Statistics
AI local search statistics, traced to source
The deep-dive: how ChatGPT, Perplexity, Gemini and Google's AI Overviews actually discover and recommend local businesses — where the evidence is genuinely rigorous, and where the industry is repeating numbers that don't survive being traced.
This page is the full version of a topic the local SEO statistics page only summarises: every platform SOCi's index now scores rather than just ChatGPT, and what Google's AI Overviews actually do for local-intent queries specifically rather than the blended average usually quoted.
It goes on to cover where each assistant's local data is really sourced from and how fast that changed in 2026, whether schema markup does what the industry claims for AI citation, and why adoption and trust have moved in opposite directions rather than together.
Two findings here changed materially within the year before it was written: the claim that Foursquare supplies most of ChatGPT's local results stopped being true when OpenAI struck a data partnership with Yelp in July 2026, and the share of Google AI Overview citations coming from a page's own top-10 ranking fell from 76% to 38% in eight months.
Both are treated below as evidence of how fast this category moves, not as settled facts. See the local SEO statistics page for the wider ranking-factor picture these findings sit inside, and the Google Maps statistics for the local-pack surface AI recommendation keeps being measured against. This page is reviewed quarterly at minimum, for exactly the reason above.
How the grading works
The grading is deliberately unforgiving. A statistic from a good organisation still earns a C if the sample size is not published beside the number, because a reader cannot check what they cannot see.
Cite freely
Primary source, dated, with sample and method published. Quote it, with its scope attached.
Cite with scope
Real research with a published sample, but narrower than the way it is usually quoted — self-reported, or one country, or one kind of business.
Cite with caution
Named source and plausible method, but the sample size or methodology is not published where the number is. Attribute it, do not lean on it.
Do not cite
Untraceable, undated, or superseded by later changes to the thing it measured. These are the numbers keeping the roundups alive.
AI recommendations are far more selective than Google's local pack — and the three assistants aren't equally selective with each other
This is the best-documented finding in this category, because the organisation behind it built and published a genuine cross-platform methodology rather than eyeballing a handful of prompts. The 2026 edition of the index is also the first to score Perplexity and Gemini alongside ChatGPT, which matters, because treating “AI recommendation” as one undifferentiated thing turns out to hide most of what's actually happening.
1.2%
Grade Aof locations were recommended by ChatGPT when asked for local businesses, against 35.9% that appeared in Google's local 3-pack for the same queries — roughly thirty times more selective.
What it does not say The scope limit matters: this covers multi-location brands with existing digital infrastructure, not independent local businesses. Whether an independent single-location business faces the same roughly 30× selectivity gap is unmeasured, and should not be assumed to be the same.
<50%
Grade Aof businesses leading Google's local search results also appear in AI-generated recommendations for the same query.
What it does not say Ranking #1 on Google is not a reliable predictor of AI inclusion. Whatever AI assistants are actually weighting, it diverges meaningfully from what wins the local pack.
7.4%
Grade Aof locations were recommended by Perplexity in the same test — roughly six times more permissive than ChatGPT's 1.2%, but still less than a quarter as generous as Google's local pack.
What it does not say The same multi-location scope limit applies here as to every figure from this study.
11%
Grade Aof locations were recommended by Gemini — the most permissive of the three AI assistants tested, though still roughly a third the rate of Google's own local pack.
What it does not say Gemini's comparative generosity tracks with the accuracy figure below — it is the only one of the three assistants grounded natively in Google's own Maps and Business Profile data, the same data source the local pack itself draws on.
68% vs 100%
Grade Awas the business-profile data accuracy SOCi measured for ChatGPT and Perplexity, against 100% for Gemini, which the report attributes to Gemini being grounded directly in Google Maps rather than retrieving from the open web.
