The 12 AI assistants compared: how each one picks local businesses

Executive summary

Key findings

1. The assistants split into two camps. In October 2026, assistants that search the web mostly agree with each other, and assistants that answer from model memory mostly agree with each other. Every assistant's closest match is in its own camp: ChatGPT's is Gemini, Grok's is DeepSeek.
2. Memory assistants lean on national brands. 9.1% of their picks are national brands, against 3.8% for web-search assistants. In home services the gap is 16% against 6%, and in pest control 59% of Grok's picks are national brands.
3. Some assistants name far more businesses than others. In a typical market Claude names 9 businesses that make the rankings; Perplexity and Google AI Overviews name 4.
4. Agreement with the crowd varies widely. Gemini named the business the other eleven agreed on in 88% of markets. Perplexity and Google AI Overviews did so in only half.
5. Mistral goes its own way. 23% of its picks are businesses no other assistant named, against 1% for Google AI Overviews.
6. Memory answers held steadier than web answers. Of the five assistants we reached the same way in September and October, Grok, which answers from memory, named 71% of its September picks again. Gemini named 40% and Google AI Overviews 33%.
7. Each assistant has blind spots. Outside California, Grok named a business that made the rankings in only 38% of legal markets and 33% of medical and cosmetic ones. Google AI Overviews did so in 64% of hotel and restaurant markets.

What it means. There is no single "AI" to be visible in. Each assistant draws its shortlist differently, so a business can lead in one and be absent from another. Assistants that search the web reflect what is on the web now; assistants that answer from memory reflect what the model learned when it was trained, which appears to favor long-established national names. These are observed patterns across 600 markets, not proof of how any one assistant is built.

Scope. 12 assistants, 600 markets (30 metros in Florida, California, New York and Texas, 20 types of business), 3 fixed questions each, asked between 1 and 5 October 2026. Month-to-month figures compare the same 245 Florida and California markets in September, and separate the assistants we reached the same way in both months from those we did not.

How do we compare the assistants?

Each month AI Syndicate Local asks every assistant the same three questions about a type of business in a metro area: the best one, the one to recommend to someone who needs one, and an affordable one nearby. A business makes a market's ranking when at least two independent answers name it. In this report, an assistant "named" a business when at least one of its answers named a business that made that market's ranking.

The twelve fall into three groups by how they answer:

• Web search: ChatGPT, Claude, Gemini and Perplexity answer with web search on. Google AI Overviews is the AI Overview Google shows for the question. Microsoft Copilot is the Copilot answer Bing shows; when Bing shows none, we reconstruct one from the same Bing results.
• Model memory: Grok, Meta AI, DeepSeek and Mistral answer from what the model learned, because no consumer live-search access is available for them.
• Voice: Siri and AAlexa cannot be asked programmatically, so we reconstruct their answers in the short, single-pick style each speaks in. Their results reflect that style, and we leave them out of the comparisons where it would mislead.

A national brand here is one of 20 businesses that are in the top three of five or more October markets, such as Orkin, Terminix, Servpro and Renewal by Andersen.

How many businesses does each assistant name?

It varies more than twofold. In a typical market, Claude names 9 businesses that make the rankings across its three answers, Gemini and Copilot 8, and Perplexity and Google AI Overviews only 4.

Google AI Overviews' shorter lists are more concentrated among market leaders: 38% of its picks were top-three businesses, against 25% for Claude. Siri and Alexa sit at the extreme because they are reconstructed in a voice assistant's style, which gives one best pick rather than a list.

For a business, a long list is easier to get onto and a short list is worth more. Being one of Claude's nine is common; being one of Google AI Overviews' four usually means being a market leader.

Which assistants agree with the crowd?

We left each assistant out in turn, found the business the other eleven named most widely, and checked whether the assistant named it too.

Gemini is the closest thing to a bellwether at 88%. Copilot and Claude follow. Perplexity and Google AI Overviews agreed only half the time, which usually means they were naming businesses the others were not.

The most agreeable assistant depends on the trade:

Hotels and restaurants have the strongest consensus of any trade: most assistants name the same well-known places.

Do the assistants fall into camps?

Yes. For every pair of assistants we measured how much their picks overlap in a market: the share of the businesses either one named that both named. Assistants that search the web overlap most with each other, and assistants that answer from memory overlap most with each other.

Every assistant's closest match is in its own camp:

The least alike pairs all cross the camps: Google AI Overviews with Mistral (8%), with Grok (9%) and with DeepSeek (9%), and ChatGPT with Mistral (9%). Even the most alike pair shares fewer than four in ten of its picks, so no two assistants give the same shortlist.

Which assistants favor national brands?

Assistants that answer from memory. Across all 600 markets, 9.1% of their picks are national brands, against 3.8% for assistants that search the web: 2.4 times as many.

Grok stands out: one in six of its picks is a national brand, against one in fifty for ChatGPT. The gap is largest in home services, where national brands operate in many cities: 16% of memory assistants' picks there are national brands, against 6% for web-search assistants. In pest control, 59% of Grok's picks are national brands.

