Engine deep-dive

How ChatGPT decides what to recommend

OpenAI publishes no ranking formula, but it does publish its rules on citation, sources and ads. This guide reads those documents against our own scans. Updated October 2026.

A chat answer listing a shortlist of vendors, with its sources underneath ChatGPT

Key findings

  • ChatGPT has two modes with completely different recommendation behavior: base (training data only) and browsing (real-time retrieval)
  • Google rankings do NOT determine ChatGPT recommendations. Different systems, different signals.
  • Structured data (FAQ schema, Organization schema) significantly increases citation probability
  • Third-party pages get cited alongside your own. Which ones ChatGPT reaches for varies by category. Measure the sources on your own buying queries, not a fixed league table of directories.
  • Most B2B service queries return generic, non-committal answers. The white space is enormous.

Two ChatGPTs, two recommendation systems

This is the most important thing to understand. ChatGPT operates in two distinct modes, and each produces different recommendations for the same query:

Base model (no browsing)

Uses training data only, with a knowledge cutoff. Cannot discover new companies or updated content. Tends to recommend well-known brands that appeared often in training data.

Browsing mode (search enabled)

Retrieves fresh web data on every query: Retrieval-Augmented Generation (RAG). Can recommend recent content, new companies and updated information. Here your content and schema directly shape what gets recommended.

The 5 signals that matter

Drawing on published industry analyses of AI search citations across B2B service categories, five signals consistently correlate with being recommended:

1. Structured data on your website

ChatGPT with browsing parses structured data more reliably than marketing copy. FAQ, Organization and Service schema tell it what you do, who you serve and what sets you apart. According to SE Ranking, about 71 percent of pages cited by ChatGPT include structured data, far more than pages without it.

2. Third-party citation presence

ChatGPT reads more than your website. It retrieves directory profiles, independent comparison articles and news coverage. Reddit is retrieved constantly and chosen rarely. One independent study of OpenAI grounding data found it cited 0.61% of the times it was a candidate (Dan Petrovic).

3. Content specificity

Generic service pages ('We provide IT solutions') almost never get cited. Specific, query-matching content does. 'IT Support for Dental Practices: What to Look For' is far likelier to be cited for 'best IT support for dental clinics' than a generic page. The page must answer the exact question asked.

4. Recency signals

In browsing mode, ChatGPT prefers recent content. Articles dated 2026 outperform identical content dated 2023. This is especially true for comparison and recommendation queries where users expect current information.

5. Entity recognition

ChatGPT recommends companies it can identify as entities. That means a Wikipedia page, a Wikidata entry, a complete LinkedIn company page and consistent naming everywhere. If ChatGPT cannot confidently identify you as a real entity, it defaults to companies it can verify.

What does NOT matter

The white space opportunity

For many B2B service queries, ChatGPT gives generic, non-committal answers. It often says "there are several good options" and names no company. That gap is the opportunity.

Ask "best MSP for healthcare compliance" and ChatGPT often lists well-known general IT companies, but no healthcare-specialist MSP. The specific query has no specific answer. The first MSP to publish a thorough, schema-marked healthcare IT compliance page is well positioned to own it.

The lead time is real but short. Search-grounded retrieval re-reads the live web, so structured-data changes can show up in cited answers within weeks, not quarters.

What you can do today

  1. Add FAQ schema to your top service pages. 5 questions matching the queries your buyers ask ChatGPT, under 80 words per answer. The single highest-ROI action.
  2. Create one vertical-specific page. "[Your Service] for [Specific Industry]": 2,000+ words with FAQ schema. Target a query where ChatGPT now gives a generic answer.
  3. Get your directory profiles right. Clutch for services, G2 for software. Directory profiles are read on vendor-selection queries; which one ChatGPT reaches for in your category is something to measure, not assume.
  4. Check your robots.txt. Allow OAI-SearchBot: OpenAI states sites opted out of it will not be shown in ChatGPT search answers. GPTBot is the separate training crawler and can be blocked independently.

How ChatGPT retrieves and cites, from OpenAI's own documents

  • From the vendor One crawler decides whether you can be cited at all. OpenAI states that sites opted out of OAI-SearchBot will not be shown in ChatGPT search answers. GPTBot, the training crawler, is a separate switch: you can refuse training and stay eligible for search. OpenAI: its crawlers
  • From the vendor Search answers carry their sources. ChatGPT search returns answers with links to the pages it drew on, and OpenAI names publisher partners whose news feeds those answers. A cited page is a page a buyer can click. OpenAI: ChatGPT search
  • From the vendor Ads sit beside the answer, never inside it. OpenAI states ads are clearly labelled, kept separate, and do not influence what ChatGPT says. There is no paid route onto the shortlist. OpenAI: ChatGPT ads
  • Our read It cites pages, so the page that answers wins. Search retrieves a page per need, and a homepage that describes everything answers nothing in particular. One page per buying query, titled the way the buyer asks it, gives search something to cite.
  • Our read We force the search on every query. The ChatGPT a buyer opens decides for itself when to search. We force it every time. Our reading shows what ChatGPT finds when it looks: the version you can change this month.

What moves your Tofu Index and your Bofu Index on ChatGPT

Your AI Visibility Index is two numbers with two jobs.