What it does not say An assistant recommending a business off inaccurate hours or a disconnected phone number arguably does the business more harm than not recommending it at all — this figure is the clearest evidence that “does the AI mention me” and “does the AI mention me correctly” are two separate questions.
| Platform | Locations recommended | Business data accuracy | How it's grounded |
|---|---|---|---|
| ChatGPT | 1.2% | 68% | Yelp data-licensing partnership (from July 2026) plus Bing-indexed web search |
| Perplexity | 7.4% | 68% | Open-web retrieval, observably weighted toward Yelp and TripAdvisor listings |
| Gemini | 11% | 100% | Native Google Maps / Business Profile grounding |
| Google local 3-pack (for scale) | 35.9% | — | Google Business Profile, Google's own local index |
Recommendation rate and accuracy figures describe multi-location brands specifically (~350,000 locations, 2,751 brands), not independent local businesses. The local-pack row is included for scale, not as a like-for-like accuracy comparison. Source: SOCi, Local Visibility Index 2026
What this means inside FOUND → TRUSTED → CHOSEN
The FOUND → TRUSTED → CHOSEN framework is the lens here. A 30× selectivity gap between what ChatGPT recommends and what the local pack shows means AI recommendation currently behaves like a harder, later filter — closer to being chosen than to being found.
Getting a Business Profile into local-pack shape earns a place in the pack; it does not automatically earn a mention from an assistant deciding which one or two businesses to name out of the field it retrieved.
The 68%-versus-100% accuracy gap adds a second, separate failure mode on top of the selectivity one: a business can be mentioned by ChatGPT or Perplexity with the wrong hours or a dead phone number, which converts an AI mention into a lost customer rather than a found one. The Google Business Profile statistics cover how often those underlying details are wrong in the first place.
Until a study measures independent, single-location businesses rather than multi-location brands, the safest reading is that AI recommendation adds a fourth surface to compete on, not a replacement for the first three. Gemini's near-perfect accuracy figure is a direct product of using data Google already controls — which is not a lever available to a business trying to influence ChatGPT or Perplexity by the same route.
Google's AI Overviews trigger far less often for local-intent searches than for informational ones
Most of what's written about AI Overviews and local business treats “does an Overview appear” as one number. The one study that actually broke local search queries down by intent found a large gap hiding inside that single number — and it is over a year old by the date of this page, in a category where a year is a long time, so treat the shape of the finding as more durable than the exact percentages.
68%
Grade Bof local business queries triggered an AI Overview on average, ranging from 57% to 80% depending on the city and industry tested.
What it does not say Published 12 May 2025 — over sixteen months before this page's own review date, in a category that has moved substantially since (see the citation-position finding further down this page). Treat the 57–80% range as a snapshot of that period, not a current figure, and expect it has shifted.
15% vs 92%
Grade Bwas the AI Overview trigger rate for narrowly local-intent queries (for example, “plumber near me”) versus informational queries in the same six industries (for example, “how to unclog a drain”) — a large gap the single 68% average figure above hides completely.
What it does not say Hybrid queries (mixing local and informational intent) triggered Overviews 97% of the time — the highest of the three intent categories, and a reminder that most real searches don't sit cleanly in one bucket.
39%
Grade Bwas the average rate at which Google's local 3-pack appeared for the same query set — an inverse relationship with AI Overview prevalence across the sample.
60% vs 40%
Grade Cof AI Overview citations for a sampled set of Houston plumber queries pointed to third-party publishers, against 40% citing an individual local business directly.
What it does not say Graded down specifically for sample size: 23 citations from one city and one trade is an illustrative subsample, not a result that should be generalised to “local AI Overviews” as a category.
The single most useful number in this dataset is not the 68% headline average — it's the gap between 15% and 92%. A search that reads as purely transactional and local, the kind aimed squarely at finding a business right now, triggered an AI Overview roughly one time in six across this sample. A search that reads as informational, even in the same trade and the same city, triggered one roughly nine times in ten.
Read plainly, that means the AI Overviews panic in local SEO is currently overweighted toward the query type least likely to actually cost a business its click.
“Plumber near me” mostly still returns a conventional local-pack-and-organic result page, while “how do I know if I have a slab leak” is the search far more likely to get answered inside an AI box before the searcher reaches a business's own site. The local search behaviour statistics set out how those two query types differ in intent before any AI layer is involved.
What this means for where to focus content effort
The practical implication is not that AI Overviews don't matter for local business — the 60/40 citation split shows a real local business can and does get named directly, not just referenced through a third-party publisher.
It's that the risk is concentrated in the informational layer of a local business's content — the how-to and explainer pages that sit above the commercial pages — rather than in the “near me” and category pages that most local SEO effort still targets first.
A business with strong service pages but no informational content is optimising for the query type this dataset says is least likely to be intercepted by an AI Overview, and under-investing in the one most likely to be.
