Welcome to Insites Local AEO Benchmark
It's a monthly digest of how ChatGPT, Gemini, Copilot, Perplexity, Grok, Google AI Overviews and Google AI Mode identify, describe and recommend local businesses. Our sample this month includes 8,016 local businesses audited in September 2026 by Insites.
What's inside
- Does AI know about local businesses?
- What does AI know about local businesses?
- How many businesses does AI never recommend?
- How often does AI recommend SMBs?
- Which industry sectors are hardest to optimise for AI?
- What's AI saying about SMBs?
- What can agencies fix and bill for?
- What's changed since last month?
- A note from the Insites team
Does AI know about local businesses?
Whether AI can find a local business, whether it gets the details right, and whether it ever recommends it: the three things that decide if the customer ever hears about them.
of local businesses were identified by at least one of the up to seven AI assistants we asked
of businesses had their name, website or phone number wrong in at least one assistant's answer
of the time AI gave the right website, where the business publishes one
How to communicate this to SMB owners
Across a sample of 8,016 local businesses in September 2026, at least one of the up to seven AI assistants we asked could identify 99.1% of them by name, but when a customer described the job instead of naming anyone, 40.3% of those businesses weren't recommended for a single one of their services, and that's the part that costs you customers.
What does AI know about local businesses?
Whether AI gives out the right name, phone number and website for a business, because a wrong detail sends the customer somewhere else.
How often AI gets a business's details right
How to communicate this to SMB owners
Where a local business publishes its phone number, AI gives the right one only 53.2% of the time, against 69.5% for the name and 79.7% for the website, and the usual cause is an old listing or directory it's still reading from. A customer given the wrong number doesn't tell you. They just call the next business on the list, so your details need to match everywhere AI looks.
How many businesses does AI never recommend?
We asked AI five local, service-specific questions a real customer would ask about each business, and a business it never recommends is invisible to anyone asking AI who to use.
40.3%
of businesses were never recommended.
How often were businesses in our sample recommended?
How to communicate this to SMB owners
Across a sample of 8,016 local businesses in September 2026, 40.3% weren't recommended by AI for a single one of the services a customer asked about, 37.2% were recommended for some of them, and only 22.5% came up for everything they do, so the question worth asking about your business isn't whether AI knows you but which of your services it will actually put you forward for.
How often does AI recommend SMBs?
Ask AI about a business by name and it usually knows who you mean. Ask AI to recommend a local business and it usually mentions a local rival.
How often each assistant knows a business, and how often it recommends it
- Knows the business when you ask about it by name
- Recommends the business when a customer asks locally for what it sells
Gemini
Google AI Overviews
Copilot
Google AI Mode
ChatGPT
Grok
Perplexity
How to communicate this to SMB owners
Gemini could identify a business 94.0% of the time when asked by name, but only recommended it in 10.8% of the instances where a customer asked for a recommendation. Being known isn't the same as being recommended, and what closes that gap is giving AI more to go on: clear service pages, consistent listings and reviews that say what you do.
Which industry sectors are hardest to optimise for AI?
How hard it is to get recommended depends on the trade, which is worth knowing before you promise a client anything.
Which sectors AI recommends least
Hardest sectors
- Interior design 56.6%
- Flooring 54.8%
- Legal 54.6%
- Cleaning 51.6%
- Finance 49.4%
Easiest sectors
- Home maintenance 33.3%
- Restaurant & hospitality 31.5%
- Dentistry 28.9%
- Retail 27.8%
- Security 19.2%
How to communicate this to SMB owners
How hard it is to get recommended by AI depends a lot on the sector, and in this sample 56.6% of interior design businesses were never recommended for any of their services, against 19.2% of security businesses. In a competitive sector, results take longer to show, and in a quieter one there's a real chance to get ahead before your competitors do.
What's AI saying about SMBs?
The criticisms AI repeats when it describes local businesses, which is what the customer reads when considering whether to get in touch.
What AI criticises businesses for most
How to communicate this to SMB owners
Most often this month, the criticism AI raises is support and help, which comes up for 14.7% of the businesses in our sample. AI builds that picture from whatever it can find, mostly reviews, so the most useful thing you can do is collect reviews that name the specific service and say why it was good.