A likely reason is what each camp draws on. A model answering from memory knows the names that were written about most widely before it was trained, which favors brands with a national footprint. An assistant that searches the web sees the review sites and local lists that rank the businesses in that city. This is our reading of the pattern, not something the data proves.

Which assistants find businesses no one else names?

A business can make a market's ranking on one assistant's word alone when that assistant names it in two of its three answers. Those solo picks show which assistants go their own way.

Mistral is the clear outlier: nearly a quarter of its picks are businesses no other assistant named. Memory assistants as a group make twice as many solo picks as web-search assistants (10.7% of picks against 5.3%). Google AI Overviews almost never does.

A solo pick can be a genuine local favorite the others missed, or a business the model remembers that is no longer the best answer, or no longer trading. We cannot tell which from the answers alone, so we read solo picks as a sign of independence, not of quality.

What changed for each assistant from September to October?

We measured the same 245 Florida and California markets in September and October. The two tables compare each assistant across the two months.

One caution comes first. Between the two measurements we changed how we reach seven of the twelve assistants (see the method notes). For those, a change between the months can come from us rather than from the assistant, so we show their figures but do not read them as the assistant changing. Gemini, Google AI Overviews, Siri, Alexa and Grok were reached exactly the same way in both months.

How we reached each assistant, and how much it said (September → October)

How its picks changed (September → October)

† In California our collection of Google AI Overviews, Microsoft Copilot and Alexa answers fell short in October. That gap, not the assistants, explains most of their lower share of markets. National-brand shares here cover the 245 compared markets only, so they differ slightly from the 600-market figures above. "Named again" counts a business as the same when it keeps its listing, website or name; see the statistical notes.

Among the five assistants we reached the same way, Grok, which answers from memory, was the most consistent. It named 71% of its September picks again (95% range 66–75%). The assistants that build answers from live web results reshuffled more: Gemini 40%, Google AI Overviews 33%, and the reconstructed Siri and AAlexa, which draw on web and maps results, 30% and 27%. That fits how they work. A model's memory changes only when it is retrained, while the pages behind a live answer change all the time. Grok is a single memory assistant, so treat this as a pattern to watch, not a rule.

Some of the five changed visibly. Gemini named fewer businesses per market (11 to 8) and agreed with the others a little less. Siri, whose reconstruction runs on Gemini, went from 6 businesses per market to 3. Google AI Overviews agreed with the others less often (61% to 43%), while Grok agreed more (44% to 55%). These may reflect changes on the providers' side, which we cannot see inside.

For the seven assistants we reached differently, read the October figures on their own. Perplexity is the clearest case: its drop from 9 businesses per market to 4, and its agreement falling from 91% to 48%, came with its move to a new interface, so it is not evidence that Perplexity itself changed how it answers people.

Did the sites each assistant cites change from September to October?

For most assistants, not much. We grouped every site an assistant cited into a few kinds: "best of" and ranking lists, review sites and directories (such as Yelp, Google Maps, Angi and legal directories), forums and social sites (such as Reddit), and businesses' own websites. Then we compared each assistant's mix in September and October, counting only markets where it cited sources in both months.

The two clearest shifts came from assistants we reached the same way both months. Google AI Overviews leaned more on Google's own listings, and Gemini on Reddit. For a business, both point to the same two places to keep an eye on: its Google Business Profile, and how people talk about it on Reddit.

These figures come from our own record of each answer's sources, which is not in the public data. Dictionary sites, which assistants cite to define a word, are left out. A shift of under about five percentage points counts as no big change.

Where does each assistant go quiet?

Most assistants named ranked businesses in almost every market. A few have clear gaps by type of business. To keep our California collection gap out of the picture, this table uses only Florida, New York and Texas (21 markets per type of business).

Each cell is the share of markets where the assistant named at least one business that made the ranking. A low share can mean the assistant declined to recommend, or that its picks were not confirmed by any other answer.

Grok rarely names a ranked lawyer, dentist or med spa, but almost always names a ranked hotel or restaurant. Google AI Overviews is the reverse for hospitality: it named a ranked hotel or restaurant in only 64% of those markets. For a business in one of these trades, it is worth knowing which assistants rarely name anyone.

What should a local business do with this?

• Measure visibility as a portfolio, not a rank. There is no single AI ranking. The useful question is not "What is my AI ranking?" but "How often am I recommended, by which assistants, for which questions, in which markets, and is that rising or falling?" We call that share of recommendation.
• Check each assistant your customers use, not "AI" in general. No two assistants share even four in ten of their picks, so position in one says little about another.
• Know which camp matters for your trade. Web-search assistants follow what is on the web now. Memory assistants follow what was written about you over years, and lean toward national names.
• Aim for the short lists. Being one of Google AI Overviews' four picks usually means being a market leader; being one of Claude's nine is easier but worth less.
• Read month-to-month moves by assistant. In this measurement, web-search answers reshuffled more than memory answers: a drop in Gemini or Google AI Overviews is common and often temporary, while a drop in Grok is more unusual.
• If you compete with a national brand, expect it to be strongest in the memory assistants, and focus on the web-search assistants, where local review sites and lists carry more weight.

