How ChatGPT finds its sources
Let us demystify how it works. ChatGPT does not "know" everything innately. It combines two sources of information:
1. Training knowledge. The GPT model was trained on a massive corpus of texts. If your company or your content was present in this corpus (press articles, Wikipedia, popular websites), ChatGPT "knows" your existence implicitly.
2. Real-time web browsing. Since 2024, ChatGPT can browse the web via Bing and its own crawler (OAI-SearchBot). When a user asks a factual or recent question, ChatGPT performs a web search and synthesises the results. This is where your optimisation matters most.
The distinction is crucial: for general questions ("what is digital marketing?"), ChatGPT draws from its internal knowledge. For specific or recent questions ("what is the best AI SEO tool in 2026?"), it uses web browsing.
According to an analysis by Authoritas (London), 73% of ChatGPT responses to commercial queries include at least one cited web source. That is your window of opportunity.
ChatGPT citation criteria
After analysing more than 5,000 ChatGPT responses on B2B commercial queries, here is what we identified at AI SOS:
| Factor | Impact | How to optimise |
|---|---|---|
| Bing indexation | Critical | Submit sitemap to Bing Webmaster Tools, verify indexation |
| Domain authority (Bing) | High | Quality backlinks, press mentions, domain age |
| Content relevance | High | Direct answer to the query in the first paragraphs |
| Content structure | Medium-High | Clear H2/H3, lists, tables, Q&A |
| Freshness | Medium | Visible update date, content updated regularly |
| Structured data | Medium | Schema.org Article, FAQPage, Organization |
| Third-party mentions | High | Being mentioned on trusted sites (press, forums, Wikipedia) |
The often-overlooked point: Bing, not Google. The majority of SEO professionals focus exclusively on Google. But ChatGPT uses Bing as its web search source. If you are not well-indexed and ranked on Bing, you are invisible to ChatGPT.
Johan Ljungberg, head of search at Precis Digital (Stockholm), confirms: "We observed a direct correlation between Bing ranking and citation frequency in ChatGPT. Sites that ignore Bing deprive themselves of a major acquisition channel."
Technical optimisations for ChatGPT
1. Bing Webmaster Tools
If you have not done so already: register your site on Bing Webmaster Tools. Submit your sitemap. Verify that all your strategic pages are indexed. This is the foundation, and 60% of the sites we audit have not done it.
2. OAI-SearchBot in robots.txt
Check that your robots.txt does not prevent OAI-SearchBot (OpenAI's crawler) from accessing your site. Many sites added AI bot blocking rules as a precaution. This is a mistake if you want to be cited.
3. llms.txt file
Implement an llms.txt file at the root of your site. It guides AI crawlers to your most relevant content and provides structured context about your company.
4. Reinforced structured data
At minimum: Schema.org Article on each blog post, FAQPage on pages with FAQ, Organization on the "About" page, Product on product pages. The more structured context you provide, the more ChatGPT can leverage your content.
Content strategy for ChatGPT
Content that works for ChatGPT has specific characteristics:
Direct answers at the start of sections. ChatGPT generally extracts the first paragraphs under a heading. If your answer is in the 4th paragraph, it will not be selected. Adopt the "Answer-First" format from the GEO method.
Precise and sourced statistics. ChatGPT loves citing precise figures. "The market grew by 47% in 2025" is much more likely to be cited than "the market grew strongly".
Structured comparisons. Queries like "A vs B" or "best [product]" almost systematically generate a web search. Comparison tables are the ideal format.
Cover conversational queries. ChatGPT generates many long and natural queries ("how do I make my site recommended by AI?"). Optimise for this type of query, not just short keywords.
Mistakes that make you invisible
Here are the most common mistakes we observe:
- Blocking OAI-SearchBot in robots.txt. You are making yourself literally invisible.
- Ignoring Bing. ChatGPT relies on Bing, not Google. If your site is not submitted to Bing Webmaster Tools, you are out of the game.
- Too commercial content. ChatGPT avoids overtly promotional content. 100% sales pages are never cited. Prioritise informative content with subtle CTAs.
- No structured data. Without Schema.org, your content is raw text for crawlers. You lose machine context.
- Outdated content. A 2022 article that has not been updated will be ignored in favour of an updated competitor. Add visible update dates.
Are you cited by ChatGPT?
We test your visibility in ChatGPT on your 20 strategic queries. Results in 48 hours.
FAQ
Does ChatGPT always cite its sources?
No, not always. ChatGPT cites its sources only when it performs a web search (browsing mode). For responses based on its internal knowledge, it generally does not cite specific sources. That is why it is important to be present both in training data (via press mentions, Wikipedia) and in web results (via Bing/AEO optimisation).
Can you pay to appear in ChatGPT?
No. There is no (yet) advertising system in ChatGPT responses. The only way to appear is organic: quality content, domain authority, and technical optimisation. This is one of the most democratic aspects of AI visibility.
What is the difference between ChatGPT and Perplexity for citations?
Perplexity systematically cites its sources with inline numbers, making them very visible. ChatGPT cites its sources at the bottom of the response or with integrated links, in a less prominent way. Perplexity uses its own index + Google, while ChatGPT uses Bing. The optimisation strategies are therefore slightly different.
My site is blocked by OpenAI. What should I do?
Check your robots.txt file and remove the directives blocking GPTBot and OAI-SearchBot. If you had intentionally blocked these bots, know that this makes you completely invisible in ChatGPT responses with web browsing. After unblocking, ChatGPT should re-crawl your site within a few days.
How long does it take to appear in ChatGPT?
For web browsing (citations from existing sites): 2 to 4 weeks after optimisations. For the model's internal knowledge: this depends on OpenAI's re-training cycles, generally every few months. Start with web browsing, it is the fastest lever.
Can ChatGPT harm my company's reputation?
Yes, this is a risk. If negative sources dominate for your brand, ChatGPT may synthesise them in its responses. That is why proactive management of your online reputation is even more critical in the AI era. Produce positive and authoritative content to drown out negative sources.
Conclusion
ChatGPT is not a gadget. It is a fully-fledged acquisition channel, with 21% market share in Europe. Companies that make themselves visible there now are taking a considerable lead over competitors who continue to ignore this channel.
The good news: the techniques are known, documented, and actionable. The bad news: every month you wait is another month your competitors reinforce their position.
The rules have changed. ChatGPT is the new playing field.