Original research · September 2026
We pulled 53,900 Google results pages for 49 kinds of local business across 250 US cities, five per state, in September 2026. Five ways of asking the same question, desktop and mobile, with the browser-rendered features most scrapers miss. This is what Google actually puts on page one, and who wins each part of it.
The one number to take away
The AI Overview appears on 4% of “near me” pages and 75% of specific-need pages. Same categories, same cities, same week. Only the phrasing moved.
The short version
The Local Pack is on almost every local results page no matter how you phrase the search. It is the one thing the wording does not change.
Add a specific need to a “best” query and you get an AI Overview three times out of four. “Near me” gets ads instead, on 44% of pages. “Best” gets Reddit, on 24%.
Half of specific-need AI Overviews contain business cards with call and directions buttons, and on the pages we spot-checked they do not always list the same businesses as the map does.
Up to 85% of local AI Overviews end by asking the searcher something, with an “Ask anything” box underneath. The pack is three clicks away. The conversation is one.
The AI Overview’s most cited sources are the business itself: its own site, its chain’s site, or its Google Business Profile. The profile alone was cited 13,777 times.
Business websites held 58% of “near me” organic results in 2024 and hold 57% now. Directories were 25% and are 24%. What changed is the page around them.
Packs for “best” searches average 4.79 stars against 4.67 for the plain search, and two thirds of their businesses are rated 4.8 or above. Review count barely moves.
Local Services Ads and sponsored pack slots never appeared in 41,650 desktop pages. Live checks found both on most high-value service searches. That is a measurement problem, and we say so.
Section 1
Each of the 49 categories was searched five ways in every city. The Local Pack barely changes. Everything around it does. Learn this gradient once and the rest of the study reads quickly.
Chart 1
Share of desktop results pages containing each feature, by how the search was phrased.
Read across the AI Overview row. 4%, 9%, 27%, 75%. Same categories, same cities, same week. The only thing that moved was the phrasing. A study that only ran “near me” queries would tell you AI Overviews are a rounding error in local search. One that only ran qualified queries would tell you they are everywhere. Both would be right about their sample and wrong about local search.
Ads run the other way. “Near me” is Google’s ad inventory, with an ad on 44% of pages and 1.2 per page on average. Add a specific need and ads drop to 16%. So the space ads used to hold at the top of a long-tail query is now the space the AI Overview fills. Nobody bids on “best plumbers in Cheyenne for water heater repair”. Google answers it instead.
The forum block follows “best”. It is on 24% of “best” pages and 29% of qualified pages, against under 3% for the plain search. People asking “best” want other people’s opinions, and Google gives them Reddit.
Page furniture, averages per page
| “near me” | “[category] [city]” | “best [category] in [city]” | + specific need | |
|---|---|---|---|---|
| Text ads | 1.19 | 0.40 | 0.47 | 0.40 |
| People Also Ask questions | 3.5 | 3.8 | 4.0 | 4.0 |
| Organic results | 8.8 | 8.9 | 8.7 | 8.6 |
Section 2
Rating is the lever, not review volume. The word “best” moves the pack’s average rating from 4.67 to 4.79, and that small shift hides a large change in the distribution underneath.
Chart 2
Local Pack composition by query type. Hotels excluded: Google serves a booking module rather than a standard pack.
| “near me” | “[category] [city]” | “best [category] in [city]” | + specific need | |
|---|---|---|---|---|
| Pack entries analysed | 35,783 | 33,541 | 34,022 | 13,328 |
| Average rating | 4.67 | 4.67 | 4.79 | 4.78 |
| Rated 4.8 or higher | 52.6% | 53.1% | 67.3% | 65.6% |
| Rated 4.5 or higher | 73.1% | 74.2% | 85.7% | 83.4% |
| Average review count | 240 | 233 | 251 | 261 |
| 100 or more reviews | 48.0% | 44.1% | 47.1% | 47.8% |
| Years in business, mean | 17.0 | 18.8 | 18.2 | 18.5 |
| Has a website link | 70.2% | 71.1% | 72.0% | 73.8% |
Chart 3
Rating distribution of Local Pack entries on “[category] [city]”.
