How ChatGPT Decides Which Coaches to Recommend

Ask ChatGPT to recommend a leadership coach for new managers in Manchester, and it will usually give you a short list of names with a line or two about each. Ask again tomorrow and the list may be different. Ask a colleague to try it, and they may see different names again.

If you’re a coach, that short list now matters. Some prospects start there instead of on Google, and the people on it get a warm introduction you didn’t.

It can also be maddening. Plenty of coaches with fifteen years of practice, a shelf of credentials and a full diary of referrals find that ChatGPT names someone who qualified two years ago. This guide explains what is publicly known about how ChatGPT picks names, what is only inference, and what the evidence suggests you can do about it.

What happens when someone asks ChatGPT for a coach

When a person types a request like “Who are the best executive coaches for women in tech?”, ChatGPT has two ways to answer.

It can answer from memory. The model has absorbed a vast amount of text during training, and some coaches, mostly well-known authors and speakers, appear in that text often enough for the model to associate them with coaching.

Or it can search the web, read what comes back and write an answer from those pages, with links to its sources. Specific, local or current requests (a niche, a city, a budget, “this year”) are the kind that tend to lead to a search, because memory alone can’t answer them well.

Most recommendation answers mix the two. The model decides what to search for, reads the results through the lens of what it already knows, and writes a short list. OpenAI hasn’t published the rules for that blend, so the rest of this guide separates what OpenAI has said from what outside research suggests.

Training data and live search: two sources of knowledge

What the model learned in training

OpenAI runs a crawler called GPTBot that collects public web content which may be used to train its foundation models¹. Whatever the model learned in training is fixed at the point training ended. It doesn’t know about the podcast you launched last spring or the niche you moved into last year.

For individual coaches, training data is a thin source. A model may know the names of a handful of famous coaches. It is unlikely to hold reliable, current detail about a coach with a local practice, and that is exactly the detail a recommendation needs.

What it finds when it searches

ChatGPT search launched for paid users on 31 October 2024, with free users to follow. OpenAI said at the time that it “leverages third-party search providers, as well as content provided directly by our partners”, and that the search model was a fine-tuned version of GPT-4o².

In practice, that means a request for a coach often turns into one or more web searches behind the scenes. ChatGPT reads the pages that come back, picks out names and facts, and shows the pages it relied on as sources. If your name isn’t on any page the search returns, it is hard for you to appear in the answer, however good your work is.

Which search index does ChatGPT use?

This is where public information gets patchy. Here is what has been said, and by whom.

1. 23 May 2023

   Bing becomes ChatGPT’s default search

   At its Build conference, Microsoft announces that Bing is coming to ChatGPT as “the default search experience”.³

2. 31 October 2024

   ChatGPT search launches

   OpenAI says it uses third-party search providers plus content from partners, without naming the providers.²

3. 2 June 2026

   Microsoft’s Web IQ

   Microsoft launches Web IQ, a Bing-based grounding service for AI tools. In an engineering post, Microsoft says the same infrastructure powers Copilot and ChatGPT. That is Microsoft’s claim; we found no OpenAI statement confirming it.⁴

4. September 2026

   A third-party map of providers

   Peec AI, which sells AI-visibility tracking, infers from response data that ChatGPT uses its own index plus at least eight outside providers. OpenAI hasn’t confirmed it.⁵

Read Peec’s analysis as third-party inference, not a confirmed map. It reports signs of Google results gathered through scraping services, Microsoft’s Web IQ, and local data from Yelp and TripAdvisor, alongside an index OpenAI appears to have built itself⁵. Much of that rests on thin evidence: the Google finding comes from a single test site, several providers are identified only by internal labels, and the Yelp and TripAdvisor link isn’t sourced.

The practical reading is simple. ChatGPT doesn’t rely on a single index you can optimise for, and the mix can change without notice. Being findable across the open web matters more than ranking in any one search engine.

What OpenAI has said, and what it hasn’t

OpenAI’s most concrete public documentation is about its crawlers. It runs three, and each does a different job¹.

