The Front Door Problem
AI Overviews and AI Mode both sit on top of Google Search — but they cite local businesses in almost opposite ways. We ran 8,000 local-service searches through both surfaces and crawled every citation to find out who actually gets the click.
What we found
-
01
AI Overviews is selective. AI Mode is not.AI Overviews surfaces for under half of local-business searches — and mostly for informational ones. AI Mode answers every single query we sent it.
-
02
AI Mode’s “front door” often leads back to Google, not the business.For “near me”-style searches, 8 in 10 of AI Mode’s citations point to Google’s own Maps data — not the business’s website.
-
03
When AI Overviews does cite a business, it rewards technical polish.Schema markup, mobile-friendliness, richer content, phone numbers on the page, social presence — AI Overviews consistently favors better-built sites. AI Mode barely notices.
-
04
Neither surface cares much about reputation.Star rating, review volume, review sentiment, domain age, backlink authority — flat or mixed across both. Being well-optimized beats being well-reviewed.
AI Overviews is a gate. AI Mode is a door that’s always open.
Across the same 3,990 local-service queries sent to each surface, AI Overviews generated an answer 48.8% of the time. AI Mode generated one 100% of the time — every query, every metro, every vertical.
The gap is almost entirely about intent. When someone asks a factual, research-style question, AI Overviews shows up more than three-quarters of the time. When someone asks the exact question a local business most wants to answer — “best X near me” — AI Overviews mostly steps aside and lets the classic local pack handle it. AI Mode doesn’t make that distinction at all.
- Don’t assume “ranking well in AI” is one problem — it’s two. Winning AI Overviews citations is mostly an informational-content game (guides, FAQs, pricing pages).
- AI Mode can’t be avoided by having thin local-intent content — it answers regardless. What it cites for those queries is the real battleground (see Finding 02).
AI Mode’s local answers mostly cite Google itself.
We classified all 86,645 citation references into eight source categories. For local-intent queries, AI Mode’s citations are dominated by google_maps_gbp — its own Maps/local-pack redirect links — rather than the business’s own site.
Flip to informational queries and the picture reverses: AI Mode’s self-citation drops from 79.8% to 22.7%, and its citations to actual business sites climb to 47.5%. AI Mode isn’t incapable of citing real sites — it just defaults to Google’s own product for exactly the queries local businesses care most about.
- For “near me” searches specifically, the fight isn’t for an AI citation — it’s for your Google Business Profile. That’s the record AI Mode is actually reading.
- An immaculate GBP (photos, hours, categories, Q&A) may matter more than site-level SEO when the query is local-intent and the surface is AI Mode.
AI Overviews rewards a well-built site. AI Mode barely notices.
We crawled 25,898 cited pages and checked six independent signals of technical and content quality. On every one that showed any spread at all, AI Overviews cited measurably better-built pages than AI Mode.
The gaps aren’t huge on any single signal — typically 5 points — but they run in the same direction every time, across six independent, separately-collected measurements. That consistency is the finding: AI Overviews behaves like a system that’s still reading and weighing the open web. AI Mode, especially for local queries, is drawing more from Google’s own pre-packaged local data (Finding 02), so page-level polish has less to grab onto.
- If your goal is AI Overviews citations: ship LocalBusiness/FAQ schema, make sure the site is genuinely mobile-responsive, and put real numbers (pricing, hours, years in business) in the body text, not just in images.
- A visible phone number on the page itself — not just in a click-to-call button buried in a header — correlates with citation for local queries specifically.
Reputation and authority barely differentiate the two.
We also checked five signals more associated with classic SEO authority and trust: star rating, review count, review sentiment, domain age, and backlink profile. None of them showed the consistent gap the technical signals did.
| Signal | AI Overviews | AI Mode | Read |
|---|---|---|---|
| Star rating (avg) | 4.77 | 4.75 | Flat |
| Review count (avg) | 1,129 | 1,266 | Flat |
| Review sentiment (−1 to 1) | 0.71 | 0.71 | Flat |
| Domain age (median, yrs) | 21.5 | 21.5 | Flat |
| GBP completeness | ≥98% | ≥98% | Flat (ceiling) |
| Referring domains (median) | 296 | 428 | Mixed — AI Mode higher |
| Backlinks (median) | 1,562 | 2,824 | Mixed — AI Mode higher |
Businesses cited by either surface already clear a high reputation bar — roughly 4.75 stars, over a thousand reviews on average, near-fully-built Google Business Profiles. That floor just isn’t where the two surfaces diverge. If anything, on backlink authority AI Mode’s citations skew slightly higher, cutting against a simple “AI Overviews rewards quality more” story — the differentiation is specifically about on-page technical execution (Finding 03), not overall domain authority.
- A stronger backlink profile or a few more reviews is unlikely to move you from “not cited” to “cited” on its own — both surfaces seem to already treat reputation as a pass/fail gate, not a ranking lever.
- Budget accordingly: technical/content fixes (Finding 03) look like the higher-leverage investment than review-generation or backlink campaigns, specifically for AI citation — not for search generally.
When AI cites a site, is it even the right one?
Cross-checking cited domains against Google’s own Business Profile records for the same business, 95.5% of business-site citations pointed to the exact domain Google itself lists as official.
High overall accuracy — but the 4.5% minority is worth a second look for any business auditing its own AI presence: it means citation errors are real, if uncommon, and not something to assume away.
