Guide · AI sources
How AI chooses the sources it cites
When ChatGPT, Perplexity or Google’s AI Overviews answer a question, they decide which pages to read, which brands to name and which links to show. Here is how that process works, how each engine differs, and which parts you can influence.
7-day Base trial · No card needed
Sources cited for “best UGC deal tracker”
One run- reddit.comForum
- g2.comReview
- notion.comCompetitor
- collabstr.comCompetitor
- ugcplanner.comYou
Sample data
In short
AI engines choose sources by searching an index, retrieving candidate pages, and using the passages that best answer the question. Brands get named when those sources, together with what the model learned in training, connect them to the question. Each engine searches and shows sources differently; Rankry records the sources each one returns for your prompts and shows which domains cite competitors but not you.
Retrieval
How do AI engines pick the sources they cite?
The details differ by engine and most of them are not public, but the broad process is the same everywhere.
- 01
Decide whether to search
For each question the engine decides whether it needs fresh information. Questions about products, prices, recent events or local businesses are more likely to trigger a search than general knowledge.
- 02
Rewrite the question into queries
The engine turns the question into one or more search queries. Google describes this for AI Overviews and AI Mode as query fan-out: running several related searches across subtopics.
- 03
Retrieve candidate pages
Queries go to a search index, the engine’s own or a partner’s. Only pages that are crawled, indexed and readable can come back.
- 04
Pick the useful passages
From the candidates, the model uses the passages that answer the question most directly. Clear statements, lists, tables and definitions are easy to use; vague copy is not.
- 05
Write and attribute
The model writes the answer, blending retrieved text with what it learned in training, and attaches some of the pages as citations or links.
ChatGPT
How does ChatGPT choose sources?
ChatGPT decides for each question whether to search the web. When it does, it rewrites the question into one or more queries and sends them to its search providers; OpenAI says these include third-party providers, and Microsoft Bing is widely reported to be one. OpenAI also runs OAI-SearchBot, its own crawler for surfacing sites in ChatGPT search. A page blocked from OAI-SearchBot is not shown in ChatGPT search answers.
From the results, ChatGPT reads a selection of pages and writes its answer with inline links and a list of sources. OpenAI does not publish how it ranks those pages, so the safest assumption is the obvious one: pages that answer the question asked, directly and specifically, have the best chance of being used.
When ChatGPT answers without searching, it relies on what the model learned in training and shows no sources at all. That part of its knowledge only changes when OpenAI releases new models.
How do AI Overviews choose sources?
AI Overviews and AI Mode are part of Google Search and draw on Google’s index. Google says a page must be indexed and eligible to be shown in Search with a snippet to appear as a supporting link, and that no additional technical requirements or special optimizations are needed. Both features may use query fan-out, issuing multiple related searches across subtopics to build a response.
That means SEO fundamentals carry straight into AI Overviews: crawlable pages, a healthy index status, and content that answers the query. Fan-out also means a page can be cited for a sub-question of the original search, not only for the main keyword. Pages that cover a topic thoroughly, with sections that each answer one specific question, give those related searches more to match.
Google’s usual preview controls apply too: a nosnippet rule can keep a page’s text out of AI Overviews. Google-Extended works differently. It limits the use of your content for AI training and grounding in some of Google’s other systems, such as Gemini, and it does not affect Google Search, so it is not a switch for AI Overviews or AI Mode.
Engine by engine
Which sources does each AI engine show?
Engines differ in how they search and in what they expose as sources. Those differences change what “being cited” means in each one.
| Engine | How it finds pages | What it shows as sources |
|---|---|---|
| ChatGPT | Decides per question whether to search; uses OpenAI’s crawler and third-party search providers | Inline links in the answer, plus the search results it consulted |
| Claude | Searches the web when the question needs it | Inline citations tied to the sentences they support, plus the search results consulted |
| Gemini | Grounds answers in Google Search when needed | Links to the Google Search results it used for grounding |
| Perplexity | Searches the web for nearly every question | A list of search results, with numbered references in the answer |
| Grok | Can search the web and posts on X | Inline citations in the answer text only |
| Google AI Overviews | Part of Google Search, using Google’s index and query fan-out | Supporting links shown with the overview |
| Google AI Mode | A conversational mode of Google Search that uses query fan-out | Supporting links and references within the answer |
Based on what each engine returns through its API or results page, as Rankry captures it. Engines change often, and an answer given without a search has no sources.
What gets cited
What kinds of sources do AI engines cite for brand questions?
For “best”, “alternatives” and comparison questions, citations usually come from a mix of these types. Which ones dominate depends on your category and the engine.
Your own site
Product, pricing, comparison and help pages. Most useful when they answer the question directly and are easy to quote.
