Ask an AI about a brand with no sense of where the human is, and you are likely to see sources like Wikipedia. That is rarely the question a customer asks. Location is central to the context an answer is built on, and assistants keep it on by default, so a brand question is usually a place question. The 155.5 million citations in this study were earned on location-grounded questions, the kind customers actually ask. Four in five pointed to a website or a listing, the two things a brand can keep current, and the mix likely shifts again at every level of geography, from national to DMA to state to city to neighborhood to the single location.
What to do. Read your AI presence with location in the question, at every level you compete at, and keep your website pages, listings, and the small directories current, because the models read them and cite what they find.
Each citation was classified by the type of source it points to. Websites are pages on the brand's own domain. Listings are public directories that carry structured business facts. Reviews and social covers platforms where customers write about the brand. News and other covers press, forums, and everything else. The study counts the first two as influenceable, since a brand writes its own pages and, when listings are managed, controls those too.
The forgotten publishers top the third-party list
Set the brand-owned domains aside and rank where else the models point. The top spot belongs to MapQuest, with almost six million citations in one quarter, 5.8 for every one that Wikipedia earned. TripAdvisor sits second and climbing, and Yelp is third and new to the list, so the whole top ten is directories and review platforms with the internet's encyclopedia as the lone exception.
One reading is validation. Business facts used to live across a crowd of small publishers, back before search converged on one engine, and the AI models appear to have re-opened that market, cross-checking that a brand's details agree on the smaller directories as well as the big platforms and citing the places where the facts line up.
MapQuest has held the top spot across quarters, while TripAdvisor, Uber Eats, Booking, and Expedia moved up and Yelp, Wikipedia, and Grubhub entered the list this quarter. Healthgrades, the Better Business Bureau, and WebMD each slipped a rank or two and stayed on it.
Four in five citations sit in the influenceable zones
Across all four models, 80 percent of the citations pointed at a website or a listing, the two zones a brand can keep current, while the other fifth came from reviews, social, press, and forums, the zones it can only nudge.
Per model, the influenceable share runs 85 percent for OpenAI, 81 for Gemini, 80 for Perplexity, and 69 for Anthropic. That is a 16-point spread between the highest model and the lowest.
Each bar splits one model's citations by source type, and the two left segments together are the influenceable share, 85 percent for OpenAI and 69 for Anthropic. A few percent of each model's citations resist classification and sit outside the four groups.
of the quarter's AI citations pointed to a source a brand can influence, holding near four in five even as citation volume nearly tripled.
A reviews-heavy model moved the average
A quarter earlier the influenceable share read higher, and one number explains the slide to 80. Anthropic pulls seven and a half times the share of its citations from reviews and social that OpenAI does, and the panel scanned 15 times more Anthropic answers than the quarter before, so a reviews-heavy model arrived at scale and the blended average moved with it.
We are not claiming the models became harder to influence, since the three models tracked from the start held their shapes and only the mix of models being measured changed.
Anthropic draws seven and a half times the share of its citations from reviews and social that OpenAI does, so which model answers now shapes how influenceable the answer is.
AI citations are compounding
The database behind this study grew fast, so raw totals flatter the trend. The honest read holds the panel to the same 770 brands tracked since the third quarter of 2025, and that cohort's citations nearly tripled in three quarters.
The sharper number is per scan. Each time a single AI answer was scanned, the cohort's brands were cited 28 percent more often than the quarter before. The growth is in the answers themselves, on top of any growth in scanning.
These bars are the whole study, so they grow with the scanning and the brand panel as well as with the citing. The like-for-like reads are the cohort numbers, a near tripling in citations and 28 percent growth per scan.
Why the models cite these sources
These are hypotheses rather than settled conclusions, and each one can be checked against what the models cite for your own brand.
- A located question wants sources that know places. With the person's location in play, the models reach for sources that map businesses to places, which is what a directory is. Strip the location away and reference pages would likely take more of the list.
- Structured sources are easy to quote. A listing carries a name, address, hours, and services as discrete fields, which is the shape an answer engine needs to compose a reply.
- Cross-checking beats trusting one source. A business fact that reads the same on the brand's site, a directory, and a review platform is a fact a model can state with confidence.
- Each model reflects its own retrieval choices. The same question resolves to brand pages in Gemini and Perplexity, to directories in OpenAI, and to review and social content far more often in Anthropic.
Three practical steps
Keep the influenceable zones current
Websites and listings drew four of every five citations measured, so accurate hours, services, and details on both is the largest surface you can work on directly.
Update the smaller directories too
MapQuest, TripAdvisor, and Yelp top the third-party list. Consistent facts across the smaller publishers is what the models appear to cross-check before they cite.
Build a review presence for the reviews-heavy models
One model already draws a fifth of its citations from reviews and social. Fresh reviews, responses, and active profiles reach the zone you only partly control.
Ask each model about your own locations and note what each answer cites. The zone split you get back is your own version of Fig 2. A full scan can come later.
Questions readers ask
Why does location change what AI cites?+
Where do AI models get their information about businesses?+
Which AI model cites reviews the most?+
Why does MapQuest show up in so many AI answers?+
Are AI citations of businesses still growing?+
How this was measured
- The scan
- Millions of search results and AI answers are scanned continuously across Perplexity, Gemini, OpenAI, and Anthropic, then parsed down to individual citations: for every question asked, which brands surface and which source each fact points to. The questions carry local intent, asked the way a nearby customer would ask them, which matches how consumer assistants answer, with location on by default. The Q1 2026 study covers 155.5 million citations across 1,623 brands.
- The cohort
- Growth figures hold the panel to the same 770 brands tracked since Q3 2025, so they read like-for-like. Per-scan growth divides cohort citations by scans run, which separates real citation growth from growth in scanning. Study-wide totals by model grew faster than the cohort because the scanning volume and the brand panel grew over the same period.
- Zones of control
- Websites: pages on the brand's own domain, full control. Listings: public directories, fully controllable when managed. Reviews and social: limited control. News, press, and forums: none. The influenceable share counts websites plus listings.
- Third-party domains
- Ranked by citation count with all brand-owned domains excluded. Movement flags compare against the prior quarter's ranking.
- Future Research
- Next cuts should separate citations by geography, query intent, and source freshness. The strongest next test is whether the same brand shifts from directories to its own pages as the question gets more specific or as the searched location changes. A second track should connect citation presence with downstream outcomes where traffic, lead, or conversion data is available.