What can agencies fix and bill for?
The AI readiness checks most local business websites fail, and every one of them is work an agency can scope, fix and bill for.
Average website readiness score across the sample
The AI readiness checks most sites fail
How to communicate this to SMB owners
The average local website scores 64.6% on our AI readiness check, and 94.9% of local websites fail the FAQ structured data check. These are technical fixes, like the markup that tells AI what you sell and where, and the file that tells it which pages to read. Unlike rankings, they're entirely in our control, so we can correct them and help your AI visibility.
What's changed since last month?
How this month compares with last, so you can see which way AI is moving before your clients ask.
September vs. August
- AI is recommending local businesses less often 26.9% to 20.9% −5.9 pts
- AI is getting more business details wrong 85.8% to 94.3% +8.5 pts
- AI is getting fewer phone numbers right 61.3% to 53.2% −8.1 pts
- AI is reading businesses' own websites more 10.6% to 13.2% +2.7 pts
- Fewer businesses are being recommended first 29.7% to 26.6% −3.1 pts
- Average website readiness 64.6% Steady
- Gemini is recommending businesses less often 23.1% to 10.8% −12.3 pts
- Google AI Overviews is recommending businesses less often 15.2% to 8.1% −7.1 pts
- AI is getting fewer business names right 75.1% to 69.5% −5.7 pts
- Fewer sites are missing local business schema 63.7% to 60.0% −3.7 pts
- AI is getting fewer websites right 82.0% to 79.7% −2.3 pts
How to communicate this to SMB owners
Gemini is recommending businesses less often, going from 23.1% to 10.8% since August. That's a change across the whole market, so it's worth tracking whether your business moves with it, which is why we measure it the same way every month.
A note from the Insites team
Everything on this page is free to quote and free to use, in your own content, a proposal or a pitch deck.
Thanks so much for reading this far. If you do quote any of it, we'd love you to link back to Insites Local AEO Benchmark: it keeps the numbers traceable to the sample they came from, and we'd really appreciate the citation.
How we work this out
Everything here comes from real audits of real local businesses. Here's how we run them, what the numbers can fairly be used for, and where they stop being useful.
How we measure it
Every audit puts a local business to up to seven AI assistants and records what comes back. This page shows a sample of 8,016 businesses from September 2026. Each audit asks two kinds of question: one that names the business, which tests whether AI can find it at all, and one that names only a trade and a town, the way a customer would, which checks whether the business gets recommended.
What we ask
Branded questions supply a name and a location and ask the assistant to identify the business, then check the name, phone number and website it returns against what the business actually publishes. That's the being-found test. Unbranded questions ask what a customer would ask, such as the best plumber in a given town, and record every local business named and in what order.
Which assistants
ChatGPT, Gemini, Copilot, Perplexity, Grok, Google AI Overviews and Google AI Mode. The two Google surfaces are counted separately because they behave differently and cite different sources. Not every business is put to all seven: in this sample 31.2% were asked all of them and 48.9% were asked three, so any figure that says "at least one assistant" is read against up to seven, not exactly seven.
How a business enters the sample
All of the sample data is based on local businesses audited by Insites customers within the Insites platform, where they're analysing their own clients' AI visibility.
How website readiness is scored
Eleven technical checks run against the business's own site, covering structured data, indexing, crawler access, content depth and freshness.
What this cannot tell you
- The sample is drawn from our customers auditing real local businesses, rather than from every local business, so read the figures as direction rather than precision.
- Assistants are non-deterministic. The same question can return a different answer an hour later, which is why a move of less than a point is marked steady rather than reported as a change.
- Trade comparisons rest on far smaller samples than the headline figures, so read them as direction rather than precision.
- The month and the quarter are aggregated separately, so a figure for the month won't always look like a slice of the quarter.
- A few measures in each export don't survive our own checks. Those are left off the page rather than quoted with a caveat.
A note on the numbers. These figures cover a sample of 8,016 local businesses audited in September 2026, each put to up to seven AI assistants. Anything that didn't survive our own checks is left off rather than quoted with a caveat.