Chart 15
Share of “[category] [city]” pages showing a single business Knowledge Panel instead of a pack. Pack and panel never appear together.
The click with nowhere to go
30%Three in ten pack businesses have no website link. That is 30% of the businesses Google is already showing with no site to send the click to.
The word “best” moves the pack’s average rating from 4.67 to 4.79. That sounds small until you look at the distribution: the share of pack businesses rated 4.8 or above jumps from 53% to 67%, and the share under 4.5 drops from 26% to 14%. Google is using “best” as a quality filter on the map, and a 4.6-star business that shows for “plumbers Huntsville” has a much thinner chance of showing for “best plumbers in Huntsville”.
Review count barely changes. Rating is the lever, not volume.
Metro packs, in cities over 1M, average 321 reviews on the plain search. Small cities under 100K average 189. Ratings are flat across tiers at 4.65 to 4.70. Bigger markets mean more reviews to compete against, not higher standards.
Hotels are the exception to everything above. Google serves its hotel booking module rather than a standard pack, so hotel pages are excluded from pack composition figures.
All pack figures describe the three organic entries. Live desktop searches for plumbers, lawyers and HVAC contractors in September 2026 showed a fourth, sponsored entry on every page checked. The collection method did not return it, so paid pack placements are not measured here.
Section 3
Category matters as much as phrasing. Google will not generate an answer for where to eat, but it will for who should build your home.
Chart 4
AI Overview rate by business category, sorted by the “best [category] in [city]” rate. The dark tick marks the same category’s rate on the plain “[category] [city]” search.
Two patterns. AI Overviews cluster on considered, high-ticket or professional decisions: who builds your house, who does your taxes, who manages your money. Google will not generate an answer for where to eat, with restaurants at 1.2%, bars at 2% and cafes at 0.4%, but it will for who should build your home, at 80%.
Where the AI Overview is low, Reddit is high. Dentists get an AI Overview on 3.6% of “best” pages and a forum block on 62%. Vets: 10.8% and 61.2%. Orthodontists, pet groomers, auto repair and gyms all follow the same shape. For anything involving your body, your pet or your car, Google hands you other people’s experiences rather than its own summary.
Adding a need to the end of a “best” query took the AI Overview rate to 75%. But we ran two kinds of need for each category, and they behave differently. An attribute qualifier describes the business: “with outdoor seating”, “open 24 hours”, “accepting new patients”. A service qualifier describes the job: “for water heater repair”, “for dental implants”, “for immigration”.
The overall AI rate is similar. What differs is what the AI Overview contains and what sits next to it. Service queries get business cards inside the answer more often, 54% against 43%, and pull in a forum block twice as often, 39% against 19%. Attribute queries slightly suppress the pack, consistent with Google not always finding businesses that match the filter.
Per category the split can be dramatic. Plumbers: 98% AI Overview for “for water heater repair”, 32% for “open 24 hours”. Locksmiths: 80% against 12%. Beauty salons run the other way: 34% for the service query, 100% for “with online booking”. Which kind of specificity triggers the AI depends on what people need to know before choosing. See any category in the explorer.
Chart 5
Feature presence on qualified queries, by kind of qualifier.
Section 4
We parsed 8,540 desktop AI Overviews. They are not paragraphs of advice. They are a second Local Pack with a conversation attached.
Chart 6
Composition of parsed desktop AI Overviews, by query type.
| “near me” | “[category] [city]” | “best [category] in [city]” | + specific need | |
|---|---|---|---|---|
| AI Overviews parsed | 492 | 1,052 | 3,305 | 3,691 |
| Contains business cards | 21.3% | 44.6% | 59.0% | 63.7% |
| Business cards per answer, mean | 0.7 | 1.7 | 2.4 | 2.3 |
| Ends with follow-up prompts | 71.5% | 82.4% | 78.0% | 84.6% |
| Citations per answer, deduplicated | 8.9 | 10.4 | 11.3 | 10.0 |
| Text length, characters | 379 | 601 | 760 | 798 |
The answer asks a question back
85%Between 72% and 85% of local AI Overviews end with an invitation to keep going, followed by questions about budget, neighbourhood or the specific job, with an “Ask anything” box directly below. That is Google steering the searcher off the results page and into AI Mode. The pack is three clicks away. The conversation is one.
