This study was independently reviewed after publication. Two citation figures were wrong, one finding has been withdrawn, and several claims have been narrowed to match the evidence. The full list of changes is in the Corrections section at the end, and all 86,645 citations behind it are now published so you can check any number yourself.
The Front Door Problem
AI Overviews and AI Mode sit on top of the same search engine. For local businesses, they cite almost opposite sources.
We ran nearly 4,000 local-service searches through each surface, classified all 86,645 citations, and crawled the 26,358 unique pages they pointed to.
3,992 AI Overviews queries · 3,998 AI Mode queries · 86,645 citations · 50 metros · 10 verticals · August 2026
Revised September 9, 2026
This study was independently reviewed after publication. Two citation figures were wrong, one finding was withdrawn, and several claims have been narrowed. The full correction history appears at the end. We have also published all 86,645 citations so you can check the numbers yourself.
What we found
1. AI Overviews appearance rates depend on how they are measured
Our collection recorded an AI Overview on 48.8% of searches. The difference was mostly driven by query intent: 21.8% for local queries and 75.8% for informational queries.
Those figures did not survive a manual validation check. A human browser saw AI Overviews much more often. Treat every appearance rate in this report as a result of our generic, non-personalized collection method—not as a settled rate for Google users.
AI Mode answered every query in the collection.
2. AI Mode’s local citations usually lead back to Google
On local-intent queries, 79.8% of AI Mode citations pointed to Google Maps or Google Business Profile data. Only 12.8% pointed to the business’s own website.
AI Overviews went the other direction: 70.7% of its local-intent citations pointed to the business’s own website, while 3.1% pointed to Google Maps or GBP.
3. AI Overviews leans slightly toward better-marked-up business pages
Among business pages already cited, pages cited by AI Overviews had schema markup, a phone number in the HTML, and concrete numbers in the body copy slightly more often than pages cited by AI Mode. The differences were 5.3, 4.5, and 3.7 percentage points.
That is a difference between two pools of cited pages. It is not evidence that adding those elements causes a citation. This study has no uncited control group.
4. Reputation is similar among businesses already cited
Ratings, review volume, domain age, and Google Business Profile completeness were effectively flat between the two surfaces. AI Mode’s cited domains had more referring domains and backlinks at the median.
Every business in this comparison had already been cited. The data cannot tell us whether reviews or authority helped a business get cited in the first place.
At a glance
| Metric | AI Overviews | AI Mode |
|---|---|---|
| Answered a query in our collection | 48.8%* | 100%* |
| Answered a local “best X in city” query | 21.8%* | 100%* |
| Answered an informational query | 75.8%* | 100%* |
| Cited the business’s own site on local-intent queries | 70.7%† | 12.8%† |
| Cited Google Maps/GBP on local-intent queries | 3.1% | 79.8% |
| Cited business pages with schema markup | 79.3% | 74.0% |
| Cited business pages with a phone number | 78.9% | 74.4% |
| Referring domains, median | 296 | 428 |
| Review count, median | 412 | 450 |
* These are collection results, not settled appearance rates. A manual browser check found AI Overviews much more often.
† “The business’s own site” is a residual classifier category. A citation-weighted audit found that about 11.7% of the remaining records in it were misclassified (95% CI 5.8–22.2%). The Google Maps/GBP figures do not carry that problem because those domains were matched positively.
Explore all 86,645 citations in the raw-data ledger
Finding 1: Our collection saw AI Overviews half the time. A browser saw more.
We sent 3,992 usable local-service queries to AI Overviews and 3,998 to AI Mode. A handful of the original 4,000 combinations per surface dropped out after retries.
AI Overviews returned an answer for 48.8% of the queries. AI Mode answered all 3,998. That second number is close to definitional for a dedicated answer endpoint.
Intent explained most of the AI Overviews difference:
| Query type | AI Overviews | AI Mode |
|---|---|---|
| “Best plumber in Phoenix, AZ”—local intent | 21.8% | 100% |
| “How much does a plumber cost?”—informational intent | 75.8% | 100% |
That looks straightforward until you compare it with a real browser.
Three weeks after collection, we ran 12 queries by hand. Ten were local searches also present in the original collection. A human saw an AI Overview on 5 of those 10. The collector had seen one.