OpenAI notes that the settings are independent: a site can allow OAI-SearchBot and block GPTBot, staying visible in search while opting out of training¹. Sites that block OAI-SearchBot can still appear as plain navigational links, but not in search answers.

Here is what OpenAI hasn’t published: how ChatGPT decides which people or businesses to name, how it orders them, how much weight it gives to reviews, lists or your own website, or how it picks the sources it shows. Any article that gives you “the ChatGPT ranking factors” is presenting inference as fact. That includes the explanations below, which we’ve labelled as inference wherever they go beyond the evidence.

# Allow ChatGPT search to show your pages
User-agent: OAI-SearchBot
Allow: /

# Optional: opt out of model training
User-agent: GPTBot
Disallow: /

What the studies suggest gets people named

With no official ranking guide, the best evidence comes from companies that track AI answers at scale. Most of them sell AI-visibility software, and almost all of their findings are correlations. Treat them as strong hints, not rules.

Mentions elsewhere track visibility closely

Ahrefs studied about 75,000 brands across ChatGPT, Google’s AI Mode and AI Overviews. The factor most closely linked to AI visibility was mentions on YouTube, with a correlation of about 0.737, followed by branded mentions across the web at 0.66 to 0.71⁶. For ChatGPT in particular, classic authority signals were weaker: branded search volume correlated at 0.352 and Ahrefs’ Domain Rating at 0.266⁶. Ahrefs itself stresses that correlation isn’t causation.

For a coach, the translation is direct. Being talked about on other people’s pages and channels seems to matter more than the size or authority of your own site.

The sources ChatGPT cites shift over time

Semrush tracked about 230,000 prompts weekly from July to October 2025. In early August, Reddit appeared in close to 60% of ChatGPT’s responses and Wikipedia in roughly 55%. By mid-September, Reddit had dropped to around 10% and Wikipedia to under 20%, while LinkedIn, Medium and Forbes gained⁷. Semrush’s data didn’t show why.

That volatility is a reason not to build your whole presence on one platform. A spread of credible sources is sturdier than a bet on whichever site ChatGPT favours this quarter.

How content is written also matters

In a 2024 research paper, a team from Princeton and partner institutions tested ways of rewriting web pages to appear more often in AI-generated answers. In their benchmark, the best methods raised visibility by up to 40%, though results varied a lot by subject⁸. It was a lab test rather than ChatGPT itself, but it supports a common-sense point: pages that state facts clearly are easier for AI systems to use.

Why a less experienced coach sometimes gets named first

What follows is our inference, not something OpenAI has confirmed. Each point is a reading of the findings above, not a tested result.

1. 1

   Their pages match the question

   If someone asks for a career coach for nurses in Leeds, a website that says “career coach for nurses, based in Leeds” is an easy match. A senior coach whose site says “I help people unlock their potential” isn’t.

2. 2

   They appear on lists the search returns

   “Best coaches in…” articles, association directories, podcast guest pages and review sites are the kind of third-party pages that turn up in searches and carry names into answers.

3. 3

   Their details agree everywhere

   Same name, same niche, same city and the same short bio on every profile make it easy to confirm that all those mentions are one person. Conflicting details make it harder.

4. 4

   Their experience is written down

   A system can only weigh credentials it can read. Years of practice that live in your head, or in a PDF nobody can find, don’t count.

5. 5

   Chance did some of the work

   AI answers vary from run to run, so one list on one day isn’t a verdict on anyone.

If that reading is right, none of it is about who is the better coach. It is about who is easier to find, match and confirm. Seen that way, experience only helps you in AI answers once it is visible on the web, in words that match how clients ask.

The same principle sits behind how Google decides who gets a knowledge panel. We cover that in What Is a Google Knowledge Panel, and Why Do Some People Get One?, and there is no evidence ChatGPT reads Google’s Knowledge Graph directly. The overlap is in the habits, not the database.