  • The Tofu Index asks whether ChatGPT puts you in the consideration set: the share of buying queries where it names you.
  • The Bofu Index asks whether it picks you: the share of head-to-head comparisons, you against a named rival, where it chooses you.
  • A free scan reports your Tofu Index on every engine we query. Fix and Dominate add the Bofu Index.
Levers on ChatGPT, and which index each one moves. Our read of the mechanism, not a measured effect size.
Lever Tofu Index: are you considered Bofu Index: are you picked
Allow OAI-SearchBot Without it you are ineligible for ChatGPT search answers, so no live page of yours can be cited. A rival whose pages can be read wins every comparison by default.
One page per buying query Gives search a page that answers the exact query, which is what gets you onto the shortlist. Little on its own. Being found is not being chosen.
An honest comparison page per named rival Adds you to the alternatives and versus answers. The strongest lever here: it hands ChatGPT your side of the trade-off, in your words, on the exact comparison it is asked to make.
Directory profiles with specific reviews Puts you on the third-party lists search retrieves for vendor queries. A review naming the outcome is what an engine quotes when it explains a pick.
Earned coverage OpenAI's publisher partnerships make news a direct input to search answers. Coverage that states what you are best at gives the engine a reason to prefer you.

What our own scans measured

We can put a number on how often ChatGPT names anybody at all.

In that study, run before every engine searched the web, Perplexity led at 7.4% and the field ran 0.4% to 4.0%, among the same firms that suspected a problem. Absence is the ordinary condition. A case study claiming your category is already saturated describes a handful of queries, not the general state of things.

A second measurement matters more here than the headline rate.

Read the MSP AI Visibility Report 2026, which shows the method and the names we could not verify.

How we count, because the denominator does the work

These rates cover buying queries only: what a buyer types when choosing a vendor, before they know who you are.

The buyer who already knows your name is not the buyer you are missing. A number that counts them measures your own marketing back at you.

What you get, and what it costs

ChatGPT is one of the 5 engines every plan queries, including the free Track plan. Engines are not plan-gated. What a plan changes is how many queries you get per scan, how often the scan runs, whether the scan adds the head-to-head comparisons behind the Bofu Index, and whether we draft the pages that close the gaps it finds.

Plan Price Queries per engine Scans Drafts a month
Track Free 5 1 a month Ideas only
Fix $99/mo 10 4 a month 10
Dominate $499/mo 25 4 a month 25

Rule is portfolio level and sales-assisted. Full pricing · How the scan works

See what ChatGPT says about your company right now

Check your AI visibility free

Frequently asked questions

Does ChatGPT use Google rankings to decide recommendations?

No. ChatGPT and Google use different systems, so a company can rank number 1 on Google and be absent from ChatGPT recommendations. ChatGPT relies on its training data, retrieval-augmented generation when browsing is on, and structured data it can parse from websites.

Can you pay to be recommended by ChatGPT?

No. ChatGPT now carries ads, and OpenAI states they are clearly labelled, kept separate from answers, and do not influence what ChatGPT says. The shortlist inside an answer is not for sale. It comes from what search retrieves and what the model already knows.

How often does ChatGPT update its recommendations?

ChatGPT with browsing enabled retrieves fresh data on every query. The base model without browsing uses training data with a knowledge cutoff. New content can appear in browsing-enabled responses within days of being indexed.

What content format does ChatGPT prefer to cite?

Industry analyses find ChatGPT most often cites clearly structured content: FAQ pages, comparison articles, listicles with specific criteria and schema-marked pages. Roughly 71 percent of pages cited by ChatGPT include structured data, according to SE Ranking. Unstructured marketing copy is rarely cited.

How do I actually get ChatGPT to recommend my company?

Be the well-structured, corroborated answer to the specific buying query. Three things move it: a page that plainly answers the query with structure and schema; presence on the sources ChatGPT trusts and retrieves (G2, Capterra, Reddit, industry roundups); and consistency, so the signal repeats. There is no submit button: you assemble enough evidence that naming you is the obvious answer.

Why does ChatGPT recommend my competitors instead of me?

Because it has more to go on for them: clearer content matched to the query and more third-party corroboration, so they are the safer answer. It is a readout of the evidence, not a fixed ranking. Find the buying queries where a competitor is named and you are not. Then build the page and the outside proof that answer those queries better.

How do I know if ChatGPT's crawler visited my site?

Check your server logs or CDN bot analytics for GPTBot and OAI-SearchBot. GPTBot gathers training data and OAI-SearchBot feeds ChatGPT search; seeing them confirms your pages are reachable, the precondition for being used. If they are absent, check that robots.txt and bot rules are not blocking them. A surprising number of sites quietly exclude the exact crawlers they want reading their content.

Does it matter whether ChatGPT is browsing or using training data?

Yes, a lot. With search or browsing on, ChatGPT retrieves live pages, so fresh, structured content can be cited within days of being indexed. Without it, ChatGPT answers from training data whose cutoff changes only on model updates. Publish for both: precise, live-retrievable pages for browsing, and durable third-party corroboration for training.

Sources and further reading

Other engine deep-dives: Claude · Perplexity · Google AI Mode and AI Overviews · Gemini · G2 · Clutch · YouTube