In Probably Genius's model, before ChatGPT names a firm the public record has to give it three things: facts it can verify, other sources vouching for the firm, and knowledge only that firm has published. We call them the three inputs, and each one maps to a record you can build.
One buyer question rarely stays one question. Google documents this for its own AI features and calls it query fan-out: the model issues several related searches before it answers. A question about a messy partnership exit can touch the problem, the specialists, the reviews and the location, and each search lands on whatever record exists.
If a familiar competitor keeps coming back for your kind of work, this is usually why. Their trail answered more of those searches. Yours can too.
What Is Documented About How ChatGPT Finds a Firm
Start with what the engines publish, because plenty of advice in this market starts somewhere else. The internals of these models are not public, and anyone quoting exact ranking weights is guessing. Three things are on the record.
Access. OpenAI's crawler documentation says OAI-SearchBot is used to surface websites in ChatGPT's search features, and that sites opted out of it will not be shown in ChatGPT search answers, though they can still appear as navigational links. Block that crawler and you have closed one of the doors ChatGPT uses to find firms.
Retrieval with sources. Perplexity's help center says it searches the internet in real time and returns answers with cited sources. When an answer engine searches, it works from whatever pages and profiles it can reach in that moment.
Fan-out. Google says AI Overviews and AI Mode may issue multiple related searches across subtopics and data sources to build a response. That is Google describing Google's products, and it is still a useful picture of why one question touches so much of your record. How does ChatGPT decide who to recommend? walks through the decision as a sequence.
One more input travels with every question: the buyer's own circumstances, as far as the buyer has shared them. The assistant works from what the buyer tells it, and the names it returns are shaped by that context.
- Check that your robots.txt allows OAI-SearchBot. OpenAI says changes take about 24 hours to apply for search.
Input One: Something It Can Verify
Probably Genius puts the whole requirement in one line. AI recommends businesses it can verify (foundation), that others vouch for (validation), and that know something it doesn't (information gain). We build all three. Foundation comes first, because the other two hang on it.
Foundation means one consistent entity: who you are, what you do, where, and for whom, described the same way on your site, your profiles and the directories that list you. The record behind it is unglamorous. A reachable website with clear service pages. A complete business profile. The same name and address everywhere.
This is where a thinner competitor often wins. Consistency is easy to read. Depth that lives in a PDF, a podcast or a partner's head is hard to read at all. A firm with less to say and a clean, matching record can look like the safer name.
Consider what a machine meets when it checks a typical firm. The website uses one name and the business profile a shorter one. An old directory listing still carries a former address. A partner site's bio describes a service the firm stopped offering years ago. Each discrepancy is small. Together they make it harder to be sure every page describes the same firm, and the competitor whose pages all agree becomes the easier answer.
- Pick one exact business name and use it everywhere.
- Give each core service its own page that says what it is, who it's for and where you do it.
Input Two: Someone Else Vouching
Validation is what other people say about you, in places you don't control. Reviews. Articles that mention you. Association and directory profiles. The referral partner who names you on their own site. Your website can describe your expertise. It can't vouch for it.
Being cited is related, and different. An assistant can quote your article as a source and still recommend someone else, because a citation shows your page was useful and a recommendation puts a name forward for the job. Why can AI cite my content and still recommend someone else? covers that split in depth.
The record behind this input is a short list of independent pages that name your firm for the specific work you want to be known for. Count them.
Validation also has to match the work. Ten reviews that say "great service" help less than two independent pages that name you for the specific problem a buyer is asking about, because the buyer's question is specific and the evidence has to meet it there.
- List the independent pages that would name you for your core work. If the list is short, that is your first project.
Input Three: Something Only You Know
Information gain is knowledge that exists nowhere else on the web until you publish it. Usually it is the judgment that happens in conversation: how you size up a case on the first call, the question clients thank you for answering, the mistake you have watched other people make and know how to prevent.
Competitors can't copy this input. It is also the easiest one to leave empty, because to you it feels obvious. Probably Genius calls that gap the Black Box Effect, and the published methodologies explain how it works.
Choosing what to publish first is a method in itself. We map every opportunity through three lenses, where AI has nobody to recommend, where the emotional formula runs hottest, and what you uniquely know, and take the open ground first.
Written well, one of these pieces does two jobs at once. It gives a human reader a real answer, and it gives an engine something specific and attributable to your name. Thought leadership and the record turn out to be the same work. A good place to start is the question clients thank you for answering.
Why This Small Channel Is Worth Getting Right
AI referrals are still a small slice of most firms' traffic, and it pays to be honest about that. In a Seer Interactive case study of a single site, AI traffic of all kinds came to 0.07% of organic search traffic. On that same site, the case study measured ChatGPT referrals converting at 15.9 percent against 1.76 percent for Google organic.
