Google's AI Overview names 3 to 4 businesses per local search. Two in three of them aren't in the local pack.
We ran 12,250 "best [category] in [city]" searches across 250 US cities and 49 business categories in the week of September 22, 2026, and compared what the AI Overview recommended with what the local pack showed on the same page. This is what came back.
One results page for "best accounting firms in Huntsville". The local pack shows three businesses. The AI Overview recommends five. All three pack businesses appear in the AI Overview. Two AI Overview businesses, Brand Blackwell & Company and Warren Averett, do not appear in the pack.
The short version
- 31%a floor · 3,836 of 12,250 searches
31% of "best [category] in [city]" searches showed an AI Overview. That's 3,836 out of 12,250, and it's a floor, since scrapers miss some of them.
- 3to4a floor · 3.9 on card answers, 3.4 on text
The AI Overview recommends 3 to 4 businesses per search. On answers built from business cards it averaged 3.9. On plain-text answers, 3.4.
- 67%measured on 714 searches
67% of those businesses are not in the 3-pack. Measured on 714 searches where every pack business had a street address we could match against.
- 64%of pack businesses
The pack still gets in. 64% of pack businesses were recommended by the AI Overview, and on 88% of searches at least one of the three made it. Only 12% of AI Overviews skipped the pack entirely.
- 80%20% two weeks earlier
80% of what the AI Overview cites is Google itself. Business entity links, not web pages. Two weeks earlier that figure was 20%.
Overlap
The AI Overview is a wider pack, not a different one
The simplest way to read the numbers: Google's AI Overview starts from roughly the same businesses as the local pack and then keeps going. It recommended all three pack businesses on 35% of searches and at least one on 88%. Then it added an average of two or three more that the pack had no room for.
So if you're in the pack, you're probably in the AI Overview. The reverse isn't true. Most of what the AI Overview recommends never appeared in the pack at all, which means pack rank on its own tells you less than half the story of who Google is putting in front of a searcher.
| None | 12% |
|---|---|
| One | 20% |
| Two | 33% |
| All three | 35% |
There's a third layer under the visible answer. The sources panel lists the businesses Google consulted while writing it, and that list runs to 15 or 20 entities on a typical search. Of the pack businesses that weren't recommended, most (26% of all pack entries) were in that consulted list. Only 10% of pack businesses were absent from the AI Overview altogether.
Share of all pack businesses
Disagreement
Where the AI Overview and the pack disagree most
Two things drive how far apart the two get: how big the city is, and what kind of decision the search is about.
City size. In the eight metro cities (over 1M people), the AI Overview recommended only 46% of pack businesses and skipped the pack entirely on 35% of searches. In small cities (under 100K), it recommended 69% and skipped the pack on 5%. Mid-size cities sat in between at 63% and 14%. More local inventory gives the AI Overview more room to disagree, and in a big city it uses it. The metro figures rest on 154 searches, so treat the exact percentage loosely; the direction across all three tiers is clean.
| Tier | Pack businesses recommended | Searches where it skipped the pack |
|---|---|---|
| Metro, over 1M, 154 searches | 46% | 35% |
| Mid-size, 100K to 1M, 1,765 searches | 63% | 14% |
| Small, under 100K, 943 searches | 69% | 5% |
Category. Where the search is about trust and money, the AI Overview goes its own way. For hospitals it recommended 41% of pack businesses, and on 30% of hospital searches it recommended none of them. Banks came in at 47%, real estate agencies at 48%, financial advisors at 56%, lawyers at 57%.
Where the search is everyday local trade, the two agree. Bars and cafes sat above 90%. Dentists and car dealers were above 85%. For most of these categories the AI Overview never skipped the pack on a single search. Run a restaurant and rank in the pack, and you're almost certainly in the AI Overview too. Run a bank branch, and being in the pack tells you very little.
| Category | Searches with both | Pack businesses the AI Overview recommended | Searches where it skipped the pack |
|---|---|---|---|
| Hospital | 107 | 41% | 30% |
| Bank | 155 | 47% | 25% |
| Real estate agency | 95 | 48% | 25% |
| Financial advisor | 107 | 56% | 16% |
| Lawyer | 97 | 57% | 27% |
| Accounting firm | 147 | 69% | 9% |
| HVAC contractor | 110 | 74% | 2% |
| Auto repair shop | 28 | 81% | 0% |
| Dentist | 23 | 88% | 0% |
| Cafe | 16 | 94% | 0% |
| Bar | 14 | 95% | 0% |
All 48 categories with a local pack are in the appendix CSV (hotels excluded, see Limitations). Categories with under 25 searches are shown with their counts on purpose. The direction is consistent, but a percentage for bars rests on 14 searches.
Notice the first column. The same query template triggered an AI Overview 163 times for home builders and 14 times for bars, out of 250 cities each. Broadly, the categories where the AI Overview disagrees with the pack most are also the categories where it shows up most. That's the combination that matters: the pack is a weaker proxy for AI visibility exactly where the AI Overview is most likely to be on the page.
Citations
What the AI Overview cites, and what changed in two weeks
We'd run the same 12,250 searches in the week of September 8. Between that run and the week of September 22, the citation layer changed shape.
