Who does Google's AI Mode cite? A source map by query type
AI Mode cites Google's own cards for local and product questions, publishers for comparisons and ranking pages for explainers. What a brand controls in each.
The short answer
It depends on the question. For local and product queries, AI Mode increasingly cites Google's own Business Profile and product cards. Comparison queries lean on publishers and review sites, and informational queries on pages that already rank, reference sites and professional networks. A brand controls its listings and pages directly, and everything else only through the sources it earns.
Key takeaways
- Profound, which sells AI visibility software, found google.com's share of AI Mode citations grew 8.4 times between 15 April and 30 June 2026, driven by Business Profile and product cards.
- For local and product questions, a brand's Google listings are now part of its AI visibility, not only its search presence.
- Comparison questions are where publishers and review sites carry the most weight, and where a brand's own pages carry the least.
- Semrush's January to April 2026 data shows ChatGPT citing about 15 sources per answer against about 3 for Gemini, so the number of slots varies widely by engine.
- Our view: map your buyer questions by type before you decide where to spend, because each type rewards different work.
In this article
The short answer: it depends on the question
Google's AI Mode cites different kinds of sources for different kinds of questions. A question about a nearby service draws heavily on Google's own Business Profile cards. A question about a product draws on product knowledge panels, retailers and reviews. A comparison draws on publishers and review sites, and a how-to question on pages that already rank.
That matters because each kind of source responds to different work. A brand can edit its Google listings today. It can rewrite its own pages this month. It cannot edit a publisher's buying guide, a review site or a community thread, and can only influence them over months.
This post is a desk analysis. It combines published studies of AI citations with our own judgement about query types, and it does not report captures of our own. Where a study measured AI Overviews or another engine rather than AI Mode, we say so, because the surfaces differ. Our explainer on tracking AI Overviews and AI Mode separately sets out why.
What the data says about AI Mode's sources
The clearest recent signal is that Google is citing itself more. Profound, which sells AI visibility software, tracked AI Mode citations between 15 April and 30 June 2026. It found google.com's share of AI Mode citations grew 8.4 times over that period, making it the second most cited domain.1
Profound traced the rise to two kinds of Google-hosted pages: Business Profile cards and product knowledge panels. It reported the effect was strongest on local-intent queries, in categories such as hospitality, home services, restaurants, real estate and healthcare, and on product queries about comparisons, compatibility and specifications.1
Other studies show how much the number of citation slots varies. Semrush, which sells an AI visibility toolkit, analysed 126 million US prompts from January to April 2026. It found ChatGPT cites about 15 sources per response, against about 3 for Gemini.2 AI Mode was not reported separately in that figure, but the spread shows why a source map has to be drawn engine by engine.
AI citations in 2026
Who gets cited is shifting, and slots vary
Our view: the google.com finding is the most practical one in this set. It means that for a large share of local and product questions, the page AI Mode quotes about you is one Google hosts and you edit.
How we drew the map
We sorted buyer questions into four types and, for each, asked which kinds of sources the published studies say AI answers draw on. We then judged how much of each a brand can change.
- Google-hosted cards
- Business Profiles, product knowledge panels and other pages on google.com that Google builds from listings, feeds and reviews.
- Publishers and review sites
- Editorial buying guides, trade press, comparison sites and review platforms.
- Community and professional networks
- Forums, Q&A sites and professional networks where people discuss products and services.
- Brand pages
- Your own site: product, service, pricing, documentation and help pages.
The four query types are local (a service near the buyer), product (a specific item or its specification), comparison (one option against another, or a best-of list) and informational (how something works or what to do). Many real prompts mix types, so treat the map as a starting point, not a rule.
Local queries: Google's own cards come first
For local questions, AI Mode increasingly cites Google's Business Profile cards. Profound reported local-intent verticals, such as home services, restaurants and healthcare, among the most affected by the rise in google.com citations.1
A Business Profile carries hours, categories, services, photos and reviews. When AI Mode answers "a dentist open on Saturday who takes new patients", much of what it says may come from those fields. A practice whose profile lists the wrong hours or no services is giving the answer wrong facts to repeat.