Where each assistant's local business data actually comes from — and how fast that has changed in 2026
This is one of the few places in this category where the ground has moved enough within a single year that a widely repeated statistic became wrong while this page's sources were still being verified. The safest way through it is to separate what's officially confirmed from what's inferred by practitioners testing outputs, because the industry consistently blurs the two.
58% / 27% / 15%
Grade Cwas the source mix ChatGPT Search drew on for local queries in 2024 — 58% business websites, 27% business mentions on other sites, 15% online directories.
What it does not say This is now closer to history than to current behaviour: OpenAI's data partnership with Yelp, dated below, did not exist when this snapshot was taken. Model and retrieval updates happen frequently enough that any single-point-in-time citation-mix study should be treated as a description of one moment, not an ongoing baseline.
330 million+
Grade AYelp reviews, across more than 8 million business listings, now feed ChatGPT's local business answers directly, following a non-exclusive data-licensing partnership between OpenAI and Yelp.
What it does not say The underlying data-licensing deal was reported first, on 23 July 2026; Yelp's own blog confirmed the consumer-facing integration — reviews, ratings, photos and business details inside ChatGPT, plus in-chat Reservations and Waitlist for thousands of restaurants — on 10 August 2026. Non-exclusive terms mean the same data could extend to other AI platforms.
0.00%
Grade Bwas the share of ChatGPT's primary-surface local citations traced to Foursquare in a large, current US test — against a widely repeated claim that Foursquare supplies 60–70% of ChatGPT's local results.
What it does not say A single agency's own testing, not independently replicated or peer-reviewed — graded down for that, even though the disclosed methodology (prompt count, markets, categories) is more transparent than most sources in this category. See the “numbers that did not survive the trace” section below for how the 60–70% Foursquare figure it's testing against originated.
No official documentation from Perplexity about how it sources or weights local business data was identified in the sources reviewed on 8 September 2026. What's observable rather than confirmed: practitioners testing Perplexity's local outputs consistently report a heavy lean toward Yelp and TripAdvisor listings and their review content, with the assistant's own explanations for a recommendation frequently quoting or paraphrasing review text directly.
That pattern is inference from watching what the assistant outputs, not a disclosed sourcing policy, and it should be cited as such rather than as a confirmed mechanism. It is the same distinction this page draws for ChatGPT, where the sourcing claim above is graded A specifically because Yelp and OpenAI both confirmed it publicly, not because a third party inferred it from testing.
What this means for where to spend directory effort in 2026
Two of the three major assistants — ChatGPT by confirmed partnership, Perplexity by consistent observed pattern — now appear to lean heavily on Yelp specifically, not on the broader open web in equal measure. Gemini is the outlier, grounded natively in data Google already controls.
The practical read is that a complete, accurate, actively-managed Yelp profile has gone from “one directory among many” to a plausible single point of leverage across two of the three assistants a local business is likely to be evaluated by. Since what those assistants quote back is largely review text, that is worth confirming directly for a given business rather than assuming — none of it is guaranteed to hold the next time a licensing deal changes.
Schema markup is not the lever most SEO advice claims it is for AI citation
This is the clearest case in this category of an entire tactic being sold on a statistic that does not survive being checked against the platform's own documentation or a controlled test. It is worth walking through in full, because the same “add schema, get cited more” claim appears on a large share of the pages selling AI-search optimisation services.
−4.6% to +2.2%
Grade Awas the change in AI citation rate after adding JSON-LD schema markup to pages that were already being cited heavily — a small, statistically significant decline for Google AI Overviews, and changes statistically indistinguishable from zero for Google AI Mode and ChatGPT.
What it does not say The study's own stated limitation matters: it tested only pages already receiving 100+ AI Overview citations before the change, so it says nothing about whether schema helps a previously-uncited page get its first citation. Within that scope, the authors' own conclusion is blunt: “we can't tell whether the schema did a tiny bit of good or nothing at all.”
76% to 38%
Grade Awas the share of Google AI Overview citations coming from a page ranking in the top 10 organic results for the same query — a general, web-wide finding (not local-query-specific) that fell by more than half in eight months, attributed to Google's January 2026 rollout of Gemini 3 as the default AI Overviews model and a heavier reliance on query “fan-out”.
