How to get your SaaS recommended by ChatGPT

• 11.6%of ChatGPT recommendations credited the tool's own websiteDerivateX, 40 B2B SaaS categories, 2026
• 0citations to G2 or Capterra in the same buyer-intent sampleDerivateX, 2026
• <1 in 100chance that ChatGPT gives the same list of brands twiceSparkToro & Gumshoe, 2,961 runs

How ChatGPT decides which tools to name

When someone asks ChatGPT for “the best CRM for a 20-person sales team”, the answer comes from two places. The first is what the model learned during training. The second is a live web search: ChatGPT looks up pages, reads them and builds the answer from what they say, with links to the sources. OpenAI says it may share search queries with third-party search providers such as Bing [4], and it runs its own search crawler, OAI-SearchBot [3].

For buyer questions like “best X for Y”, the live search does most of the work, and what it finds is mostly other people's pages. In DerivateX's 2026 study of 40 B2B SaaS categories, 92.3% of the tools ChatGPT named came with a citation, but only 11.6% of recommendations credited the tool's own website [1]. The rest credited a third party.

The pages that got cited had a recognisable shape: every one used a numbered or bulleted list, 78% had a year in the title, 68% had a comparison table and 56% had an FAQ section [1].

Studies disagree on the details, and it is worth knowing why. Profound's analysis of 680 million citations from August 2024 to June 2025 found G2 among ChatGPT's most-cited domains across all kinds of questions [6], while DerivateX found no G2 or Capterra citations at all for buyer-intent software prompts. Different samples, different periods, different prompts. The conclusion that holds across both is the useful one: your own site is a small part of what ChatGPT reads about you.

1. Make sure ChatGPT can read your site

This is the only step that is fully under your control, and it is surprisingly often broken.

• Allow OAI-SearchBot in robots.txt. OpenAI is explicit: sites that opt out of OAI-SearchBot “will not be shown in ChatGPT search answers”, and changes take about 24 hours to apply [3].
• Decide on GPTBot separately. GPTBot collects content for model training. OpenAI treats the two settings as independent, so you can block training and still appear in search [3].
• Check your firewall, not just robots.txt. Bot protection on a CDN or WAF can block AI crawlers even when robots.txt allows them. Look for OAI-SearchBot and ChatGPT-User in your server or CDN logs.
• Serve the important text in HTML. Pricing, features and “who it is for” should be readable without running JavaScript.
• Get indexed by Bing. Given that OpenAI names Bing as a search provider, Bing Webmaster Tools is fifteen minutes of cheap insurance.

2. Write down the prompts your buyers ask

Keywords are the wrong unit here. Buyers ask ChatGPT full questions, usually with their situation attached. Write down 50 to 150 of them, grouped by stage:

• Category: “best [category] for [segment]”, “[category] tools for teams under 50”.
• Comparison: “alternatives to [market leader]”, “[you] vs [competitor]”.
• Decision: “is [your product] good for enterprise?”, “[your product] pricing”.

Use your sales notes, support tickets and the exact wording from demo requests. The method page describes how I build and freeze a prompt set before measuring anything.

3. Measure before you change anything

Ask ChatGPT one question once and you learn very little. SparkToro and Gumshoe had 600 volunteers run the same 12 prompts 2,961 times across ChatGPT, Claude and Google's AI. There was less than a 1 in 100 chance of getting the same list of brands twice, and roughly 1 in 1,000 of getting it in the same order [5].

What did hold steady was how often a brand appeared across many runs. The study's conclusion is that a visibility percentage across dozens of prompts, each run several times, is a reasonable metric, and that a “ranking position in AI” is not [5]. So run each prompt several times, record which brands appear, and report the share of answers that name you, with a margin of error.

4. Get onto the pages ChatGPT already cites

This is the step that moves the number most, because it targets what ChatGPT actually reads.

1. Run your prompts and open the Sources list under each answer [4].
2. Collect every cited URL in a sheet. Within a few dozen prompts, the same domains start repeating.
3. For each page, check three things: are you listed, is what it says about you current, and is the page still updated.
4. Write to the authors with something they can use: an honest comparison, current pricing, a trial account, or data they do not have. A correction to an outdated description is the easiest win.
5. Where Reddit threads are cited, take part as yourself, with your affiliation stated. Planted recommendations get spotted and removed, and they cost more trust than they win.