You’re not optimizing for one AI. You’re running two different campaigns — a content campaign for AI Overviews and a Google Business Profile campaign for AI Mode — and most local businesses are only funding one of them.
Every signal in this study points the same direction. AI Overviews behaves like a search engine: it reads your site, and it visibly rewards the ones that are technically sound and content-rich. AI Mode, for the exact searches that drive phone calls — “best plumber near me” — mostly isn’t reading your site at all. It’s reading your Google Business Profile and handing the answer back to Google’s own Maps data. Neither surface cares much whether you have 200 reviews or 2,000. Both already assume you cleared that bar just to be in the conversation.
- Add LocalBusiness + FAQ schema markup — AI Overviews cited schema-marked pages 5+ points more often.
- Put real numbers in the body text: prices, response times, years in business — not locked in images or PDFs.
- Confirm the site is genuinely mobile-responsive, not just “not broken” on a phone.
- Write the pages that answer the question directly — pricing guides, “how to choose a …” content — since that’s where 76% of AI Overviews’ local answers surface.
- Audit your Google Business Profile like it’s your homepage — because for 8 in 10 local AI Mode answers, it functionally is.
- Keep categories, hours, photos, and Q&A complete and current — this is the record AI Mode is actually reading.
- Don’t expect a website redesign to move AI Mode’s local citations — it’s largely not looking at your site for these queries.
- Check that your GBP-listed website matches what you’d want cited — 4.5% of citations in this study pointed to the wrong domain entirely.
The businesses that win here won’t be the ones with the best reviews. They’ll be the ones who noticed these are two different games before their competitors did.
Methodology
Collection
Queries were built from the same 50-metro × 10-vertical scaffold as our companion Citation Ledger study, crossed with 8 query templates (4 local-intent, 4 informational-intent) — 4,000 combinations per surface. AI Overviews data was collected via DataForSEO’s Google Organic Live Advanced endpoint with async AI Overview loading; AI Mode via DataForSEO’s dedicated Google AI Mode Live Advanced endpoint. 3,992 and 3,998 queries respectively returned usable data after resumable retries on transient server errors (≥99.7% completion on both).
Classification
All 86,645 citation references were classified into 8 source categories (business’s own site, Google Maps/GBP, local/general directories, social platforms, review platforms, news media, industry-vendor content, government/association) using a domain taxonomy audited against this dataset’s own highest-frequency defaults.
Page-level signals
25,898 unique cited pages were crawled directly to extract schema markup, HTTPS, mobile viewport tags, word count, statistics presence, and E-E-A-T heuristics (About/Contact links, credential language, named-staff mentions, phone/email presence). A second, independent full crawl was run for the E-E-A-T pass.
Business-level signals
Rating, review count, and review sentiment were collected via the Google Places API for a statistically-sized sample of the 1,000 most-cited businesses (all four surface×intent groups exceed n=449, above the standard threshold for a 95% confidence interval at ±5% margin). Review sentiment is a lexicon-based heuristic over Google’s own review snippets, not a trained model. Backlink authority, domain age (WHOIS), and social-platform presence were collected for the top 2,000 most-cited domains by citation frequency (all groups exceed n=365–653).
Scope notes
Social-platform follower counts/activity and third-party local-citation-directory counts were not pursued — the former requires platform-specific API access not available for this study, the latter a paid citation-tracking subscription. Backlink and domain-age figures are reported as medians, not means, since a small number of mega-domains (Reddit, Yelp, Facebook) skew averages by orders of magnitude; category breakdowns and technical signals are restricted to business_own_site citations where noted, to isolate business-level patterns from directory/social noise.
Limitations
- Single snapshot, not a trend line. All data was collected over a few days in August 2026. AI Overviews and AI Mode outputs are known to vary run-to-run for the same query; we did not repeat-test this dataset for consistency the way our companion Gemini study does. Aggregate patterns across thousands of queries are more stable than any single citation, but a repeat run at a different time could show different specific results.
- No JavaScript rendering in the page crawl. Page-level signals (schema markup, content length, statistics, E-E-A-T heuristics) were extracted from a plain HTTP fetch of each page’s initial HTML. Sites that inject schema markup or body content via client-side JavaScript would be undercounted on those specific signals — a real source of noise, though it would affect both surfaces’ citations roughly equally rather than favor one.
- Sampled, not exhaustive, on five signals. Backlink authority, domain age, and social presence were collected for the 2,000 most-cited domains; rating, review count, and sentiment for the 1,000 most-cited businesses — both sized to clear standard statistical thresholds per surface×intent group, not to cover every citation in the dataset. Long-tail businesses cited only once or twice are underrepresented in those five signals specifically (all other findings use the full dataset).
- Third-party collection, not a live user session. AI Overviews and AI Mode data was collected via DataForSEO rather than a logged-in Google account, so results reflect a generic, non-personalized session — not what any specific real searcher with their own history and location would see.
- Heuristic, not verified, on two signals. Review sentiment is a lexicon-based word-count score over Google’s review snippets, not a trained sentiment model. The domain-category taxonomy is a maintained but manually-built classifier; we found and fixed real errors during this study (a subdomain-matching bug, several misclassified B2B/directory domains) and it should be treated as a best-effort labeling, not ground truth.
- US home-services scope. All queries covered 10 home-service and professional verticals (plumbers, roofers, HVAC, electricians, dentists, personal injury lawyers, etc.) across 50 US metros, in English. Findings may not generalize to other business categories, countries, or languages.
See the stats/study in context: Steady Demand Research Index
