Most Cited Sources in ChatGPT: What the Studies Show

Which websites does ChatGPT cite most often?

Across the published studies, the same kinds of sites keep showing up near the top: Wikipedia, Reddit, large news publishers, and big retail and marketplace sites for shopping questions. The exact order changes from study to study. Profound (June 2025), Ahrefs (a November 2026 edition of its "50 Most-Cited Websites in ChatGPT") and Everything PR (a 2026 "50 Most-Cited Websites" list) each use different questions, time windows and counting rules. So treat any single ranking as a snapshot, not a fixed leaderboard.

What is a "citation" in ChatGPT?

When ChatGPT searches the web to answer you, it often attaches links to the pages it used. Those links are citations. Studies collect them in bulk: run a large set of questions, record every linked page, group the pages by website (the domain), and count.

Two details matter. First, citations only appear when ChatGPT uses search. For questions it answers from what it already learned in training, there are usually no links, so those answers never enter the data. Second, a citation is not an endorsement or proof the page was the best source. It is a record that the page was pulled into the answer.

Where do the studies agree?

Three published studies are the main references people use: Profound's research on how ChatGPT, Google AI Overviews and Perplexity source information (published 5 June 2025), Ahrefs' list of the 50 most-cited websites in ChatGPT (November 2026 edition), and Everything PR's 2026 list of the 50 most-cited websites in AI answers. I am not reproducing their percentages here. The figures shift with every refresh, and each depends on its own method. Open the original pages for exact numbers.

What repeats across this kind of research:

The pattern makes sense. ChatGPT needs pages that are easy to read, clearly about one thing, and widely trusted or widely discussed. Reference sites, discussion forums, established media and product catalogs fit that.

Why do the rankings differ between studies?

If you have seen one list put Wikipedia first and another put Reddit first, nothing is necessarily wrong. The differences usually come from five choices.

1. The query set. A study built from shopping questions will rank retail sites higher. One built from general knowledge will favor reference sites. One built from software questions will favor documentation and forums.
2. The time window. Citation behavior changes as OpenAI updates ChatGPT and its search. A June 2025 snapshot and a November 2026 snapshot are measuring different systems.
3. How citations are counted. Some studies count every link. Some count each website once per answer. Some count the share of answers that include a domain. Those methods can reorder the same data.
4. Domain grouping. Does "news" mean each publisher separately, or all publishers together? Are subdomains merged into the main site? Different choices produce different leaderboards.
5. Which product is tested. ChatGPT with search on, the API, and the free versus paid app can behave differently. Profound also compared ChatGPT against Google AI Overviews and Perplexity, and platforms do not cite the same way.

A useful habit: before trusting a ranking, look for its question count, date range and counting rule. If those are missing, the ranking is hard to rely on.

How should you read a "top 50" list?

Use it for direction, not precision. Three questions help.

• What types of sites dominate? Reference, forum, news and retail is more informative than whether Reddit is third or fifth.
• Is my industry represented? General lists are dominated by sites that cover everything. Specialist sites often win inside their own niche and never reach a global top 50.
• How old is the data? Anything more than a few months old deserves a recheck on your own questions.

How can you check which sources ChatGPT cites for your topic?

You do not need a paid tool for a first pass. This takes about an hour.

1. Write 20 to 30 questions your customers, colleagues or readers would really ask. Mix types: "what is," "best," "compare," "how do I."
2. Run each one in ChatGPT with web search on. Use a fresh chat each time so earlier answers do not influence the next.
3. Record every cited domain in a spreadsheet: one row per citation, with columns for question, domain, full URL and date.
4. Count by domain. A pivot table does this in a minute. Note both how many times a domain appears and how many different questions it appears in.
5. Group the results into types: reference, forum, news, retail, vendor sites, your own site, competitors.
6. Repeat in a month. A single run tells you little. Two runs tell you what is stable.

If you want a starting prompt, use this wording and change the bracketed part:

[Question a real customer would ask about your topic]
Please search the web and cite your sources.

Asking for sources makes search more likely, but it can also nudge the model toward citing more than it would unprompted. Run some questions with this line and some without, and note which version you used.

What does this mean for you?

If you work in marketing, PR or content, the lists show that earned coverage and community presence matter. Being discussed on Reddit, covered by established outlets, or accurately described on Wikipedia (following its rules, including those on conflicts of interest) may matter as much as your own pages. Check which of those already mention you.

If you run a small business or sell products, look at which retail and review sites appear for your product questions. Make sure your listings there are accurate and current.

If you are a researcher or analyst using ChatGPT, the lists are a reminder to open the cited links. Heavy reliance on forums and reference sites means an answer can be fluent and still rest on opinion. Verify anything you will put in front of a client or your boss.

If you are just curious, the takeaway is simple: ChatGPT leans on sources that are large, readable and widely referenced. Treat its citations as a starting point for reading, not a final answer.

Could you build your own citation tracker?

A spreadsheet works for one round of checking. It gets tedious when you want to repeat the test monthly, compare topics or share results with a team. A small internal tool fits that job: a form for logging questions, a table of cited domains, and a chart showing which sources rise or fall over time. With Remy, you can describe that tracker in plain English and it builds and deploys it as a working web app with a database and sign-in, so your team uses one shared page instead of passing files around. You still run the questions and record the citations yourself, or decide how to automate that step.