On the earlier run, 20% of AI Overview citations were links to Google's own business entities and 80% were external web pages. Two weeks later, that had flipped: 80% Google entities, 20% external. External citations per AI Overview fell from about 9 to about 4.
Share of AI Overview citations. Week of September 8: 20% Google business entities, 80% external web pages. Week of September 22: 80% Google business entities, 20% external web pages.
The external mix itself barely moved. Businesses' own websites were the most common outside source both times (50%, then 46% of external citations). Directories were next (29%, then 26%). Reddit was steady at just under 5% of external citations. Google didn't swap one source for another; it added its own entity links on top and cited fewer outside pages per answer.
| Week of September 8 | Week of September 22 | |
|---|---|---|
| Share of external citations | ||
| Business's own website | 50% | 46% |
| Directories | 29% | 26% |
| 4.8% | 4.7% | |
| Share of AI Overviews citing the source | ||
| Yelp | 47% | 35% |
| 23% | 17% | |
For any single source, that shows up as lost presence. AI Overviews citing Reddit fell from 23% to 17% of answers. AI Overviews citing Yelp fell from 47% to 35%. The Reddit story that's been running in AI search coverage this month is real, and it isn't specific to Reddit. Every external source lost ground to Google itself.
Yelp remains the single largest external domain by count. When the AI Overview reaches outside Google for a local answer, the business's own site is where it goes most often.
What this means for how you report on local visibility
If you report pack rank to a client as their local visibility number, you're now reporting on a third of what Google shows. In metro markets and in high-consideration categories it's less than that.
A few things worth building into local reporting now:
Track the AI Overview as its own surface. Whether the business is recommended, whether it's in the consulted list, or whether it's absent are three different states, and the pack predicts them poorly in the markets that matter most.
Stop identifying businesses by website domain. Google's move to opaque /goto links in August means the pack's website link no longer resolves to a domain, and the AI Overview's citations now mostly point at Google entities rather than the business's site. We matched businesses by name and street address for this study. Any tracker still keyed on domain will report gaps that aren't there.
Watch the citation share, but read it correctly. A falling citation share for a source can mean Google stopped citing external pages generally, which is what happened here. Compare it against the total external citation count before drawing conclusions about any one site.
Check the business's own site. It remains the most-cited external source for local answers. Location pages that state services, hours and address plainly are what the AI Overview is reading.
Methodology
- Queries.
- "best [keyword] in [city]" for 49 business categories across 250 US cities (the five most populous in each state), 12,250 searches, desktop, US English, collected in the week of September 22, 2026 through the ScrapingDog SERP API with city-level location targeting. An earlier run of the same queries in the week of September 8 is used only for the citation comparison.
- Identifying recommended businesses.
- The AI Overview's business cards arrive from the API with the name stripped, so we recovered identity three ways: Google entity links embedded in the answer text, table rows, and the AI-written address bullets under each card, matched to the local pack by street address (with a rating-plus-review-count fallback where the pack showed tenure instead of an address). Plain-text answers went through a name-extraction pass using Claude Haiku. Manual checks: 25 of 25 text matches and 20 of 20 card matches confirmed correct.
- Recommended vs consulted.
- "Recommended" means the business appears in the visible answer: a card, a table row, or named in the text. "Consulted" means it appears only in the sources panel. The two are reported separately throughout.
- Citation classification.
- Every cited domain was classified into a type (business's own site, multi-location brand, directory, review aggregator, forum, social, and so on) using a hand dictionary for the most frequent domains and Claude Haiku for the long tail, with a human-audited sample.
Limitations
- AI Overview detection is a floor.
- Third-party benchmarks put this provider's AI Overview capture below the best in class, and our retry logic closes only part of that gap. One of our five manual spot checks showed an AI Overview live where the API had returned none. The 31% presence figure should be read as "at least".
- Pack capture gap.
- On 6% of non-hotel searches the API returned no local pack where a live search showed one. Those searches are excluded from every pack-based statistic. Hotels are excluded entirely, since Google shows a booking module in place of a pack.
- Stripped names.
- Some card headers and heading-style business names are dropped by the API and can't be recovered. The "businesses recommended per AI Overview" figures are floors for the same reason.
- Two points, not a trend.
- The week-of-September-8 and week-of-September-22 runs are two snapshots. They show a change; they don't show its shape over time.
- Wrong-city leakage.
- 4.7% of pack entries were businesses in a same-named city in another state (Bowling Green, Ohio for a Kentucky search). The AI Overview did the same on occasion. We report it as observed.
Appendix
Downloadable CSV: all 48 categories with a local pack with search counts, AI Overview presence, pack businesses recommended, and searches where the pack was skipped. City-tier and citation-type tables included.
Download the CSVaio-vs-pack-categories.csv · 48 categories
Synup Research. "Google's AI Overview names 3 to 4 businesses per local search. Two in three of them aren't in the local pack." 12,250 searches, week of September 22, 2026. Published September 28, 2026. https://www.synup.com/research/ai-overview-vs-local-pack
Use the findings in your local SEO report: watch Niladri explain the study and download an editable visibility worksheet.