What it does not say This is not a local-query-specific finding — Ahrefs' own report does not break the result down by query type, and no local-specific version of this analysis was identified in the sources reviewed on 8 September 2026. It is included because the scale of the shift, in eight months, is itself the relevant fact for this page: whatever governs AI citation is not stable enough for a 2025 finding to be trusted as a 2026 description.
| Factor | Score |
|---|---|
| Presence on expert-curated “best of” lists | 179 |
| Dedicated page for each individual service | 170 |
| Prominence on key industry-relevant domains | 167 |
| Quality and authority of unstructured citations (mentions with no link) | 160 |
| Authority of third-party review sites | 156 |
New scoring dimension for 2026, introduced because AI assistants draw on different data than Google's local pack. Whitespark's own methodology note flags this column as harder for its 47-person expert panel to score consistently than the established local-pack factors, since different platforms have very different data access — ChatGPT has no Google Business Profile data, while Google's own AI Mode and Gemini do. Source: Whitespark, Local Search Ranking Factors 2026
Google's own documentation on AI features is unambiguous, and it is worth quoting directly rather than paraphrasing: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary,”
And, more specifically still, “There's also no special schema.org structured data that you need to add.” That is Google Search Central's own guidance, last updated 10 December 2025 — a primary source stating plainly that the mechanism a large share of “AI SEO” advice sells does not exist as described.
The claim that circulates instead — that adding schema and FAQ markup produces something like a 44% lift in AI citations — traces to a misattribution: the real, publicly findable BrightEdge statistic carrying that number describes something unrelated (how much more often Google's AI Overviews criticise brands compared with ChatGPT), and no BrightEdge study establishing a 44% schema-citation lift could be located.
The Ahrefs controlled test above is the closest thing to real evidence on the question, and it found the opposite of a meaningful lift. None of which makes schema pointless — it does the jobs the local business schema guide describes, which are not this one.
What this means for where local businesses should actually spend effort
Read together, the table above and the debunked schema claim point the same direction: the AI-visibility factors that scored highest with Whitespark's expert panel are almost entirely about being mentioned favourably somewhere other than your own website — a curated list, an industry-relevant third-party domain, a review site — rather than about markup added to pages you control. Three of the top five factors are variations on being cited elsewhere, not on-page technical work.
That is a genuinely different priority list from most “how to rank in ChatGPT” guides, which lead with schema and structured data because that's a task an agency can bill for and a business owner can point to as done. Earning a mention on an industry “best of” list or from a genuine local publication is slower and less billable, and the evidence above says it currently matters more.
This site's own local SEO for AI search guide works through what a business can actually do about that in practice.
Consumer AI usage for local recommendations is real and moving fast, but trust in it is a genuinely split verdict
Self-reported adoption of a genuinely new behaviour tends to run high, because respondents round occasional or curious use up to routine use. The adoption figure below should be read with that in mind — and read alongside the trust figures next to it, because a large share of people who say they've used AI for this do not yet say they trust it.
45%
Grade Cof consumers say they've used ChatGPT or another AI tool for local business recommendations, up from 6% the prior year.
What it does not say A near-eightfold year-over-year jump in self-reported behaviour is the kind of shift that deserves scepticism about the precise number even when the direction is obviously credible. This almost certainly captures a real trend and probably overstates its current size. AI ranked third among sources consumers cited for local recommendations, behind Google (71%) and Facebook.
40% vs 32%
Grade Cof US consumers say they trust AI platforms to provide business recommendations, against 32% who say they do not — a genuine split rather than the runaway confidence some AI-adoption coverage implies. A further 42% say they trust AI “as much as traditional reviews” for local recommendations.
What it does not say BrightLocal's own framing describes this trust level as “surprisingly high” for a new technology; a 40/32 split with the remainder presumably undecided reads, on the numbers themselves, as contested rather than resolved.
82% / 23%
Grade Cof consumers say they read AI-generated review summaries, but only 23% say they'd rely on the summary alone to make a decision without checking star ratings or full reviews; 59% use the summary as a starting point before cross-referencing.
What this means for measuring AI's real influence
Adoption and trust are two different measurements, and the gap between them is the more useful number here. Forty-five per cent having tried AI for a local search says something about exposure; forty per cent trusting the output, against thirty-two who don't, says something closer to actual influence on a decision.
