TL;DR
- AI tools cite sources in two stages: they search an index (Bing, Google, Brave, or their own), then pick the passages that answer the question best. In an AirOps study of 15,000 prompts, ChatGPT cited only 15% of the pages it retrieved.
- Google rankings still help, but less than they did. Ahrefs found 38% of AI Overview citations came from top-10 pages in early 2026, down from 76% in July 2025.
- Brand mentions track AI visibility more closely than backlinks. In an Ahrefs study of 75,000 brands, branded web mentions correlated at 0.664 with AI Overview visibility, versus 0.218 for backlinks.
- Answer-first writing gets quoted. A Kevin Indig analysis reported by Search Engine Land found 44.2% of ChatGPT citations came from the first 30% of a page.
- Schema and llms.txt don’t move the needle on their own. Ahrefs measured no major citation uplift from adding schema on any AI platform.
What Does It Mean When an AI “Cites” a Source?
When ChatGPT, Perplexity, or Google’s AI features cite a source, they are showing you a web page they retrieved live and used to write the answer. This process is called retrieval-augmented generation (RAG). Amazon Web Services defines RAG as having a language model reference “an authoritative knowledge base outside of its training data sources before generating a response,” and notes that the output “can include citations or references to sources.”
Google calls the same idea grounding. Its AI optimization guide says its generative features rely “on our core Search ranking systems to retrieve relevant, up-to-date web pages.”
That distinction matters for business owners. A model’s training data decides what it “knows” in general, but citations come from search results fetched at the moment someone asks. So getting cited is mostly a search visibility problem, and you can work on it the same way you work on SEO.
How Do AI Tools Choose Sources, Step by Step?
AI tools choose sources in four steps: they rewrite the question into searches, retrieve pages from a search index, score individual passages, and cite the few that best support the answer. Most retrieved pages never make the final cut.
- Rewrite and fan out. The tool turns your prompt into one or more search queries. OpenAI’s help page for ChatGPT search says it “typically rewrites your query into one or more targeted queries” for its search partners. Google describes “query fan-out” as issuing multiple related searches across subtopics.
- Retrieve. Each query pulls candidate pages from an index. Which index depends on the platform (see the table below).
- Score passages. The tool judges sections of pages, not whole pages. Perplexity explains that it splits documents into “fine-grained units” that are “individually surfaced and scored against the original query.”
- Select and cite. The model writes the answer and links to the passages it used.
The drop-off between steps 2 and 4 is steep. In the AirOps study (March 2026), ChatGPT retrieved 548,534 pages across 15,000 prompts and cited only 15% of them. It ran follow-up searches for 89.6% of prompts, and 32.9% of cited pages showed up only in those follow-up searches. Being findable gets you into the pool. Being the clearest answer gets you cited.
Where Does Each AI Platform Get Its Sources?
Each AI platform pulls sources from a different search index and uses its own crawler, so your visibility can differ a lot from one tool to the next. This table is built from each company’s own documentation, plus one news report for Claude.
| Platform | Where sources come from | Crawler to allow | How citations show |
|---|---|---|---|
| ChatGPT search | Its own crawling plus search partners (OpenAI) | OAI-SearchBot (OpenAI) | Inline links and a Sources list |
| Perplexity | Its own index of “hundreds of billions of webpages” (Perplexity) | PerplexityBot (Perplexity) | Numbered inline citations |
| Google AI Overviews and AI Mode | Google Search index and ranking systems (Google) | Googlebot | Supporting links beside the summary |
| Gemini app | Google Search grounding (Google AI for Developers) | Googlebot | Sources button, not on every answer (Gemini Help) |
| Claude | Web search, reported to use Brave Search (TechCrunch) | Claude-SearchBot (Search Engine Journal) | Direct citations (Anthropic) |
| Microsoft Copilot | Bing search (Microsoft Learn) | Bingbot | Prominent source links (Bing) |
Bing matters more than most business owners think. A Seer Interactive test of 100 queries (February 2025) found that more than 87% of ChatGPT search citations matched Bing’s top 20 organic results. It’s a small sample, but if you have never checked your site in Bing Webmaster Tools, you may be invisible to two of these six tools.
Do Google Rankings Still Matter for AI Citations?