On “best” and qualified queries, six in ten AI Overviews contain business cards: a business name, a one-line description, a phone number, a rating, and the Call, Directions and Website buttons from Google Business Profile. Two or three per answer. This is a Local Pack wrapped in generated text, and on the pages we spot-checked it does not always list the same businesses as the map pack below it.
Figure B
Stylised. Labels mark the parts we detected and counted.
Business cards 2.3 per answer. Name, description, phone, rating, and the three GBP buttons.
Chart 7
Share of deduplicated AI Overview citations by source type. The three orange segments are the business’s own footprint and are grouped together.
The business itself: “near me” 45% · “[category] [city]” 61% · “best [category] in [city]” 60% · + specific need 60%
Add the first three segments. On “[category] [city]” searches, 61% of what the AI Overview cites is the business itself: its site, its parent brand’s site, or its Google Business Profile. On qualified queries it is 60%. Directories of all kinds get 16 to 24%. Reddit, which sits at position 3 on half of “best” pages, gets 4 to 6% of AI citations.
The single most cited domain is google.com, 13,777 times, and every one of those is a Google Business Profile viewer link. The AI Overview is built on the same profile data as the map pack.
Chart 7b
Share of AI Overview citations pointing at a domain that also ranks in the organic top 10 on the same page.
Chart 8
Citation count by domain, with the share of those citations that also rank in the organic top 10 on the same page.
| Domain | Citations | Pages | Also on page 1 |
|---|---|---|---|
| google.com (Google Business Profile) | 13,777 | 840 | 0% |
| yelp.com | 8,184 | 3,290 | 97% |
| reddit.com | 3,628 | 1,820 | 86% |
| facebook.com | 2,207 | 1,077 | 76% |
| youtube.com | 1,143 | 577 | 0% |
| instagram.com | 628 | 307 | 46% |
| health.usnews.com | 616 | 245 | 95% |
| houzz.com | 595 | 269 | 95% |
| mapquest.com | 524 | 249 | 69% |
| angi.com | 469 | 236 | 82% |
| en.wikipedia.org | 465 | 303 | 90% |
| zillow.com | 455 | 223 | 98% |
| tripadvisor.com | 418 | 225 | 92% |
| expertise.com | 415 | 205 | 59% |
| locations.oreillyauto.com | 413 | 155 | 100% |
Section 5
Every organic result in the top 10 was classified into one of 14 site types: 72,718 distinct domains, the top 520 by hand and the rest by a language model with the page title as context.
Chart 9
Organic top-10 composition by site type. Colours match chart 7, so the same type reads the same in both.
Chart 9b
Organic top-10 composition on “[category] near me”, 2024 study against 2026 study.
Business websites, local plus chain, held 58% of “near me” results in 2024. They hold 57% now. Directories were 25% then and 24% now. Organic did not get taken over by directories, and local businesses did not lose their share of the links. The page got taller and the links moved down.
Independent local businesses take 40% of top-10 results on “[category] [city]”, their best showing of any query type, and 40.5% of the top three positions. On “near me” chains take a bigger share, 26%, because store locators rank well for it. As queries get more specific, chains fall away, 26% to 21% to 16% to 17%, and listicles and forums take the space.
By city size the local-business share is flat: 38% in metros, 40.5% in mid-sized cities, 40% in small cities. Directories are slightly stronger in small cities, 13.3% against 9.4% in metros, where there are fewer businesses with websites to rank.
Chart 10
Share of results pages carrying each domain in the organic top 10, with its mean position.
| Domain | “near me” | “[category] [city]” | “best [category] in [city]” | + specific need |
|---|---|---|---|---|
| yelp.com | 84.7% | 76.8% | 86.7% | 87.6% |
| reddit.com | 9.4% | 24.9% | 49.2% | 53.1% |
| facebook.com | 23.7% | 34.4% | 34.7% | 32.0% |
| tripadvisor.com | 11.1% | 11.2% | 12.9% | 7.5% |
| thumbtack.com | 11.5% | 5.4% | 8.0% | 7.6% |
| bbb.org | 9.8% | 7.5% | ||
| angi.com | 7.5% | 5.1% | 8.5% | 4.5% |
| instagram.com | 8.0% | 5.9% | 7.1% | |
| houzz.com | 5.2% | 5.7% | 5.7% | 5.5% |
Section 6
We flagged a pack business as a chain when the same name appeared in five or more of the 250 cities.
