Skip to main content

How ChatGPT chooses its sources (and how to become one)

By The TadaSEO team4 min read

ChatGPT picks its sources in two steps: it runs a web search when the question calls for it, then it reuses the clearest passages from the pages it read. To become one of them, your site must be open to its robots, rank well in Bing, and contain precise answers that can be quoted as they are.

This article is part of our GEO guide, which details the full method for getting cited by ChatGPT.

When does ChatGPT search the web?

ChatGPT does not browse the web for every question. When asked for a stable definition (“what is a deposit invoice?”), it often answers from what it learned during training. When the question is about something recent, local or commercial (“which accountant for a sole trader in Austin?”, “best quoting software in 2026”), it runs a search, reads a few pages and cites its sources under the answer.

Those are precisely the questions that matter to a business: the ones where a customer is looking for a provider, a product or a tool. For these questions, what counts is what ChatGPT finds at answer time.

Where do the pages ChatGPT reads come from?

Several analyses agree: ChatGPT Search relies heavily on Bing’s index, complemented by its own search robot, OAI-SearchBot. In one sample studied, 87% of ChatGPT’s citations matched Bing’s top results for the same query (Stackmatix, Yoast).

Three practical consequences:

  1. A site missing from Bing is unlikely to be cited. Many small businesses have never checked their Bing indexing. Bing Webmaster Tools can import your Search Console property in a few minutes.
  2. Your Bing ranking matters. SEO basics (precise titles, useful content, fast pages, links) therefore help with ChatGPT too.
  3. ChatGPT’s robots must be able to get in. If OAI-SearchBot or ChatGPT-User are blocked by your robots.txt or firewall, ChatGPT cannot read your page, even if Bing knows it. Our robots.txt checker tells you in seconds.

Which passages does ChatGPT reuse?

ChatGPT does not cite a whole page: it extracts pieces that answer the question. The passages it reuses have things in common:

  • They answer directly. The answer is in the first or second sentence, not after three paragraphs of introduction.
  • They stand on their own. You understand the passage without reading the rest of the page: the subject is named.
  • They contain facts. A price, a lead time, an area, a figure with its source, a product name.
  • They are structured. Lists, tables, headings phrased as questions.

The researchers who coined the term GEO measured the effect of this kind of rewriting: adding quotations, statistics and sources to a text increased its visibility in generated answers by up to 40% (arXiv).

What role does brand reputation play?

When ChatGPT answers without searching the web, or has to choose between several brands, what it “knows” about you weighs heavily. An Ahrefs study of 75,000 brands found that the factors most linked to a brand’s presence in ChatGPT were mentions on YouTube, then mentions across the web in general (Ahrefs).

In other words, ChatGPT more readily recommends a brand that others talk about: industry directories, comparisons, customer reviews, articles, videos. Your own site is not enough.

Why does ChatGPT cite a competitor instead of you?

When you ask a question and a competitor is cited instead of you, look at the sources under the answer. The most common cases:

  • The competitor has a page that answers the exact question, and you do not. For example a “Pricing” page with prices shown, when yours are hidden behind a contact form.
  • The competitor appears in a comparison or directory that ChatGPT read. It is not their site being cited, but a third-party page that mentions them.
  • Your site is not readable: blocked robots, content shown only with JavaScript, page missing from Bing.
  • Your page is old. A large share of pages cited by ChatGPT were updated recently (Frase).

Each of these cases can be fixed, and that is what makes measurement so useful: it tells you exactly what to do.

What can you do this week?

  1. Check robot access: OAI-SearchBot, ChatGPT-User and Bingbot allowed, firewall included.
  2. Check your Bing indexing with Bing Webmaster Tools and submit your sitemap.
  3. Rewrite the opening of your three most important pages so the answer fits in the first two sentences, with precise facts.
  4. Ask ChatGPT your questions and note who is cited and which sources are used. Those sources are your target list: comparisons to join, directories to complete.
  5. Repeat every week. Answers vary by person, time and place: only regular measurement under the same conditions shows a reliable trend.

To take stock quickly, run the free audit: it checks AI robot access, tags and speed on your site in 30 seconds. And to track your questions every week without spending time on it, there is ChatGPT tracking.

Sources

Be the first found on Google and cited by ChatGPT.

Run the free audit: 30 seconds, no credit card.

  • No credit card
  • Results in 30 seconds
  • 5 free credits
Audit my site