Write down the questions your buyers ask before they know you, ask ChatGPT each one with web search on, and record every domain it cites. Sort those domains into review sites, listicles, communities, trade press and company sites. Then earn a place on each type in the way that type rewards. That is the whole method.
Most advice on this topic stops one step earlier. It tells you to be on review sites, get press, post in communities and write clear pages. All true, and all generic. What it leaves out is which review site, which list, which community and which publication matter in your market, and the only reliable way to find that out is to look at what the assistant actually cites when your buyer asks.
This note walks through the method in five steps, gives you a template for each kind of source, and ends with how to check whether any of it worked. It takes an afternoon to run the first time and an hour a month after that.
Why does ChatGPT recommend some companies and not others?
An assistant can only recommend what it has read about.
ChatGPT names companies from two places. One is memory: what the model learned about your market during training, which changes only when a new model is released. The other is the live web: the pages it finds and reads when it searches in answer to a question. Our sibling note, Remembered or retrieved: the two ways an AI assistant can name your company, covers the difference in detail.
Memory is slow to move and hard to influence directly. The live web is where you can act this quarter. When ChatGPT searches, it reads a handful of pages and writes its answer from them, and it usually shows which ones. Those citations are a list, published in plain sight, of the pages that shaped the recommendation.
OpenAI says that ChatGPT search draws on third-party search providers as well as content from its partners, and that site owners should make sure they are not blocking OAI-SearchBot, the crawler it uses to surface sites in ChatGPT search. It also says placement is never assured. So there is no shortcut and no form. There are the pages the assistant already trusts for your buyer’s question, and whether you are on them.
Those citations are a list, published in plain sight, of the pages that shaped the recommendation.
How do you find the sources ChatGPT trusts in your niche?
Ask the buyer’s question, not yours, and write down every domain in the citations.
Five steps. The first time takes an afternoon.
- Write ten buyer questions. Use the words a buyer uses before they have heard of you: category questions (“what are the best options for this, for a company our size?”), problem questions (“how do teams like ours usually solve this?”) and comparison questions (“what should we look for when choosing a provider?”). Leave every company name out, yours and your competitors’.
- Ask each one in ChatGPT with web search on, in a fresh chat with memory and personalisation off. If you have time, ask the same questions in Perplexity, Gemini and Claude with search on too. Their sources overlap but are not identical.
- Record every cited domain, one row per citation: the question, the domain, the full URL, and whether your company or a competitor appears on that page.
- Count how often each domain appears across all ten questions. The domains that come up again and again are the ones that shape answers in your niche. A domain that appears once may be noise.
- Classify each domain into one of five types: review and directory sites, listicles and roundups, communities and forums, trade press and analysts, and company sites (yours, competitors’, partners’).
Ask each question more than once if you can. Answers vary between runs, and so do citations. A domain that appears on two different days for the same question is a stronger signal than one that appears once.
Our sibling note, Does ChatGPT recommend my company? A one-hour check you can run today, covers the question-writing and the recording sheet in full. This method adds one column to that sheet, the cited domain, and one step after it, the classification.
What are the five kinds of source, and how do you earn a place on each?
Each type of source has its own gatekeeper, and each gatekeeper wants something different.
Once your domains are classified, the plan writes itself, because each type is earned in a different way. A review site wants reviews from real customers. An editor of a roundup wants a reason to add you. A community wants useful answers, not promotion. A trade publication wants something new. Your own site wants to answer the question plainly.
| Source type | How to find it | How to earn a place | How to check |
|---|---|---|---|
| Review and directory sites (for example G2, Capterra, Trustpilot, or the directories specific to your trade) | Domains in your citations that list many vendors with ratings or profiles | Claim and complete your profile; ask real, recent customers to review you; keep category, pricing and description current | Your profile appears in the cited URL, and the facts on it are right |
| Listicles and roundups (“best tools for…”, “top agencies for…”) | Cited URLs whose titles are lists; note who wrote them and when they were last updated | Contact the author or editor with a short, specific case for inclusion; offer a fact they can check; some lists are paid placements, so note which | You are named on the cited page, and the page is still cited next month |
| Communities and forums (Reddit, industry forums, Q&A sites, Slack or Discord archives that are public) | Cited threads; search the same community for your category’s questions | Answer questions helpfully under your own name, disclose who you work for, follow each community’s rules on self-promotion | Your useful answers exist in the threads that get cited, without being removed |
| Trade press and analysts | Cited articles from publications and analyst firms in your field | Offer original data, a first-hand case or an expert comment to the journalists who cover your category; answer requests for sources | You are quoted or named in a cited article |
| Company sites (yours, competitors’, partners’) | Cited pages on company domains; note which pages of competitors get cited | Publish plain pages that answer each buyer question; allow OAI-SearchBot; ask partners and integrations to list you | Your pages appear in citations for the buyer questions they answer |
Fill in the table for your own niche. In the first column, replace the examples with the domains you actually found. Most niches have three or four review sites and two or three roundups that do most of the work, and knowing which ones is worth more than any general list.