Profound's advice is to complete every profile field and to manage the reviews that appear on Google's cards, because they now sit in front of the testimonials on your own site.1 That advice is sound and cheap to act on.
What a brand controls here: almost everything on the card, except the reviews themselves. What it influences: the volume and recency of reviews, by asking happy customers. What it cannot change: how Google weighs one profile against another.
Brands with many locations face a scale version of the same problem. Each location has its own profile, and each can drift: a closed branch still listed as open, a service offered in some branches but listed in all. Our view: audit the profiles in bulk before writing any new content, because a single wrong card can be quoted in every answer about that area.
Product queries: panels, retailers and reviews
For product questions, AI Mode draws on product knowledge panels alongside retailers and reviews. Profound reported that product queries involving comparisons, compatibility or specifications were among those most affected by Google's own cards.1
Product knowledge panels are built from many inputs, including merchant feeds and the retailers that list the item. So a manufacturer's visibility depends partly on how consistently its retailers describe the product. A specification that differs between your site and two retailers gives the answer three versions to choose from.
What a brand controls: its own product pages and the feeds it submits. What it influences: retailer listings, through the content it supplies them, and reviews. Our guide to why AI engines skip some products goes through the feed and catalogue fixes in order.
Our view: for a product brand, the cheapest AI visibility work is often a data clean-up across feeds and retailers. It is not glamorous, but it removes contradictions that no amount of new content can outweigh.
Comparison queries: publishers and review sites carry the weight
Comparison questions are where third parties matter most and a brand's own pages least. A buyer asking "X vs Y" or "best tool for Z" is asking for an outside view, and answers tend to reflect it.
These queries also show AI summaries very often. Seer Interactive studied 53 brands from January 2025 to February 2026 and found that 95.4% of "X vs Y" queries showed an AI Overview.4 That figure is for AI Overviews, not AI Mode, but it confirms that comparisons are prime ground for AI answers on Google.
Visibility in AI answers is also concentrated, and more so in some categories. In Semrush's January to April 2026 US data, the top three brands held 41.4% of AI visibility in finance and 82.9% in news and media.2 In a concentrated category, the sources that name the leaders carry most of the weight.
Share of AI visibility held by the top three brands
What a brand controls: a fair, factual comparison page of its own, with dated prices and specifications. What it influences: inclusion in publishers' guides and review sites, through briefings, data and product access. What it cannot change: an editor's verdict. A fair comparison page on your own site will not outrank a publisher's verdict, but it gives the answer accurate facts to set beside it.
Informational queries: ranking pages, reference sites and networks
For how-to and explainer questions, AI answers on Google still draw heavily on pages that rank, though ranking is not enough. Ahrefs, which sells SEO and AI visibility tools, studied 863,000 results pages and 4 million cited URLs and found 38% of pages cited in AI Overviews ranked in Google's top 10 for the same query; the data period was not stated.3
That leaves most AI Overview citations coming from outside the top 10. Our explainer on ranking and AI Overview citations looks at what that overlap means for SEO budgets.
Reference sites and professional networks also show up widely. Semrush's weekly snapshots from 14 July to 12 October 2025 found Reddit and Wikipedia were ChatGPT's two most cited domains, and LinkedIn appeared in nearly 15% of AI Mode responses.5 For a B2B brand, that makes the professional network a source worth tending, since the posts and articles of its own people can be cited.
What a brand controls: explainer pages that answer the question in the first lines, with dates and named authors. What it influences: what its experts publish on professional networks, and whether reference sites describe it accurately. Our review of the evidence on schema, llms.txt and community sites covers which page-level tactics have support.
Where do brand pages still win?
Brand pages win where the brand is the only reliable source of a fact. Prices, specifications, service areas, opening policies, integration lists and documentation are things no publisher knows better than you. When a buyer's question turns on one of those facts, a clear page on your own site is the natural source to cite.