Use is broad and still growing. In Pew Research Center's survey of February 17 to 23, 2026, 44% of US adults said they use ChatGPT, including 61% of adults aged 18 to 29. Adobe's analysis of over a trillion retail visits found AI-referred traffic up 393% year over year in the first quarter of 2026. Retail and single-site figures describe direction. They don't describe your market.
On that one site, the few visitors who arrived from ChatGPT were worth far more than their volume suggested. Your own numbers will differ. The case shows only that a small channel can still carry weight.
How to Check Without Guessing
The temptation is to ask ChatGPT once and treat the answer as a verdict. Answers vary by engine, by phrasing and by day. We test the five engines your buyers actually ask, ChatGPT, Gemini, Perplexity, Claude and Grok, with the questions they actually ask, and read the result as a rate.
You can run a simpler version yourself. How can I check whether ChatGPT recommends my business? shows how, and How do I measure whether AI visibility work is actually working? explains what to track over time. For a scored read of your record, the free Recommendation Check covers 109 checkpoints across eight zones.
If the gap is real, The Answer program builds all three inputs for you. It starts with the Genius Interview, then creates 12 done-for-you thought-leadership articles every month, planned to produce approximately 60 clear Knowledge Entries, with PR distribution, citations and reviews building validation. Every piece is scored 0-100 through the Integrity Gate: nothing publishes under 80. On pricing: $2,500 setup + $1,500/month, twelve-month program.
You run the business. We make sure AI knows why to recommend it.
Frequently Asked Questions
What does ChatGPT weigh when a buyer asks it who to hire?
Probably Genius reads the record through three inputs: facts an assistant can verify about a firm, independent sources vouching for it, and knowledge only that firm has published. The model's internal weights are not public. What OpenAI does document is that ChatGPT's search answers depend on pages its crawler can reach, so the record you leave is the raw material it works from.
Why does ChatGPT name a competitor whose work is not better than mine?
Because the name follows the record it can verify, and a competitor's record may be clearer even when the work is weaker. Consistent facts, independent mentions and published expertise are easy for a machine to read. Judgment that lives in client conversations is hard to read at all. The fix is to put your consistency, your validation and your own knowledge on the record, so the better work becomes the easier one to verify.
Can ChatGPT recommend a business whose website it cannot reach?
Its search answers cannot show that site. OpenAI's documentation says sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though they can still appear as navigational links, and that robots.txt changes take about 24 hours to apply for search. If your site blocks the crawler, fixing that comes first, because every other input depends on being reachable.
Is being cited by ChatGPT the same as being recommended?
No. A citation means one of your pages was used as a source for an answer. A recommendation means your firm's name was put forward for the job. An assistant can cite your article and still recommend a competitor whose record is easier to verify. Citations are a good sign that your knowledge is useful. Recommendations also need the foundation and validation inputs in place.
How can I check, without guessing, whether ChatGPT names my firm?
Ask the questions your buyers actually ask, with their situation included, more than once and across several engines, then count how often your name appears. A single answer is an anecdote. The free Recommendation Check does this across ChatGPT, Gemini, Perplexity, Claude and Grok, and scores the record behind the answers across 109 checkpoints in eight zones.
Related Reading
- How does ChatGPT decide who to recommend?: the hub this piece accompanies, with the decision step by step.
- The Recommendation Check: the free 109-Point AI Visibility Diagnostic.
- Our methodologies: where Probably Genius publishes the ideas behind the three inputs.
Sources
- Overview of OpenAI Crawlers (OpenAI). OAI-SearchBot surfaces websites in ChatGPT's search features; sites opted out of it will not be shown in ChatGPT search answers, though they can still appear as navigational links; robots.txt changes take about 24 hours to apply for search. platform.openai.com
- AI features and your website (Google Search Central, updated December 10, 2025). AI Overviews and AI Mode may use query fan-out, issuing multiple related searches across subtopics and data sources to develop a response. developers.google.com
- How does Perplexity work? (Perplexity Help Center). Perplexity searches the internet in real time and returns conversational answers with cited sources. www.perplexity.ai
- Americans and AI 2026: Chatbots, Smart Devices and Views on Impact (Pew Research Center, June 17, 2026). Survey of 5,119 US adults, February 17 to 23, 2026: 44% say they use ChatGPT, including 61% of adults aged 18 to 29 (Pew notes the 2026 question wording also asked about other chatbots). www.pewresearch.org
- Case Study: 6 Learnings, 1 site, How Traffic from ChatGPT Converts (Seer Interactive, June 3, 2025). GA4 data from one Seer client, October 2024 to April 2025, with ChatGPT converting at 15.9% and Google organic at 1.76%, and AI traffic at 0.07% of organic. www.seerinteractive.com
- AI traffic to US retailers rose 393% in Q1, and it's boosting their revenue too (TechCrunch, April 16, 2026, reporting Adobe data). Adobe's US retail data: first-quarter 2026 AI traffic 393% above the same quarter a year earlier, across more than a trillion visits. techcrunch.com