Which sources should you work on first?
Start where the gap is widest and the gate is lowest.
You will not work on every domain at once, so rank them. Two questions do most of the ranking. How often does this domain appear across your buyer questions? And are your competitors on it while you are not? A domain that appears for most of your questions and names three competitors but not you is the first one to work on.
Then weigh the effort each type tends to take.
- Your own site and your review profiles are usually the fastest, because you control them or can start today.
- Communities take steady time over months, and only work if the help is genuine.
- Roundups depend on an editor saying yes, and some are paid placements, which you should know before you pitch.
- Trade press is the slowest and the most durable, and it needs something new to say, such as original data or a first-hand case.
A reasonable first month is one review campaign with real customers, one rewritten page on your own site for your most important buyer question, and one pitch to the editor of the roundup that appears most often. That is small enough to finish and specific enough to check.
What should you fix on your own site?
Other people’s pages carry the recommendation. Your page has to be there to be read.
The assistant needs to be able to reach your pages, and to understand them when it does. OpenAI’s documentation describes three crawlers: OAI-SearchBot, used to surface sites in ChatGPT search; GPTBot, used to crawl content that may be used to train its models; and ChatGPT-User, which fetches pages when a user’s request calls for it. They are controlled separately in robots.txt, so you can allow the search crawler while blocking the training one. Check your firewall and bot protection too, since they can block a crawler that robots.txt allows.
Then check that each buyer question has a page that answers it. Say plainly who you serve, what you do, what it costs or how pricing works, and what makes you different, in text on the page rather than in images or scripts. A page that states its answer in the first paragraph is easier to quote than one that builds up to it.
Finally, make the facts match everywhere. If your review profile, your LinkedIn page, your directory listings and your own site describe you in four different ways, the assistant has four stories to choose from. Pick one description and use it consistently.
How do you know if it is working?
Same questions, same wording, same modes, once a month.
Re-run the ten buyer questions each month exactly as you asked them the first time, and record the same columns. Two changes matter. Are you now on pages that get cited? And are you named in the answers themselves? The first usually moves before the second.
ChatGPT adds utm_source=chatgpt.com to the links it sends people through, according to OpenAI, so visits from its citations can be separated in your web analytics. That tells you whether anyone clicked. Whether those visitors went on to book a call or buy is a question for your own funnel data, and it is the one worth answering, because a citation nobody clicks and a click nobody converts are both easy to mistake for progress.
Keep the counts honest. Ten questions asked once is a small sample and answers vary between runs. Write results as fractions, such as named in 3 of 10, and compare month to month rather than reading much into a single run.
You do not get recommended by asking ChatGPT to recommend you. You get recommended by being on the pages it reads when your buyer asks.
See the use case for a growth-stage SaaS company: How does a B2B SaaS company get named first, not just named?
Questions people ask next
Can I pay or apply to be recommended by ChatGPT?
Not for organic answers. OpenAI says placement in ChatGPT search is never assured and there is no submission form for recommendations. What you can do is make your site reachable by OAI-SearchBot and earn a place on the third-party pages ChatGPT cites for your buyers’ questions.
Which review sites matter most for AI recommendations?
It depends on your niche. Find out by asking your buyers’ questions with web search on and counting which review and directory domains appear in the citations. Two or three usually appear far more often than the rest.
Should I block GPTBot?
That is a separate decision. GPTBot relates to model training; OAI-SearchBot relates to appearing in ChatGPT search. OpenAI documents them as separate crawlers, so you can block one and allow the other in robots.txt.
How long before this shows up in answers?
Changes to pages that ChatGPT already cites can show up as soon as those pages are re-read. Changes to what the model remembers wait for a new model. Re-run the same questions monthly and compare, rather than expecting a fixed timeline.
Can Throughline find the sources ChatGPT trusts in my niche?
Yes. Its Visibility Report asks your buyers’ questions across ChatGPT, Claude, Gemini, Perplexity, Copilot and Google’s AI Overviews and lists the pages each answer was built from, with every number sourced and dated. It then drafts page briefs and community replies for you to approve.