That holds across all four query types. A local answer may quote your profile for hours but your site for what a service includes. A comparison answer may rely on a publisher's verdict but take the current price from your pricing page. An informational answer about setting up your product should come from your own documentation.
The common failure is not missing pages but hidden facts. Prices behind a contact form, specifications in a PDF, service areas only in an image: an answer engine cannot quote what it cannot read. Our own site follows the same rule, with published prices and answer-first pages.
Our view: list the five facts a buyer most needs before choosing you, and check that each one sits in plain text on a crawlable page, with a date. It is the smallest piece of work in this post and often the one with the clearest effect on accuracy. Our guide to fixing what ChatGPT gets wrong about a company covers the same discipline from the correction side.
The source map in one table
Put together, the map shows where to spend for each type of question. The judgements in the last column are ours.
| Query type | Sources AI answers lean on | What the brand controls |
|---|---|---|
| Local | Business Profile cards, reviews | The profile fields; review requests |
| Product | Product panels, retailers, reviews | Product pages, feeds, retailer content |
| Comparison | Publishers, review sites, comparison pages | A fair comparison page; briefings |
| Informational | Ranking pages, reference sites, networks | Answer-first explainers; experts' posts |
Our synthesis of the studies cited in this post; the brand-control column is our judgement.
Where a brand's effort pays back fastest
Brand control →
Weight in AI Mode answers →
Our view: most brands should start in the top-right quadrant. Listings and product data are quick to fix, AI Mode cites them often, and the work also helps ordinary search.
How to check the map for your own category
Run your own buyer questions and record the sources, rather than relying on averages from other categories. The studies above are broad, and your category's mix can differ sharply.
Map your category's sources in five steps
- Write the prompts
Group real buyer questions into local, product, comparison and informational.
- Run them repeatedly
Run each prompt on several separate days, because answers vary.
- Record every source
Log the cited URLs and classify each one into the four source kinds.
- Mark what you control
Tag each source as yours, influenced or outside your reach.
- Fix in order
Start with sources you control that appear often.
A Sigzen AI audit does this across six engines: ChatGPT, Gemini, Perplexity, Claude, Copilot and Google AI. It uses official APIs where they are offered and a licensed AI-answer data provider otherwise, and runs each prompt on three separate days. The output includes a citation-source map for your category, which is this post's table filled in with your own data.
What this map cannot tell you
It cannot tell you how often a source is used inside an answer without being shown as a citation. Every study here counts visible citations, and an answer can draw on more than it links.
It also mixes surfaces. Profound's data is AI Mode. Seer's and Ahrefs' are AI Overviews. Semrush's sources-per-response figure covers ChatGPT and Gemini. We have kept each figure with its surface, but the four query types are our framing, not a split any of these studies made. And the vendors behind most of the data sell AI visibility tools, which gives them an interest in the category looking important.
Finally, it will date. Google changes AI Mode often, and the google.com shift Profound measured happened within a single quarter. The surface is also large: Google said at I/O on 19 May 2026 that AI Mode had passed one billion monthly users, a company statement without a published method.6 Re-run your own map at least every quarter, and treat any single snapshot, including this one, as a starting point.
Sources
- Profound, google.com in AI Mode citations, 15 Apr–30 Jun 2026; industry breakdown from 32M+ Business Profile citations (Jul 2026)
- Semrush, 2026 AI Visibility Index: 126M US prompts, Jan–Apr 2026 (Jun 2026)
- Ahrefs, AI Overview citations and the top 10: 863,000 SERPs, 4M cited URLs (Mar 2026); data period and geography not stated
- Seer Interactive, AI Overviews and CTR, 2026 update: 53 brands, Jan 2025–Feb 2026 (Apr 2026)
- Semrush, most-cited domains in AI search: weekly snapshots 14 Jul–12 Oct 2025 (Nov 2025)
- Google, Search updates at I/O 2026: AI Mode monthly users, company statement (May 2026)