The review-summary figures make the same point from a different angle — the large majority reading AI summaries and the small minority willing to act on one alone describes an assistant used as a filter that narrows the field, with the final decision still routed through the same reviews and ratings it summarised. That is a materially smaller claim than “AI recommendations are replacing search,” and it is the one this data actually supports.
Adoption of AI for search rose through 2026. Trust in it fell over the same period — and the two studies measuring each are not describing the same population
This is a genuinely new section, not a repeat of the adoption figures above, because it answers a different question the industry rarely separates out: not how many people have tried AI for search, but whether the ones who have are becoming more or less confident in it. The two most useful studies on this reach opposite-sounding headline numbers for a specific, checkable reason.
82% to 54%
Grade Bwas the share of US consumers who said AI-powered search was more helpful than traditional search, a year earlier versus in 2026 — a 28-percentage-point fall measured by repeating an identical question to a fresh sample.
What it does not say This measures AI search broadly, not local business recommendation specifically — no equivalently designed year-over-year trust study scoped to local search alone was identified in the sources reviewed on 8 September 2026. It's included because it is the clearest directional trend data available on whether trust in AI search is rising or falling, and the direction is the opposite of what most AI-adoption coverage implies.
3% to 17%
Grade Bwas the share of consumers who rated AI search as actively less helpful than traditional search, in the same year-over-year comparison — nearly a sixfold increase in outright AI scepticism in twelve months.
75%
Grade Bof people who already use AI tools for local search rate their trust in AI-generated local recommendations at 4 or 5 out of 5 — a much higher figure than BrightLocal's 40% trust reading above, because this question was asked only of a self-selected group of existing AI users, not a general population.
What it does not say The gap between this 75% figure and BrightLocal's 40% is not a contradiction — it's a selection-bias demonstration. Asking only people who already use AI whether they trust AI structurally produces a higher number than asking a general population the same question, and the two surveys should never be quoted against each other as though they measured the same thing.
5%
Grade Bof AI users say they proceed directly from an AI answer to a purchase or booking without further checking; 53% say they verify through a separate Google or Bing search first, and 49% visit the business's own website directly before acting.
What this means inside FOUND → TRUSTED → CHOSEN
The verification figures above are, in effect, a direct measurement of the TRUSTED stage happening even after an AI mention. Only one AI user in twenty proceeds straight from an AI answer to a purchase; roughly half re-run the check through Google, and roughly half go to the business's own website before deciding.
An AI assistant naming a business is closer to a FOUND-stage event than a TRUSTED or CHOSEN one — it puts a business in front of a customer, but the customer's own trust check still happens afterward, on the business's website and reviews, exactly as it always did. That is the same argument this site's framework makes about paid search and the local pack, extended to a new channel.
None of these findings suggest AI recommendation shortcuts the trust-building step, and the declining year-over-year helpfulness figures from Fractl suggest the opposite trend if anything — a customer arriving from an AI mention in late 2026 is, on this evidence, more likely to double-check than one arriving a year earlier.
A business whose website or review profile can't survive that follow-up check turns an AI mention into a wasted opportunity rather than a converted one, which is the AI-search version of the failure this site's research keeps finding at the CHOSEN stage generally.
What the evidence covers
Every figure here describes US-based or predominantly-US research, with the partial exception of Yext's 2026 survey, which has a genuine global sample but whose detailed breakdowns (income bands, verification behaviour) are drawn mainly from its US-specific wave.
No published study measuring AI local recommendation behaviour for India or comparable markets was identified in the sources reviewed on 8 September 2026. Given that assistants draw heavily on web, review and directory signals that already skew toward US sources (Yelp, Foursquare, US-weighted web content), there is good reason to expect AI local visibility works differently in markets where WhatsApp and Instagram carry more of the discovery load.
The numbers that did not survive the trace
These appear in current statistics roundups. Each one fails on provenance, and quoting them transfers the failure to you.
40.16%
Do not citeof local business queries trigger Google's AI Overviews.
Traced to Circulates with suspiciously precise decimal formatting, source unclear.