Yes, Google rankings still matter for AI citations, but a top-10 position is no longer the main way in. Ahrefs found that 76% of AI Overview citations ranked in the top 10 in July 2025. Its March 2026 update, covering 863,000 keywords, put that figure at about 38%, and linked the drop to query fan-out.
The gap is wider outside Google. For long-tail prompts, Ahrefs found (August 2025) only 12% of AI-cited URLs ranked in Google’s top 10 for the original query:
| Assistant | Citations ranking in Google’s top 10 |
|---|---|
| Perplexity | 28.6% |
| Gemini | 8.6% |
| Copilot | 8.2% |
| ChatGPT | 8% |
Rankings still give you an edge. In the AirOps study, pages ranking first on Google were cited 43.2% of the time when retrieved, 3.5 times more often than pages outside the top 20. Results also vary by industry. BrightEdge reported that 75.3% of healthcare AI Overview citations also ranked organically, against 22.9% for ecommerce.
The practical takeaway: rank for the main question and for the follow-up questions around it. That’s the case I make in what GEO is and how it differs from AEO.
Why Do Brand Mentions Matter More Than Backlinks?
Brand mentions matter more than backlinks because AI tools learn which businesses are trusted for a topic from how often other sites talk about them, not only from who links to them. In an Ahrefs study of 75,000 brands (May 2025), these factors correlated with appearing in AI Overviews:
| Signal | Correlation with AI Overview visibility |
|---|---|
| Branded web mentions | 0.664 |
| Branded anchor text | 0.527 |
| Branded search volume | 0.392 |
| Domain Rating | 0.326 |
| Backlinks | 0.218 |
The places that get mentioned most are community and video platforms. A Peec AI study of 30 million sources found Reddit was the most-cited domain across five AI platforms, followed by YouTube and LinkedIn. Those patterns shift, though. Semrush tracked ChatGPT’s Reddit citations falling from about 60% of responses to about 10% after mid-September 2025.
Links still count for Google rankings, which feed AI retrieval. I cover that balance in do backlinks still matter in the age of AI search.
What Makes a Passage More Likely to Be Cited?
A passage is more likely to be cited when it answers the question early, sits under a heading that matches the question, names specific people and things, and includes evidence such as statistics or quotes. These patterns show up across several studies.
- Put the answer near the top. In Kevin Indig’s analysis of 1.2 million AI answers, 44.2% of ChatGPT citations came from the first 30% of a page, 31.1% from the middle, and 24.7% from the final third.
- Use question-style headings. The same study found 78.4% of question-linked citations came from H2 headings phrased as questions.
- Be specific. Heavily cited text had about 20.6% entity density, meaning names of brands, places, products, and people, per the same analysis.
- Add evidence. The GEO research paper from Princeton and partners (KDD 2024) found that adding citations, quotations, and statistics increased source visibility by over 40% across queries. Keyword stuffing gave “little to no improvement.”
- Keep it current. Ahrefs found URLs cited by AI assistants were 25.7% fresher on average than URLs in organic results.
One caution about that GEO paper: some articles quote its 115.1% figure as a general boost. The paper itself reports it only for websites ranked fifth in search results, while the top-ranked site’s visibility fell 30.3% on average. Lower-ranked sites gain the most from citing sources.
Every article on famamun.com, including this one, follows that structure: a question as the H2, a direct answer in the first sentence, and a named source linked beside every statistic. I write this way for readers first, and the same clarity makes a passage easy for an AI tool to quote.
Does Schema Markup or llms.txt Help You Get Cited?
No, neither schema markup nor llms.txt has been shown to increase AI citations by itself. In an Ahrefs test (May 2026) of 1,885 pages that added JSON-LD schema against 4,000 control pages, citations changed by -4.6% in AI Overviews, +2.4% in AI Mode, and +2.2% in ChatGPT. Ahrefs concluded that schema “produced no major uplift in citations on any platform.”
Google says the same thing directly. Its AI features documentation states there’s “no special schema.org structured data that you need to add” and no need for “AI text files.” I still add schema to client sites because it supports rich results and keeps business facts consistent, but I don’t sell it as an AI citation tactic. I explain the llms.txt side in do you need an llms.txt file.
What Technical Access Does Your Site Need?