Competitor sites
Competitors’ own pages, including their comparison pages, which may describe you on their terms.
Review sites
Platforms such as G2, Capterra and Trustpilot, where listings and reviews give AI structured, third-party opinions.
Forums and communities
Reddit threads, Q&A sites, YouTube and LinkedIn, where real users discuss what they picked and why.
Reference sources
Wikipedia and similar references, used for facts about companies, products and categories.
Editorial
Publications, blogs, roundups and “best of” lists written by third parties.
Your levers
What can you influence, and what can’t you?
Some parts of source selection are within your reach and others are not. Spend your effort on the first group.
| Factor | Can you influence it? | How |
|---|---|---|
| Whether AI crawlers can reach your pages | Yes | Allow search agents in robots.txt; keep key content in the HTML |
| Whether your pages are indexed | Mostly | Submit sitemaps to Google and Bing; fix crawl, canonical and noindex issues |
| How quotable your pages are | Yes | Answer-first sections, clear headings, tables, FAQs and current facts |
| Coverage on review sites and forums | Partly | Ask customers for reviews, join discussions honestly, give list authors accurate information |
| How consistently you are described | Mostly | Use the same name and description on your site and every profile |
| Which pages an engine prefers | No | Engines don’t publish their ranking; you can only improve your candidates |
| What the model learned in training | Not directly | It changes with new model versions; today’s public content can shape future training |
| Whether an engine searches at all | No | Engines decide per question; track many prompts to see the pattern |
Find your gaps
How do I find out which sources AI uses for my prompts?
You can see this for yourself for any prompt. It takes longer by hand, but the method is the same.
- 01
Run your buyer prompts
Ask each engine the same set of questions in a clean session, and open the sources or links for every answer.
- 02
Group sources by domain
List the cited domains and mark each one as yours, a competitor’s, a review site, a forum, a reference or editorial.
- 03
Mark the gaps
Note the domains that appear in answers naming competitors but not you. Those are your citation gaps.
- 04
Prioritize the reachable ones
Review sites and forums are usually easier to act on than a competitor’s own site. Start where you can get listed, reviewed or mentioned.
- 05
Track it with Rankry
Rankry’s Sources page does this for your current report: every cited domain with its type, a Yours, Gap or Mixed status, a “Gap analysis only” toggle, a priority, top URLs per domain and a filter by engine.
Guides
Keep reading
How does ChatGPT choose its sources?
When ChatGPT decides to search, it rewrites your question into queries, sends them to its search providers, reads a selection of the results and links the pages it used. OpenAI does not publish how it ranks pages, but a page has to be reachable by OAI-SearchBot or indexed by its search partners to be considered at all.
How do AI Overviews choose sources?
AI Overviews draw on Google’s search index. A page must be indexed and eligible to appear with a snippet, and Google may run several related searches, called query fan-out, to build the overview. Google says no special optimization is needed beyond that and does not publish how it picks supporting links.
Does AI always cite its sources?
No. When an engine answers from what the model learned in training, without searching, there are no sources to show. Engines also differ in what they expose: some list the search results they consulted, others only show inline citations for specific sentences.
Why does AI cite my page but not recommend my brand?
The page was useful for a fact or definition, but other sources pointed the model toward different brands as the answer. State plainly on the page who your product is for and how it compares, and earn mentions on the third-party sources that recommendation answers rely on.
Why do AI engines cite different sources for the same question?
They search different indexes, rewrite questions differently and weigh pages differently. Google’s AI features use Google’s index, ChatGPT uses OpenAI’s crawler and third-party providers, and Grok can also draw on posts from X. Track each engine separately.
Can I see which sources AI uses for my brand?
Yes. Open the sources or links in each answer to your buyer prompts and list the domains. Rankry does this for every tracked prompt: its Sources page lists the domains cited in your current report, their type, whether they mention you or only competitors, and the top URLs per domain.
Does domain authority decide what AI cites?
Not in any way an AI company has documented. Well-known, well-linked sites are often retrieved because they do well in search indexes, but specific, relevant pages from smaller sites also get cited when they answer the question better. Relevance to the exact question is the safest thing to optimize for.
Do AI engines prefer recent content?
For questions about prices, products and current options, engines that search can favor pages that look current. Keep dates, prices and product details up to date, and show when a page was last updated.
Is Reddit really a major AI source?
In many categories community threads show up among the sources for recommendation questions, but how much depends on your category and the engine. Check your own prompts rather than general studies. Rankry’s Sources page classifies Reddit and similar sites as Forum sources, so you can see their share in your own data.
See the sources behind your AI answers
Rankry tracks your buyer prompts across AI engines and shows which domains cite competitors but not you.
7-day Base trial · No card needed