Auto parts is the most chain-dominated pack in the study: three national brands in 138 to 230 of the 250 cities. Pharmacies, grocery, banks and gyms follow. For home services, professional services and health, the pack is overwhelmingly independent businesses.
Chart 11
Chain names by number of cities where they appear in a pack.
Section 7
Ads follow lead value, and that part is clean. The paid surface we could not capture and the AI Overviews a standard scraper cannot see are the same problem: what your collection method renders decides what you think is on the page.
Chart 12
Share of “[category] near me” pages carrying at least one text ad.
Ads follow lead value. Marketing agencies, financial advisors and lawyers have an ad on eight in ten “near me” pages. Restaurants, bars and cafes have none. The most frequent advertiser domains are aggregators and lead-generation sites, Thumbtack, NerdWallet, Zocdoc, Expedia and Forbes advisor content among them, alongside a few national brands.
The limitation we are reporting, not a finding
Verification searches in September 2026 showed the LSA unit on “plumbers Houston” and “hvac contractors Phoenix”, two sponsored cards above the pack with Get quote, Book and Get phone number buttons, and not on “lawyers Chicago”. The same checks showed a sponsored fourth pack entry on all three, and a newer “What people are saying” block of social posts on the lawyers page. So LSAs are present on live high-value service searches and absent from our data. That is a limitation of the collection method, not a fact about Google, and we report it as one rather than claiming zero penetration.
The measurement gap
The standard scraper also returned zero forum blocks and zero Knowledge Panels on the mobile run. If a study reports single-digit AI Overview rates for local search and does not say how it rendered the page, that is the reason. Google defers about half of local AI Overviews to a second request; on first fetch, 40 to 50% of those came back empty, and a fresh request recovers most. After retries, 2 pages in 53,900 have an AI Overview we could not read. On long queries the rendering browser sometimes captured results for a partial query, on 11% of qualified queries at the first attempt; every page was validated against the category keyword and the city and re-fetched if it failed. Nine pages that never passed are excluded.
Section 8
We ran the “[category] [city]” query again in mobile mode. The mobile collection method cannot render ads, forum blocks or Knowledge Panels, so the comparison covers the pack, organic and the AI Overview only.
The pack is the same size on both: three organic businesses. The fourth entry that mobile mode returned on 18% of packs is most likely the sponsored slot desktop mode discarded, but the mobile data does not label it, so we cannot confirm that and do not count it. The AI Overview appeared slightly less often on mobile, 6.3% against 8.6%, on the same queries in the same cities.
Chart 13
Desktop against mobile on “[category] [city]”, same cities and categories.
| Desktop | Mobile | |
|---|---|---|
| Local Pack present | 91.4% | 90.9% |
| Packs with three entries | 100% | 82% |
| Packs with an unlabelled fourth entry | 0% | 18% |
| AI Overview | 8.6% | 6.3% |
| Organic results, mean | 8.9 | 9.3 |
Section 9
The page you are competing on now has four surfaces that matter, and they are fed by the same three things: your profile, your rating, and your own site.
On nine pages in ten. Your rating is the lever, not your review count: “best” queries filter on 4.8 and above. Thirty percent of pack businesses have no website link, which is a click Google cannot send anywhere.
It cites your own site and your Google Business Profile before it cites anyone else. Six in ten citations on a plain search are the business’s own footprint. The business cards inside it are built from the profile. If the profile is thin, the card is thin.
It has not changed. Independent sites still take 40% of the top ten on “[category] [city]”. Yelp is at position 3 whatever you do. Reddit is at position 3 on “best”.
For high-value leads, and mostly bought by aggregators. Worth knowing that on “near me” in a professional category you are bidding against lead-generation sites, not your neighbours.
For agencies and multi-location brands managing hundreds of locations, the operational read is simpler. The profile data feeds the pack, the AI cards and the AI citations, so listing accuracy and review rating at every location are now the inputs to three surfaces, not one.