Your site needs to let each platform’s search crawler in and be indexed with a snippet in Google. Blocking the wrong crawler in robots.txt can remove you from an AI tool’s answers without you noticing.
| Crawler | What it does | Block it? |
|---|---|---|
| OAI-SearchBot | Surfaces sites in ChatGPT search (OpenAI) | Allow if you want ChatGPT citations |
| GPTBot | Collects content for model training (OpenAI) | Your choice; it doesn’t control ChatGPT search |
| PerplexityBot | Surfaces sites in Perplexity results, not used for training (Perplexity) | Allow |
| Claude-SearchBot | Indexes content for Claude’s search; blocking may reduce visibility (Search Engine Journal) | Allow |
| Google-Extended | Controls Gemini training and grounding, not Search inclusion (Google) | Your choice; it doesn’t affect AI Overviews |
For Google’s AI features, Google states that a page must be indexed and eligible to show with a snippet. A nosnippet tag or a noindexed page removes you from AI Overviews. Also check your CDN or firewall: OpenAI notes that your host must allow traffic from its published search bot IP addresses.
How Do Local Businesses, Law Firms, and Medical Practices Get Cited?
Local businesses get cited mostly through their listings and reviews, while law firms and medical practices also need strong organic rankings for their practice-area questions. BrightLocal’s August 2026 study of 60,970 local AI checks found Google Business Profile made up 28.63% of local citations across AI Overviews, AI Mode, and ChatGPT, with Yelp at 9.53% and the most-used source for ChatGPT.
That means a local firm’s first job is keeping its Google Business Profile, Yelp listing, and website details identical. When I worked with Immigration Lawyers USA, one of the first fixes was consolidating three competing Google Business Profile listings into one, then building a five-hub site structure so each practice area had pages that answer clients’ real questions.
For healthcare, rankings carry more weight: BrightEdge found 75.3% of healthcare AI Overview citations also ranked organically. I explain how assistants handle local recommendation prompts in how AI assistants answer “best service near me” questions.
During my year as an SEO expert for doctorsfaculty.com, a healthcare education platform, compliance-aware content mattered as much as rankings. On health topics, a citation only helps you if the passage being quoted is accurate.
How Can You See Which AI Platforms Cite You?
You can see some AI citations directly in Bing Webmaster Tools, and track the rest by testing prompts on each platform every month. Bing’s AI Performance report, launched in public preview in February 2026, shows total citations across Copilot and Bing AI summaries, the grounding queries behind them, and which of your pages were cited. Google’s Search Console reports AI Overview and AI Mode impressions, which I cover in how to optimize for Google AI Overviews.
Test more than once. Semrush found (June 2026) that ChatGPT’s Instant and Thinking modes shared only 25.6% of cited domains, and Thinking mode cited 4.5 sources per answer on average against 2.6 for Instant. One check in one mode tells you very little. My routine is in how to track whether your content is being cited by AI platforms.
If you’d like to know where you stand right now, my free AI visibility audit checks whether ChatGPT and Google AI Overviews already mention your business. You can also see my AI SEO and AEO services or book a strategy call.
Frequently Asked Questions
How do LLMs decide which sources to cite?
They search an index, score passages against the question, and cite the clearest matches. In an AirOps study, ChatGPT cited only 15% of the pages it retrieved.
Does ChatGPT use Google or Bing to find sources?
ChatGPT search uses its own crawler, OAI-SearchBot, and partners with other search providers, according to OpenAI. A Seer Interactive test found over 87% of its citations matched Bing’s top 20 results.
Do I need to rank in Google’s top 10 to get cited by AI?
No. Ahrefs found only about 38% of AI Overview citations came from top-10 pages in early 2026, because AI tools also search related follow-up questions.
Does schema markup help me get cited by ChatGPT?
Not by itself. An Ahrefs test found adding schema produced no major uplift in citations on any AI platform, though schema still supports rich results.
Will blocking GPTBot remove my site from ChatGPT?
No. OpenAI says GPTBot collects training data, while OAI-SearchBot surfaces sites in ChatGPT search. Keep OAI-SearchBot allowed if you want to be cited.
Why does Perplexity cite different sites than ChatGPT?
They use different indexes. Perplexity runs its own index, and Ahrefs found 28.6% of Perplexity citations ranked in Google’s top 10, compared with 8% for ChatGPT.
