What is an AI citation?
An AI citation is a source an AI answer engine shows alongside its answer, usually as a link, a numbered reference or a source card, to show where the information came from. When Perplexity answers "what is the best CRM for a small team" and lists five web pages under the answer, each of those pages has earned an AI citation.
Google calls these "supporting links" in AI Overviews and AI Mode [1]. ChatGPT, Perplexity and Claude show them as numbered references or source lists. The format differs by product, but the idea is the same: a specific page credited as a source for a specific answer.
The AI citation is the unit of success in answer engine optimization. It plays the role a ranking position plays in SEO. There is no industry-standard definition yet, so this entry sets out what counts and what does not.
What counts as an AI citation?
An AI citation counts when a page is linked or explicitly named as a source for the answer. Search Insider uses three tests:
- The source is identifiable. The answer links to, or names, a specific page or document.
- It is attached to the answer. The link sits inside or under the generated answer, not in an ad, a sponsored slot, or a separate list of normal search results.
- It is visible to the user. A source the system used but did not show is not a citation, because no one can see or click it.
Three things are often counted as citations but are not:
- A brand mention. "HubSpot is a popular CRM" names a brand but credits no source. That is a mention, which is a different metric (see share of model).
- A training influence. If a model learned a fact from a page during training, that page is not cited. Nothing in the answer points to it.
- A normal search result below the answer. On Google, the ten blue links under an AI Overview are rankings, not AI citations.
How do AI engines choose which pages to cite?
AI engines choose citations by running searches, retrieving pages from their own index, and linking the pages whose passages they used to write the answer. That process is called grounding. The details differ by engine:
| Engine | Where citations come from | Can a site opt out? |
|---|---|---|
| Google AI Overviews and AI Mode | Google's search index, via query fan-out [1] | Yes, Search generative AI control in Search Console |
| ChatGPT search | OpenAI's search index, built by OAI-SearchBot [2] | Blocking OAI-SearchBot removes a site from search answers, except navigational links [2] |
| Perplexity | Perplexity's index, built by PerplexityBot [3] | Blocking PerplexityBot, though Perplexity-User may still fetch pages a user asks about [3] |
| Claude | Anthropic's search crawling (Claude-SearchBot) and live fetches (Claude-User) [4] | Blocking those bots, per Anthropic's documentation [4] |
Two conditions sit underneath all of these. The page must be reachable by the engine's search crawler, and it must be in the engine's index. For Google, the stated requirement is a page "indexed and eligible to be shown in Google Search with a snippet," with "no additional technical requirements" [1].
What is the difference between an AI citation and a backlink?
An AI citation is a link shown inside a generated answer, while a backlink is a link from one web page to another. A backlink is permanent until someone edits the page. An AI citation exists only in that answer, for that user, at that moment. Ask the same question tomorrow and the cited sources can change [8].
The two are related. Search engines use backlinks to judge which pages deserve to rank, and AI engines that ground answers in search results tend to cite pages that already rank. Search Insider found AI Overview citations "moved with the organic results rather than independently" when keywords were edited [9].
Do AI citations send traffic?
AI citations send some traffic, but far less than the citation count suggests, and no AI company publishes click-through rates. A Pew Research Center study, as summarized in Wikipedia's AI Overviews article, found only 1% of users clicked a source link directly from an AI Overview [10].
Google's own reporting confirms the gap in data. Search Console's generative AI report, rolled out worldwide by August 31, 2026, shows how often a site's pages appeared in AI Overviews, AI Mode and Discover AI features, but reports impressions only, with no clicks or queries [6]. Google also notes that all links within an AI Overview share a single position [7], so a citation's place in the list is not reported as a ranking.
That makes an AI citation a visibility metric first and a traffic metric second.
How many sources does an AI answer cite?
A Google AI Overview cites about six sources on average, based on Search Insider data. In Search Insider's study of 1,000 searches, an AI Overview appeared on 947 and produced 5,834 citations across 17,240 ranking pages [8]. That works out to about 6.2 cited sources per AI Overview, and means most ranking pages on any given search are not cited.
Search Insider has not published equivalent counts for ChatGPT, Perplexity or AI Mode, and no AI company publishes them. Any per-engine average quoted elsewhere comes from a third-party sample and should be read with its method.
How do you track AI citations?
To track AI citations, run a fixed list of real questions through each AI engine on a schedule, record every source each answer links to, and count how often your domain appears. The three parts that matter:
- A fixed prompt set. The same questions every time, written the way customers ask them. Changing the questions changes the result.
- Repeat runs. AI answers change over time [8] and can differ from one run to the next, so one run is an anecdote. Run each question several times and report the share of runs where you were cited.
- Per engine, per date. A citation in Perplexity and a citation in AI Mode are different events. Keep them separate.
Commercial tools automate this. Profound, for example, tracks which websites influence AI answers about a brand across ChatGPT, Perplexity, Claude, Gemini, Google AI Overviews, Copilot and others [11]. Google Search Console adds impressions for Google's AI features only [6].
Counting your own citations against every citation in the answer set gives citation share of voice, the metric for comparing yourself with competitors.
When were AI citations introduced?
No single date marks the first AI citation, but the idea became measurable in November 2023, when researchers introduced generative engine optimization and scored how prominently sources appeared in generative answers [5]. AI citations reached Google Search with the US launch of AI Overviews on May 14, 2024 [10].
2023: citations become measurable
The GEO paper by Aggarwal and colleagues, first published on November 16, 2023 and presented at KDD 2024, reported that optimization could increase a source's visibility in generative engine responses by up to 40% [5].
2024: Google adds AI Overviews
Google launched AI Overviews in the US on May 14, 2024, after testing the feature as the Search Generative Experience from May 2023 [10]. Supporting links in AI Overviews brought AI citations to the biggest search engine.
2026: Google reports AI impressions
In 2026, Google added a Search Console report of impressions in AI Overviews, AI Mode and Discover AI features, the first first-party measure of how often a site appears in Google's AI answers [6].
How do you get more AI citations?
You get more AI citations by ranking for the questions and sub-questions people ask, and by writing passages an engine can lift whole. Google says no special optimization is required beyond standard SEO [1].
What helps in practice:
- Let the right crawlers in. Allow each engine's search crawler, which is separate from its training crawler [2][3][4].
- Answer the question in the first sentence of a section. Engines quote passages, not pages.
- Rank first. Search Insider's data shows Google's AI citations follow organic rankings closely [9].
- Improve the page, not the label. Adding a year to an unchanged title gave no measurable AI Overview citation advantage once pages of the same age were compared [8].
What is not known about AI citations?
No AI company publishes how it selects citations, how many it shows, or how citations convert to clicks. Studies of AI citations measure correlation in a sample of questions on a given date. They show what tends to be cited, not why. A tool claiming to know the exact formula behind any engine's citations is claiming more than any public evidence supports.