AI answers change every time

If you’ve tested ChatGPT once and given up, this part matters.

As reported by Search Engine Journal, SparkToro and Gumshoe had hundreds of volunteers run 12 recommendation prompts nearly 3,000 times across ChatGPT, Claude and Google’s AI. One of the prompts asked for digital marketing consultants, a service category much like coaching⁹.

According to SEJ’s report, the researchers concluded that a name’s position in an AI answer means little, while how often it appears across many runs is more consistent⁹. The study wasn’t peer-reviewed, and Gumshoe sells AI-tracking tools, which the authors disclosed. It is one study, but it is a good reason to test more than once.

So when you check yourself, ask the same question several times, in fresh chats, with slightly different wording. Count how often you appear. Being named in six of ten answers tells you more than being first in one.

Four myths about AI recommendations

OpenAI documents the difference itself¹. Check which one your site is blocking before assuming either.

Search rankings still help, because a page has to be found before it can be read. They are one input among several²⁶.

What you can actually do

You can’t apply to be recommended, and nobody outside OpenAI can promise a mention. You can make yourself much easier to find, match and confirm. The steps below follow from the evidence above. None is guaranteed to work on its own.

Start by checking what ChatGPT says today

Ask the questions your clients would ask: your niche, your city, the outcome they want. Run each a few times in fresh chats and note who appears, how often and which sources are cited. Those sources are a map of the pages that shape answers in your niche. Our free AI Mention Checker runs a similar check for you.

Then work through these

1. 1

   Say exactly who you help, where

   Put your niche, your typical client and your location in plain words on your homepage and About page. Use the phrases clients use, not internal jargon.

2. 2

   Write your experience down

   Years of practice, accreditations, notable clients you can name, books, talks and results belong on your site in clear, factual sentences.

3. 3

   Make your details match everywhere

   Same name, title, niche, city and photo on LinkedIn, coaching directories, podcast pages and speaker bios. Our guide to how Google reads your LinkedIn headline shows why the wording matters.

4. 4

   Get onto the pages search returns

   Credible directories and associations in your field, reviews from real clients, “best coaches” roundups, interviews and guest articles. A few strong ones beat many weak ones.

5. 5

   Show up on video and audio

   Ahrefs found YouTube mentions had the strongest link to AI visibility. Podcast guest spots with show notes on the web give AI search a page to read.

6. 6

   Don’t block the search crawler

   Make sure robots.txt allows OAI-SearchBot, so ChatGPT search can show your pages.

7. 7

   Add structured data

   Person markup on your About page states your name, role and profiles in a form machines read reliably.

8. 8

   Recheck every few months

   Sources and answers shift. Re-run your questions and look at which pages are now being cited.

Step three is where most coaches lose ground without noticing. An old speaker bio with your previous niche, a directory listing in a city you left and a LinkedIn headline that says something else all compete with your current story. Our name search checklist walks through the places prospects, and search systems, look first. If structured data is new to you, Structured Data in Plain English explains it without the jargon, and How Google Reads Your LinkedIn Headline covers the profile most coaches update least.

Where to start

ChatGPT doesn’t pick coaches on merit, and it doesn’t follow a published ranking. As far as anyone outside OpenAI can tell, it names the people it can find, match to the question and confirm across several sources, with a fair amount of chance on top.

That is mostly good for experienced coaches, because it rewards things you already have and can write down. Start by asking ChatGPT the questions your clients ask, or run the AI Mention Checker. Then make your niche, location and experience consistent everywhere they appear. If you’d like help, see how we work with coaches and consultants, or book a free audit.

Keep reading

Knowledge PanelsWhat Is a Google Knowledge Panel, and Why Do Some People Get One?Shobin K Jose · 19 Oct 2026 · 23 min readGuidesThe Name Search Checklist: 12 Things Prospects See Before They CallShobin K Jose · 6 Oct 2026 · 18 min readGuidesStructured Data in Plain EnglishShobin K Jose · 18 Sep 2026 · 18 min read