Expect most of the list to be smaller sites than you think. In DerivateX's data, niche and independent sites out-cited all recognised media combined [2].

5. Publish pages built to be quoted

Vendors' own content is a large share of what ChatGPT cites: in DerivateX's classification, 51.1% of the 188 citations it classified went to a software company's own content, usually a “best tools” or comparison post [2]. Your competitors are already writing the pages that recommend them. You can write yours, and do it more honestly.

• Category pages: “Best [category] tools for [segment] in 2026”, including competitors, with a clear statement of when each one is the better choice.
• Alternatives and versus pages for the two or three tools you lose deals to.
• Use the format that gets cited: a numbered list, a comparison table, the year in the title and a short FAQ [1].
• Keep them current. A list that still shows last year's pricing gives the model a reason to prefer someone else's.

A page that only praises your product is easy to ignore. A page that says “if you need X, choose the competitor” reads like a source.

6. Say who you are for, the same way everywhere

ChatGPT answers questions that contain a situation: team size, industry, budget, stack. It can only match you to that situation if the pages it reads describe you that way. Write one plain sentence about who you are for and what you replace, and use it on your homepage, pricing page, LinkedIn page, review profiles and directory listings.

Mentions matter more than links. In Ahrefs' study of 75,000 brands, branded web mentions correlated with visibility in Google's AI Overviews at 0.664, against 0.218 for backlinks [7]. That study covers AI Overviews rather than ChatGPT, and correlation is not causation, but it points the same way as the citation data: being talked about beats being linked to.

7. Re-measure on a schedule

Run the same frozen prompt set again after each round of changes, monthly at most. Compare shares and margins, not single answers. If your share moved by less than the margin of error, nothing measurable happened yet, and that is a normal result for the first month. Changes to cited pages can show up in search-based answers fairly quickly; what the model learned in training changes only when a new model is released.

What I would not spend the first month on

• llms.txt files. Cheap to add, but I have not seen solid evidence that they change what ChatGPT recommends. The studies are in does llms.txt work?
• Schema markup as a lever. Structured data is good hygiene; on its own it is not why a tool gets named.
• Mass-produced AI articles. They add pages nobody cites and risk your Google rankings.
• Tracking a single “AI rank”. As the SparkToro data shows, the order changes on almost every run [5].

Questions

How long does it take to get recommended by ChatGPT?

Fixing crawler access takes about a day to apply on OpenAI's side. Getting onto cited pages and seeing it in your measured share usually takes one to three months, because you depend on other people updating their pages. What the model learned in training changes only with new model releases.

Does my own website matter if ChatGPT mostly cites other pages?

Yes, for two reasons. It is where list authors and the model check your positioning, pricing and features, and your own comparison pages are part of the vendor content ChatGPT does cite. It just should not be the only place you work on.

Should I block GPTBot?

That is a choice about model training, not search. OpenAI treats GPTBot and OAI-SearchBot as independent settings, so you can block GPTBot and still appear in ChatGPT search, as long as OAI-SearchBot is allowed.

Do G2 and Capterra reviews help?

The data is mixed. One large citation analysis found G2 among ChatGPT's most-cited domains overall, while a 2026 study of buyer-intent software prompts found no G2 or Capterra citations at all. Reviews still shape how other sites describe you, so keep profiles current, but do not treat them as the main lever.

Want this measured for your category?

The AI visibility audit runs your buyers' prompts across ChatGPT, Claude, Perplexity and Google AI Overviews, lists the pages that get cited, and turns it into a prioritised backlog. The scope is free and comes within one working day.

Get a free scope →

Sources

Checked on 31 October 2026. Figures are quoted as published; samples and periods differ between studies.

1. DerivateX — B2B SaaS AI Citation Study (published 31 May 2026)
2. DerivateX — Where ChatGPT gets software recommendations (18 June 2026)
3. OpenAI — Overview of OpenAI crawlers
4. OpenAI Help Center — ChatGPT search
5. SparkToro — AIs are highly inconsistent when recommending brands or products (27 January 2026)
6. Profound — AI platform citation patterns (June 2025, updated August 2025)
7. Ahrefs — AI Overview brand visibility factors, 75,000 brands (May 2025, updated April 2026)