Interactive
Every category’s own view, across all four ways of asking. The URL updates as you pick, so you can send a colleague straight to the view you are looking at.
Home and trades
Which kind of specific need triggers the AI Overview
Synup, September 2026. Desktop. 250 pages per category for the first three query types, 100 for the qualified pair, split evenly between a service and an attribute qualifier. Blank values mean the measure does not apply: hotels return a booking module rather than a pack, and banks rarely show ratings.
Methodology
Everything here is open on the page rather than behind a disclosure, because the limitations are part of the finding.
49 Google Business Profile categories, the same set as the 2024 study, each searched five ways: “[category] near me”, “[category] [city]”, “best [category] in [city]”, the same with a qualifier, and “[category] [city]” again in mobile mode. Two qualifiers per category, one drawn from Google’s attribute taxonomy and one from its service lists, run in the largest city per state.
The five most populous cities in each of the 50 states, 250 in total, from New York, NY at 8.26M to Sitka, AK at 8.5K. Reported by tier: metro over 1M, mid 100K to 1M, small under 100K. The design over-represents small cities relative to population, deliberately: this is a study of how Google composes pages, not of where searches happen.
53,900 results pages, collected in September 2026, via a browser-rendered SERP API with city-level geo-targeting on every request and 20 results per page. Deferred AI Overviews were fetched within their two-minute window. Every page was validated against the category keyword and target city and re-fetched on failure. Nine pages were excluded.
Feature presence from the structured response. Local Pack fields cleaned from the provider’s mislabelled output: rating, review count, category label and years in business. AI Overview citations deduplicated and normalised. Business cards detected from the flattened Call, Directions and Website markup.
72,718 distinct domains from the organic top 10 and AI citations. The 520 most frequent, 52.6% of result rows, were classified by hand into 14 site types. The rest were classified by a language model given the domain, its frequency and a sample page title. A 300-domain random sample of the model-classified set was checked by hand: 298 of 300 agreed, 99.3%. Both misses were third-party pages read as the business itself.
A pack business is counted as a chain if the same business name appears in five or more of the 250 cities. Name matching is exact after normalisation, so a location using a different trading name is counted as independent.
Chart 14
Pack, AI Overview, forum and ad rates by city tier. Each cell is pack / AIO / forums / ads.
| City tier | “near me” | “[category] [city]” | “best [category] in [city]” | + specific need |
|---|---|---|---|---|
| Metro, over 1M | 97.7 / 3.6 / 0 / 39.5 | 95.5 / 3.6 / 5 / 24.3 | 96.6 / 30.9 / 31.8 / 23.6 | 90.5 / 75.7 / 29.1 / 18.5 |
| Mid, 100K to 1M | 97.4 / 4.2 / 0.2 / 42.9 | 92.7 / 7.5 / 3 / 16.7 | 93.7 / 25.4 / 26.3 / 20.3 | 92.3 / 75.1 / 28.9 / 15.8 |
| Small, under 100K | 97.4 / 3.7 / 0.2 / 44.9 | 88.8 / 11 / 2.3 / 14.4 | 90.4 / 29.4 / 19.4 / 16.5 | 89 / 76.9 / 28.6 / 17.4 |
Download and cite
Reuse any chart with attribution. The aggregate files below are the same numbers the charts are drawn from.
CSV
All 49 categoriesPack, ad, Knowledge Panel, AI Overview, forum and business-card rates per category, plus the qualifier split and pack rating averages.
Download categories.csvCSV
Page-one features by query typeChart 1 as a table: every feature’s presence rate across the four ways of asking.
Download features.csvCSV
AI Overview citations by source typeChart 7 as a table: citation share by source type and query type.
Download citations.csvSuggested citation
Synup, “What’s Ranking in Local Search: 2026”, September 2026. 53,900 Google results pages across 49 business categories and 250 US cities. synup.com/research/local-serp-studyLink to this study
Every section and every chart has its own anchor, so you can point at the exact number you are writing about.
Synup research, September 2026. Our 2024 study asked which websites rank. This one asks where a local business actually gets found. Back to the top