A figure carrying two decimal places usually signals a real underlying dataset — but no page repeating this number links to where it was actually measured, what query set was used, or over what time period. The precision is performing rigour it hasn't shown its work for. A genuine, sourced study on this page's “Google's AI Overviews” section put the comparable average closer to 68% across a small US sample — a different figure entirely, from a source that discloses its method. No substantial published study establishing this specific 40.16% figure was identified in the sources reviewed on 8 September 2026.
60–70%
Do not citeof ChatGPT's local business results come from Foursquare.
Traced to A Spanish SEO consultant's test of 50 prompts across five Spanish cities, published May 2025, found Foursquare-sourced businesses topping roughly 60–70% of results in that specific sample.
The original finding was small, single-country and clearly scoped by its own author. Independent tracing found its qualifiers stripped from the claim within nine days of publication, after which it spread as an unscoped, universal fact about ChatGPT. A large, current US test — 2,880 prompts across 12 metro areas and 12 categories, run in 2026 after OpenAI's July 2026 data-licensing partnership with Yelp — found 0.00% Foursquare citation share on ChatGPT's primary answer surface. Whatever was true of a small Spanish sample in May 2025, it is not a description of how ChatGPT sources local results in September 2026.
44%
Do not citemore AI search citations from adding schema markup and FAQ blocks to a page.
Traced to Widely attributed to BrightEdge, but no BrightEdge study establishing this specific figure could be located. A real, publicly findable BrightEdge statistic does carry the number 44% — for something unrelated: how much more often Google's AI Overviews criticise brands compared with ChatGPT.
This looks like a straightforward misattribution that then hardened into received wisdom through repetition, and it sits directly against the available evidence rather than merely lacking support for it. Google's own Search Central documentation states plainly that “there's also no special schema.org structured data that you need to add” for AI features, and a controlled Ahrefs test of 1,885 pages that added schema found a small decline in AI Overview citations and changes statistically indistinguishable from zero on AI Mode and ChatGPT. A statistic this specific, this widely repeated, and this directly contradicted by both the platform's own documentation and a controlled study is close to the clearest case of a zombie statistic on this entire site.
How this page is made and how to cite it
The selectivity findings were traced to SOCi's published methodology directly. The Google AI Overviews local-query breakdown was traced to Whitespark's disclosed methodology and flagged for its age and small sample.
The ChatGPT sourcing findings distinguish an officially confirmed partnership (Yelp) from a single agency's own disclosed-but-unreplicated testing (the Foursquare disproof), and both are graded accordingly. The schema-markup section rests on Google's own official documentation plus a controlled Ahrefs study, checked directly rather than taken from a secondary summary.
The consumer-adoption and trust sections keep BrightLocal's general-population figures separate from Yext's AI-user-only figures, specifically because conflating the two produces a false picture of how settled consumer trust in AI recommendations actually is. Three zombie entries document figures that circulate in 2026 roundups without surviving a trace to their origin, including one — the schema-markup “44% lift” claim — that is directly contradicted by both official documentation and a controlled study, not merely unsupported.
Last verified: 8 September 2026. Reviewed quarterly at minimum — this category moves fast enough that a stale review here is worse than a stale review elsewhere on the site.
Preferred citation Parashar, S. (2026). AI local search statistics, traced to source. shrutiparashar.com/statistics/ai-local-search/
Questions
- Is it harder to get recommended by AI than to rank on Google?
- For multi-location brands, substantially. SOCi's Local Visibility Index 2026 analysed roughly 350,000 locations across 2,751 multi-location brands using a standardised prompt across six platforms, and found 1.2% were recommended by ChatGPT against 35.9% appearing in Google's local 3-pack — about thirty times more selective. Perplexity (7.4%) and Gemini (11%) are both more generous than ChatGPT but still far more selective than the local pack. The study covers multi-location brands specifically; whether independent local businesses face the same gap is unmeasured.
- Does ranking #1 on Google mean you'll be recommended by AI?
- No. The same SOCi study found fewer than half of Google's local-pack leaders also appeared in AI-generated recommendations for the same query — meaning whatever signals AI assistants weight most heavily diverge meaningfully from what wins a Google ranking.
- Which AI platform is easiest to get recommended by?
- Gemini, on SOCi's 2026 figures — 11% of tested locations were recommended, against 7.4% for Perplexity and 1.2% for ChatGPT. Gemini is also the only one of the three grounded natively in Google's own Maps and Business Profile data (measured at 100% business-data accuracy, against 68% for both ChatGPT and Perplexity), which plausibly explains both its higher recommendation rate and its accuracy advantage — it is drawing on data Google already controls and keeps current.
