When ChatGPT answers a question with sources attached, it has run one or more web searches, read a handful of pages and written an answer from what it found. Which companies it names depends on three things: whether your pages can be reached, whether they answer the question that was searched, and whether other sources back them up.
OpenAI does not publish a ranking formula, and anyone who claims to know the exact weighting is guessing. What can be said with confidence comes from OpenAI's own documentation and from watching how answers change over time. This guide sticks to that.
Two sources of knowledge
ChatGPT answers from two places. The first is what the underlying model learned during training, which has a cut-off date and reflects whatever the web said about a subject up to then. The second is live search, which ChatGPT uses when a question needs current or specific information, or when the user asks it to search.
For commercial questions such as "which agency should I hire" or "best software for X", ChatGPT frequently searches. That is the part you can influence most directly, because it reads your pages as they are today.
The crawlers involved
OpenAI documents three crawlers, and they do different jobs:
- OAI-SearchBot finds and indexes pages so they can appear in ChatGPT's search results. OpenAI states that sites which block it will not be shown in ChatGPT search answers, though they may still appear as plain links.
- ChatGPT-User visits a page when a user's request needs it, for example when ChatGPT opens a link during a conversation.
- GPTBot collects content that may be used to train future models.
These are controlled separately in robots.txt. A company can block GPTBot from training while still allowing OAI-SearchBot, which is the setting many businesses want. The common mistake is a blanket block, often switched on by a hosting provider or bot protection service, that stops all three.
What happens when ChatGPT searches
A single question is usually turned into several searches. Ask for the best payroll software for a small UK charity and the model may look separately for UK payroll software, charity-specific features, pricing for small organisations and reviews. Each search returns its own set of pages, and the final answer is assembled from whichever brands appear across them.
This is why ranking for one main phrase is not enough. Our piece on query fan-out covers the effect in more depth.
What makes a page worth quoting
From the answers we track for clients, the pages ChatGPT draws on share some recognisable traits:
- They answer the searched question directly. A clear statement near the top, such as who a service is for and what it includes, is easier to use than a page that builds up to its point.
- They contain specifics. Locations, sectors served, prices or price drivers, named people, dates and concrete examples give the model something it can repeat with confidence.
- They are readable without JavaScript. If the content is only rendered in the browser, the crawler may receive an empty shell.
- They agree with other sources. When a company's own description matches what directories, reviews and articles say, the claim is easier to trust. Conflicting facts across sources tend to lead to hedging or omission.
Why third-party sources matter so much
For "best" and "which should I choose" questions, ChatGPT often cites lists, comparison articles, review platforms and forum threads rather than company websites. A company's own site tells the model what the company claims. Independent sources tell it whether others agree. In most categories we work in, the set of third-party sites that keeps being cited is small enough to map, and being accurately represented on those sites is one of the most effective things a company can do.
Practical steps
- Check robots.txt and your firewall for OAI-SearchBot and ChatGPT-User. Decide separately about GPTBot.
- Load your key pages with JavaScript disabled and confirm the main content is there.
- Rewrite service pages so the first paragraph says plainly what you do, for whom and where.
- Ask ChatGPT your buyers' questions with search on and note the sources it cites. Those are the sites to be present on.
- Make your company facts identical across your site, LinkedIn, directories and review profiles.
Frequently Asked Questions
Does ChatGPT use Google search results?
ChatGPT runs its own search using OpenAI's crawler and third-party search providers. Its results overlap with Google's but are not the same, so a strong Google ranking does not guarantee a mention in ChatGPT.
Can I pay to be recommended by ChatGPT?
Organic answers cannot be bought. OpenAI has begun testing clearly labelled ads in ChatGPT, which are separate from the answer itself.
Should I block GPTBot?
That is a business decision about whether your content may be used for training. Blocking GPTBot does not remove you from ChatGPT search, which relies on OAI-SearchBot.
How quickly does ChatGPT pick up changes to my site?
Live search reads pages as they are when crawled, so changes can appear within days or weeks. What the model learned in training only changes when a new model is released.