What sources does ChatGPT use when it recommends products in my category?
Quick Answer: Run 10 to 20 buyer prompts in ChatGPT and read the cited pages, or let a tracker like Welt Analytics collect them daily. If your domain is missing, publish pages for those prompts. If a rival's or publisher's page is cited, work on that source. This matters to any e-commerce, SaaS or agency team whose buyers ask AI assistants.
Key takeaways
- Keep a dated log of the domains cited each time ChatGPT recommends products in your category, so you compare weeks instead of single runs.
- Assign each cited page an owner on your team who rereads it after every price, spec or range change.
- Fix wrong facts on pages that are already cited before you write new pages from scratch.
- Report the source list next to your visibility numbers, so a drop has a named page behind it.
What sources does ChatGPT use when it recommends products in your category?
When ChatGPT searches the web, it draws on specific pages and shows them as citation links you can open and log. OpenAI's academy page puts it plainly: "ChatGPT search allows ChatGPT to pull in the latest information from the internet directly into your conversations." The same page says search takes answers beyond ChatGPT's built-in training knowledge, to topics like market trends and competitor activity [1]. OpenAI's help center also documents an offline web search that uses OpenAI's indexed and cached web content for eligible ChatGPT workspaces [2]. That is one more reason to read the links instead of guessing where an answer came from.
In a product category, these are the kinds of pages worth looking for in the citations:
- Retailer and brand product pages, including yours and your competitors'.
- Comparison and review articles that rank or contrast products.
- Forum threads where buyers ask each other for picks.
- Publisher buying guides and roundups.
This is a list of what to look for. Only your own prompts show which kinds appear in your category and how often.
A sources report is the list of pages cited next to each answer, collected prompt by prompt and day by day. One run is a snapshot. A report shows a trend.
How can you see which pages ChatGPT cites for your category?
You see them by running the prompts your buyers ask and opening every citation link. OpenAI's own instruction is: "Click the citation links in the response to review the original sources." A small globe icon next to the response shows that search was used for that answer [1].
- Write 10 to 20 prompts the way a buyer would ask them, such as
best trail running shoes for wide feetorwhich CRM suits a 10 person sales team. - Open a new ChatGPT chat for each prompt and run it.
- Check for the globe icon. If there is none, the answer was not built with search, so log that prompt as uncited.
- Click each citation link. Record the URL, its domain, and every brand the answer names.
- Count how many prompts cite each domain.
- Repeat steps 2 to 5 on a second day, because a single run is one snapshot.
Expected result: a list of cited domains, each with a count, that shows whether your own domain appears and which competitors or publishers shape the category answer.
To read the list, check where your domain lands. Your domain cited on several prompts on both days is a good result. If your domain is absent, your pages are not being cited for those prompts yet. That answers the related question of which pages on your site ChatGPT cites. A list made up only of competitor and publisher domains means those pages currently frame your category for the buyer.
One mistake skews everything: testing prompts that contain your brand name. A prompt that names your brand does not show what a buyer who asks openly sees. Write the prompts without your brand, then add a separate brand prompt only if you need it.
A tracker runs the same check on a schedule and keeps the history, which a spreadsheet struggles with past a few dozen prompts.
What should you do with the sources you find?
Act on the source that is cited, because the citation shows which page currently answers your buyer. Use one row per case:
| What the source list shows | How to confirm | Action |
|---|---|---|
| Your own pages are cited | Your domain appears in the log on both test days | Keep them current and add the questions the prompts still ask |
| A competitor's comparison article is cited | The link opens a competitor-owned page that compares products | Publish your own comparison for the same prompt, using facts you can source |
| A forum thread is cited | The link opens a community or forum thread | Answer the question in the thread honestly and link only where it helps |
| A publisher guide is cited | The link opens a buying guide or roundup | Check that your product is listed and correct, and ask for a correction if not |
| No page cited for your brand | Your domain never appears in the log | Publish a clear category page and article on your own domain, then re-check |
These actions change what is available for ChatGPT to find.
For the competitor rows, it helps to show colleagues or a client the evidence before you start. This guide on how to show which competitors AI assistants recommend walks through presenting that comparison.
Do all AI engines use the same sources?
Do not assume ChatGPT's source list holds for the other engines. Check each engine your buyers use. If Gemini or Google AI Mode matters to your category, run the same prompt set there. A tool that tracks how ChatGPT, Gemini and Perplexity talk about your ecommerce saves running every engine by hand.
The most common error is treating one run in one engine as the full picture. OpenAI describes ChatGPT search as one that "Prioritizes the latest available information; ideal for breaking news or time-sensitive data." [1] A list that favours fresh pages can change from one day to the next. So repeat the prompts on several days and in each engine you care about before you rewrite a page or chase a publisher.
Cadence matters too. Welt Analytics checks ChatGPT, Gemini, Google AI Overviews and Google AI Mode daily, and Claude and Perplexity weekly [3]. A weekly-checked engine cannot show a one-day change. Read Claude and Perplexity results as weekly snapshots, and judge day-to-day moves only on the four daily engines.
Small point: a source that vanishes for one day and returns the next is usually noise.
When does a tracking tool beat checking by hand?
A tool is worth it when your prompt list grows or you need history. If the hand check takes more than an afternoon, or you need to know when a source first appeared, a tracker does that work. Our comparison of the best tool to track my products in ChatGPT, AI Overviews and AI Mode covers the selection itself.
Welt Analytics reports a brand's visibility, share of voice, position, sentiment, sources and competitors in AI answers. It scans your site and sets the first prompts. As of 2026-11-13, Starter costs 89 EUR per month for 50 prompts and 1 brand across all 6 engines, with unlimited seats. Listed prices exclude VAT, and there is a 7-day free trial with no card required [3].
On the Growth and Pro plans, Welt Analytics also writes articles and publishes them on a blog on your subdomain. That covers the article half of the last row in the table. On ownership, its site says: "Every article we publish belongs to you." [3]
To see the pages AI answers cite for your category, collected each day for the daily engines, start a trial at Welt Analytics.
Questions
Can I make ChatGPT cite my product pages?
Nobody can promise that. You control two things: clear, factual and current pages on your own domain, and corrections to third-party pages that mention you. After either change, run the prompt set again and see whether the cited domains shift.
How often should I re-check which sources ChatGPT uses?
Several times across a week at minimum, since single answers vary. For history at day level, a daily tracker covers ChatGPT. A weekly-checked engine such as Claude or Perplexity cannot show a change that happened on one day.
Do I need a paid tool to see cited sources?
No. The manual check above is enough for a small prompt set. A tool earns its place once the list grows or you need a record over time, and a free trial lets you test one before paying.
Which AI tools let an agency team have unlimited seats?
Welt Analytics includes unlimited seats on Starter and lists them on Enterprise. For how agencies run this across several clients, see [how agencies track client brands across ChatGPT, Gemini, Claude and Perplexity](https://blog.weltanalytics.com/how-do-agencies-track-client-brands-across-chatgpt-gemini-claude-and-perplexity).