- Does Google's AI Overviews feature actually affect most local searches?
- It depends heavily on the specific query, more than most coverage of this topic acknowledges. A Whitespark case study (540 queries, 3 US cities, 6 industries, published May 2025) found AI Overviews triggered for only 15% of narrowly local-intent queries like “plumber near me”, against 92% of informational queries in the same trades and 97% of hybrid queries. The 68% average figure quoted elsewhere blends all three intent types together and hides that gap. This study is now over a year old in a fast-moving category, so treat the shape of the finding — informational queries trigger AI Overviews far more than transactional local ones — as more durable than the exact percentages.
- Does adding schema markup actually help a business get cited by AI?
- The evidence says no, despite how often the opposite is claimed. Google's own Search Central documentation states there is “no special schema.org structured data that you need to add” to appear in AI Overviews or AI Mode. A controlled Ahrefs test that added JSON-LD schema to 1,885 pages already receiving heavy AI citation found a small decline in Google AI Overview citations and changes statistically indistinguishable from zero for AI Mode and ChatGPT. The widely repeated claim that schema produces a 44% lift in AI citations traces to a misattributed BrightEdge statistic that appears to describe something else entirely, and no study establishing that specific figure could be located.
- Where does ChatGPT actually get its local business information from?
- As of 2026, primarily Yelp. OpenAI and Yelp confirmed a non-exclusive data-licensing partnership (reported 23 July 2026, with Yelp's own blog confirming the consumer-facing integration on 10 August 2026), covering more than 330 million reviews across 8 million-plus business listings, plus in-chat Reservations and Waitlist for thousands of US and Canadian restaurants. A large current test (2,880 prompts, 12 US metros, 12 categories) found essentially zero citations traced to Foursquare, despite a widely repeated claim that Foursquare supplies 60–70% of ChatGPT's local results — that claim originated from a small, scoped May 2025 study in Spain and does not describe ChatGPT's current sourcing.
- How many consumers actually use AI to find local businesses, and do they trust it?
- 45% say they've used ChatGPT or another AI tool for a local recommendation, according to BrightLocal's 2026 survey (n=1,002 US adults) — up from 6% the year before, a jump large enough that the precise figure should be treated with caution even though the underlying trend is real. Trust is a genuinely split verdict rather than settled: the same survey found 40% of consumers trust AI platforms for recommendations against 32% who don't. A separate Yext survey found 75% trust among people who already use AI regularly for local search — a much higher number, but that's because it only asked existing AI users, not a general population, which structurally inflates the result.
- Is trust in AI search rising or falling?
- Falling, on the clearest year-over-year evidence available, even as raw adoption rises — the two numbers are not the same thing and move independently. A Fractl/Search Engine Land study that repeated an identical question to comparable US samples found the share who rated AI-powered search as more helpful than traditional search fell from 82% a year earlier to 54% in 2026, while the share who rated it as actively less helpful grew from 3% to 17%. This measures AI search broadly rather than local-business recommendation specifically, but it is the clearest trend data available on the direction trust is moving.
- If an AI assistant recommends a business, do customers act on it directly?
- Rarely, without checking first. Yext's 2026 survey found only 5% of AI users proceed straight from an AI answer to a purchase or booking; 53% verify through a separate Google or Bing search, and 49% visit the business's own website before acting. That makes an AI mention closer to a discovery event than a closing one — the customer's own trust check still happens afterwards, on the business's website and reviews, much as it always has.
- Does this data apply outside the United States?
- Not reliably. Every figure on this page describes US-based or predominantly-US research; the Yext survey is the one exception with meaningful non-US respondents, and even there the detailed breakdowns available are US-specific. No published study measuring AI local recommendation behaviour for India or comparable markets was identified in the sources reviewed on 8 September 2026 — and because these assistants draw heavily on US-centric platforms (Yelp, Foursquare, US-weighted web content), there's good reason to expect AI local visibility behaves differently in markets where WhatsApp and Instagram carry more of the discovery load.
Related reading: our research methodology , Local SEO statistics , Local search behaviour statistics , the local SEO for AI search guide .
For the numbers on adjacent topics, see the rest of the archive.