These are patterns across 600 markets, not guarantees for any one business.

How this data is used. This research feeds the AI Syndicate platform, so our clients have the leading edge. Clients see where they stand in their market's AI Local ranking, and how they have moved, inside their dashboard. Their markets are measured more often than the rest of the index, with up to three times as many answers behind each reading. What we learn across the whole index, such as which assistants name whom, which sites they cite and how each assistant changes from month to month, is built into the twelve-assistant measurement the platform runs for each client's own business and questions. When an assistant changes how it answers, or how we can reach it, the platform changes with it.

How was this measured, and what can it not tell you?

Method. The October 2026 edition of AI Syndicate Local asked 12 assistants three fixed questions in each of 600 markets, with the city named in the question and no user location or account history. A model extracts business names from each answer, spellings of the same business are merged, and a chain's locations that share one website count as one business. Rankings need at least two independent answers. The full method is on the methodology page.

What changed between September and October. Besides the assistants' own answers, two things changed. In October we re-asked any question that went unanswered the first time, which raised the share answered from 73% to 92% in the compared markets. And we changed how we reach several assistants. Perplexity retired the interface we used in September, so its October answers come through its new one. Meta AI moved to Meta's own interface. DeepSeek and Mistral had fixes that let far more questions through. ChatGPT, Claude and Microsoft Copilot had smaller changes to how answers are fetched. Gemini, Google AI Overviews, Siri, Alexa and Grok were reached the same way in both months.

What "named" covers. This report sees the businesses that made a market's ranking, with the list of assistants that named each one. A business an assistant named only once, that no other answer confirmed, is not in the published data, so each assistant's full list is somewhat longer than shown here.

What this cannot tell you.

• Citation changes by assistant come from our own record of each answer's sources, grouped into a few kinds of site. The public data pools sources across assistants, so these cannot be rebuilt from it.
• Siri and Alexa are reconstructions, and Copilot is when Bing shows no Copilot answer.
• Grok, Meta AI, DeepSeek and Mistral answer from model memory because we cannot reach their live search; their consumer apps may search the web and answer differently.
• In California, Google AI Overviews, Copilot and Alexa returned far fewer usable answers in this edition, which is a gap in our collection. Trade comparisons use the other three states for that reason.
• Month-to-month figures count only businesses that made the rankings in both months. For the seven assistants we reached differently in October, a change between the months may come from us, not the assistant.
• Some trade groups are small (as few as 21 markets in the three-state table), so their figures are directional.

What stays private. The public index publishes each market's rankings and the assistants that named each business. The individual answers and our records of what changed on ranked businesses' websites are not published.

Statistical notes

• Unit of measurement. Shares of markets count each market equally. Shares of picks pool every assistant's picks across all markets. A market is one metro area and one type of business.
• Questions. Each market gets three fixed questions, each asked once of each assistant, and the three count equally. Because each is asked once a month, we cannot yet separate real change from answer-to-answer variation.
• Missing answers. A question that returns no answer is left out, not counted as naming no one. Each assistant's figures cover only the markets where it named a ranked business.
• Agreement. For each assistant we leave it out, take the business the other eleven named most widely, and check whether the assistant named it. Its own vote never counts toward the consensus it is measured against.
• Overlap. For two assistants in one market: the businesses both named, divided by the businesses either named. We average that over the markets where both named someone, then over the pairs in a group.
• National brand. One of 20 businesses in the top three of five or more October markets.
• Ranges. 95% ranges come from resampling markets 2,000 times, for example national-brand shares of 9.1% (8.1–9.9%) for memory assistants and 3.8% (3.4–4.2%) for web-search assistants, a ratio of 2.4 (2.3–2.6).
• Matching businesses across months. Two entries are the same business if they share a listing, a website, or a name once "the", "and", "of" and legal suffixes are ignored, or if one name of two or more words is the start of the other. Two different websites are two businesses unless the names are identical.
• Typical values. "Typical" means the median market.

How can I cite or reuse these numbers?

Quote any figure in this report with credit to AI Syndicate Research and a link to this page. You may quote up to 50 words of text, or share a single chart with its source line intact. Reproducing the report or large parts of it needs written permission. A suggested citation:

A PDF copy and a plain-text Markdown version are available. The rankings behind the report, with the assistants that named each business, are published under a CC BY 4.0 licence at aisyndicate.com/ai-local/, with machine-readable copies at /ai-local/data.json and /ai-local/all.md. Journalists and researchers who need a cut of the data that is not here can write to support@aisyndicate.com.

Key figures

Every key figure in this report, in one table. The same figures are in findings.json and findings.csv, free to reuse under a CC BY 4.0 licence. Ranges are 95% intervals from resampling markets.

Sources

1. AI Syndicate Local: the October 2026 edition (600 markets) — AI Syndicate. Read Oct 6, 2026.
2. AI Syndicate Local methodology: how the index is measured — AI Syndicate. Read Oct 6, 2026.
3. AI Syndicate Local, machine-readable rankings with the assistants that named each business (CC BY 4.0) — AI Syndicate. Read Oct 6, 2026.