We then put the same queries back through the collector five times each on the day of the manual check. Its appearance rate was 6%, close to the August result and nowhere near the browser result. The difference was between measurement methods, not dates.
AI Mode matched the manual check on all 12 queries.
One query also changed inside the same browser session: an AI Overview appeared, then disappeared when we repeated the query ten minutes later. This surface moves. Any single appearance rate is a snapshot.
What this means for local businesses
- Do not treat “ranking in AI” as one problem. AI Overviews appeared much more often for informational questions in our collection. Those are the questions guides, FAQs, and pricing pages are built to answer; this study did not test whether those formats improve citation rates.
- AI Mode answered every local query in our collection. The important question is what it chooses to cite.
Finding 2: AI Mode’s local answers mostly cite Google itself
We classified all 86,645 citations into ten source categories.
For local-intent queries, the two surfaces pointed in different directions:
| Citation destination | AI Overviews, local intent (n=3,869) | AI Mode, local intent (n=40,883) |
|---|---|---|
| Google Maps/GBP | 3.1% | 79.8% |
| Business’s own site | 70.7% | 12.8% |
| Social | 12.4% | — |
| Other | 16.9% | 7.4% |
For informational queries, AI Mode behaved differently. Its Google Maps share fell to 22.7%, while citations to business websites rose to 42.3%. AI Mode can cite a business website. On local-intent queries, it usually chooses Google’s own listing data instead.
The pattern changes more by vertical than by city
| Vertical | AI Mode Google Maps share, local intent |
|---|---|
| Dentist | 88.0% |
| Auto repair shop | 87.0% |
| Plumber | 85.1% |
| Locksmith | 84.5% |
| Roofer | 82.9% |
| HVAC contractor | 82.9% |
| Pest control | 81.8% |
| Electrician | 76.5% |
| House cleaning | 65.4% |
| Personal injury lawyer | 43.4% |
The metro results stayed within a much tighter range: 71.2% in Pittsburgh at the low end and 84.8% in Birmingham at the high end. Every metro was above 70%.
Vertical was different. Eight of ten categories landed between 76% and 88%. House cleaning fell to 65.4%, and personal injury law to 43.4%. A single 79.8% average should not be treated as the number for every business category.
What this means for local businesses
- Check your vertical before acting on the headline average.
- For “best X in city” searches, the listing behind the AI Mode result may matter more than earning a direct website citation.
- Keep the primary category, service area, hours, and listed website accurate. Category is an established local ranking input. We did not test which profile attributes drive AI Mode selection, so the other fields should be treated as good listing hygiene—not proven AI Mode levers.
- We inferred AI Mode’s use of local-index data from its links. We did not observe its selection mechanism.
Finding 3: AI Overviews leans slightly toward better-marked-up business pages
We crawled the cited pages and checked eight signals. For this comparison, we restricted the data to citations classified as a business’s own website and counted each page once.
Three signals differed by more than two percentage points. Five were flat by that rule.
| Page signal | AI Overviews | AI Mode | Difference |
|---|---|---|---|
| Schema markup | 79.3% | 74.0% | +5.3 |
| Phone number in HTML | 78.9% | 74.4% | +4.5 |
| Concrete numbers in body copy | 72.8% | 69.1% | +3.7 |
| Viewport meta tag | 94.8% | 95.1% | −0.3, flat |
| About page linked | 72.7% | 72.0% | +0.7, flat |
| Credential language | 66.9% | 67.3% | −0.4, flat |
| Named staff mentioned | 25.3% | 24.0% | +1.3, flat |
| HTTPS | 99.5% | 99.6% | −0.1, at ceiling |
The denominators were 8,494 unique AI Overviews pages and 10,445 unique AI Mode pages.
Our first analysis made this finding look stronger. It counted citation instances instead of unique pages and mixed different page categories together. Because AI Mode cites more directories and forums, that made its pages look less mobile-friendly.
Once we compared business page with business page, the mobile difference disappeared: 94.8% for AI Overviews and 95.1% for AI Mode. We withdrew both the finding and the recommendation built on it.
The remaining result still depends on a classifier category with known error. We tested whether contamination was uneven: the proxy was 4.6% for AI Overviews and 5.9% for AI Mode. Removing those pages left the schema difference unchanged at 5.3 points.
See the stats/study in context: Steady Demand Research Index
