 90 days) present, and do the actual words in those reviews mention "panel upgrade" and the town name?
Is the shop named in any of the local "best electrician in [town]" roundups the local paper, chamber, or a nearby review site publishes?
Is there a page on the site for each real service in each real town, or one generic "we serve the tri-state area" page?

None of that is about schema or headings or a shiny new theme. It is about being present, consistently, in the sources the AI is going to check anyway. That work is unglamorous. It also matches what our numbers say actually predicts making it through the gate. For a fuller walk through the shape of that work, local SEO in the AI era covers the ground methodically.

Why on-page fixes did not survive the analysis

Here is the intuition. The retrieval gate is a filter that runs before the answer. It is asking, roughly: for this question and this town, which sources should I pull from. That decision is based on presence in the engine's reading list (Foursquare, Google Maps, Yelp, Brave, Bing Places, category directories, best-of roundups) and on whether your site already ranks for the question in Google.

By the time your page's headings or word count get a vote, the retrieval gate has already decided whether you are in the room. The page-level work matters after the gate, for how the AI describes you once it has chosen to include you. For a lot of local businesses, the room is still empty. Fix that first.

If you are curious about how these categories differ (on-page work, page-level readability for AI, presence in listings), the piece on local citations vs AI citations walks through the term collision carefully, because "citation" means two different things depending on which world you are standing in.

What to do with this if you run a local business

Three moves, in order.

First, run the free AI Visibility Check for your business. It asks the same category questions across the four main engines and shows you which sources they cited for your town. If you are in the 84%, that is the diagnosis. If you are in the 15.7%, the report tells you where and how.

Second, before you spend money on page-level "AI optimisation," get your presence right in the places the engines actually read. Google Business Profile complete and accurate. Bing Places claimed. Yelp claimed and current. Industry directories consistent. Reviews recent and specific to the services and towns you serve. A page on your site for each service in each town you actually work in. This is boring, familiar work. Our measurement says it is where the leverage is.

Third, be skeptical of any local AI visibility service that will not explain the difference between association and causation. If they promise "do this and AI will cite you," they are either not measuring the gate or not telling you what their measurement actually shows. The honest position is that Google presence for the query is far and away the strongest predictor we have, and that the causal chain from A to B has not been proven. That is still an incredibly useful thing to know, because it points at real work. It is not the same as a guarantee, and anyone selling you a guarantee should scare you.

Frequently asked questions

What does the 84% number actually measure?

It measures how often a business was not among the sources an AI answer used, across 9,454 question-by-business observations we ran on the four main AI engines. It is a snapshot of the retrieval gate, not a prediction of any specific business's outcome. If you want to see where your own business sits, check if AI recommends your local business.

If Google presence predicts AI citation, is fixing my SEO the answer?

Improving your Google presence for the exact questions buyers ask is the single strongest lever we have identified. What we cannot tell you is that fixing it causes an AI citation, because we measured association and not causation. Treat it as the most likely place to find leverage, and treat any promise of a guaranteed outcome as a red flag.

Do reviews matter for AI recommendations?

Reviews matter for two reasons. They are part of what platforms like Yelp and Google Business Profile score you on, which affects your presence in the reading lists AI engines actually read. And the words in your recent reviews (specific services, specific towns, specific problems solved) are readable by the engines that pull from those platforms. Old, generic reviews do less work than recent, specific ones.

Is the word "citation" the same in AI and local SEO?

No, and this trips people up. In local SEO, a citation means a listing of your name, address and phone number on a directory. In AI search, a citation means an engine naming your business and pointing to your page in its answer. Same word, two different things. The pieces you see on our site are careful to say which one they mean.

What about my competitor who ranks nowhere but keeps getting recommended?

That happens, and it usually points to one of three things. They are strong on a directory the engine leans on that you have overlooked. They are named in a local best-of roundup that keeps getting quoted. Or the engine is treating your town's category as thin and reaching for whatever it can find. The pattern behind AI recommending a competitor in your town is worth reading in full.

If you have not looked yet, the fastest way to see where you actually stand is to run the free AI Visibility Check. It asks the category questions a real buyer in your town would ask, records which sources each engine used, and tells you whether you were in the 15.7% or the 84%. From there, the work becomes concrete: which listings need cleaning up, which directories you are missing from, which service-and-town pages do not exist yet. That is the work the measurement says matters.