Short answer: AI models choose sources that directly answer the question, come from sites they trust, and agree with other sources. To see which sites ChatGPT, Gemini and Perplexity cite for your brand, ask your customers' questions repeatedly, record every cited link, and compare the domains cited when you appear with those cited when a competitor appears instead.
Key takeaways
- AI answers that use web search are built from a small set of cited sources.
- Those sources are usually third-party sites — reviews, media, marketplaces, video, forums — not just brand websites.
- Comparing your cited sources with a competitor's reveals a citation gap: the places to improve.
- Track AI Citations by domain over time, alongside Mention Rate.
How AI models choose sources
When an AI assistant searches the web, it runs one or more searches, reads the results, and writes an answer from the pages it finds most useful. Pages are more likely to be used when they:
- Answer the question directly, near the top of the page.
- Are specific — numbers, names, dates and clear comparisons.
- Come from a trusted site for that topic.
- Agree with other sources, so the AI can state facts with confidence.
- Are current, especially for products and prices.
Different assistants weigh these differently, which is why ChatGPT, Gemini and Perplexity often cite different domains for the same question.
The main source types
| Source type | Examples | Typical role in AI answers |
|---|---|---|
| Brand websites | Product, pricing and help pages | Facts about the brand itself |
| Review & comparison sites | Review platforms, "best of" articles | Recommendations and rankings |
| News & industry media | Trade press, newspapers | Credibility and recent events |
| Marketplaces & retailers | E-commerce listings | Product availability and prices |
| Video & community | Video platforms, forums | Real-world experience and how-tos |
| Reference | Encyclopedias, government and institutional sites | Definitions and factual grounding |
How to see which sites AI cites for your brand
- List your customer questions (unaided — no brand name).
- Run them repeatedly in each AI model with web search. Pranas asks each question 20 times per AI Scan.
- Collect every cited domain, for answers that mention you and answers that do not.
- Group the domains by type using the table above.
- Compare with competitors: which domains are cited when they are named and you are not?
In Pranas, this is the AI Citations view: Citation Rate, Citation Share and the domains cited for you and for each competitor.
Closing the citation gap
| Gap found | What to do |
|---|---|
| Competitor cited from review sites | Get listed and reviewed on those platforms |
| Competitor cited from media | Pitch data or stories to those publications |
| Competitor cited from marketplaces | Complete and update your listings |
| Your own site rarely cited | Add direct-answer pages for the questions customers ask |
| Outdated pages cited | Update facts, prices and dates |
Then re-run the same questions monthly to see whether your citations and Mention Rate improve. For context, the average brand in Pranas data was named in 28.3% of unaided AI answers — see benchmarks by industry.
Frequently asked questions
How do AI models choose which sources to cite? They favour pages that answer directly, are specific, trusted, consistent with other sources and current.
How can I see which websites ChatGPT cites for my brand? Run repeated questions with web search and record cited links, or use Pranas AI Citations.
Does my own website get cited? Sometimes — but third-party sites are often cited more. Track your own-domain share.
Next step: Run a free AI Scan to see which websites AI cites for your brand and competitors. Related: AI citation optimization · Track your brand in Perplexity · How to optimize content for AI citations
About the author. The Pranas Research Team analyses thousands of AI Scans across ChatGPT, Gemini, Google AI Overview, Claude, Perplexity and DeepSeek to measure how AI search represents brands.