ChatGPT visibility
ChatGPT ranking factors: what is documented, what is observed, what is just claimed
Published: · Last reviewed: · By Nick
Short answer
There is no official list of ChatGPT ranking factors, because ChatGPT does not rank pages the way Google does. It writes answers, and brands appear in them as a side effect of what the model learned and what it read. Still, some things are documented by OpenAI: which crawlers exist, that search mode reads live pages, that ads are separate. Some are consistently visible in saved answers: being on the sources it cites, being described the same way everywhere, answering questions directly. And some are repeated everywhere without evidence: llms.txt, keyword density, schema as a direct factor. This page sorts eighteen factors into those three groups, so you spend effort where it is justified.
Key takeaways
- Only four factors are documented by OpenAI, and they are all about access, not preference.
- Most of what moves visibility is observed, not documented: presence on cited sources, consistency, quotability, freshness.
- Several popular "factors" have no evidence behind them. They are harmless, but they should not be your first week of work.
- Model-only answers and search-mode answers respond to different factors on different timescales.
- The only ranking signal you can measure directly is your own rate of appearance across repeated prompts.
Prompts this guide answers
- “ChatGPT ranking factors”
- “how does ChatGPT decide which businesses to recommend”
- “ChatGPT SEO ranking factors”
- “ChatGPT local ranking factors”
- “Is there a list of what makes ChatGPT recommend one company over another?”
- “Does schema markup or llms.txt make ChatGPT rank my site?”
Why is "ranking factors" the wrong model, and still a useful question?
Google scores pages against a query and orders them. ChatGPT predicts the next word. When it writes "three agencies worth looking at are...", the names that follow are the ones most strongly tied to the context in its training and, in search mode, in the handful of pages it just read. There is no index of brands with scores attached. The how to rank in ChatGPT guide tells you what to do about that. This page explains why.
The question is still useful, because the inputs to that prediction are partly under your control: what is written about you, where, how consistently, and whether the model can reach it. Calling those inputs "factors" is fine as long as you remember they are probabilistic. Every factor below is graded Documented (OpenAI says so), Observed (visible across saved answers, with a check you can run) or Claimed (common advice, no evidence we can point to).
The 18 factors, graded
One row per factor. "Applies to" says whether it matters for answers from the model alone, for search-mode answers, or both.
| # | Factor | Grade | Applies to | How to check it yourself |
|---|---|---|---|---|
| 1 | OAI-SearchBot can crawl your pages | Documented | Search mode | robots.txt, CDN bot rules, server logs |
| 2 | GPTBot allowed (training data) | Documented | Model-only, future models | robots.txt |
| 3 | ChatGPT-User can fetch a page on request | Documented | Both | Ask ChatGPT to summarize a URL on your site |
| 4 | Main content present in HTML without JavaScript | Documented (crawler behaviour) | Search mode | View source; compare with the rendered page |
| 5 | Your brand appears on the sources ChatGPT cites for your category | Observed | Search mode | Run 20 prompts, list the citations, check your presence |
| 6 | Consistent name, category and description across the web | Observed | Both | Audit your site, LinkedIn, directories, review sites |
| 7 | A plain one-sentence statement of what you are, high on the page | Observed | Both | Ask ChatGPT "what does this company do" with your URL |
| 8 | Pages that answer the buyer’s question directly under a question heading | Observed | Search mode | Map tracked prompts to pages |
| 9 | Presence in existing comparison and "best of" content | Observed | Both | Search your category plus "best"; are you on the lists? |
| 10 | Specific, checkable facts (numbers, dates, locations) | Observed | Both | Read your own pages for vagueness |
| 11 | Freshness of time-sensitive pages | Observed | Search mode | Date audit of pricing and spec pages |
| 12 | Review-platform and community presence (G2, Capterra, Reddit, forums) | Observed, depends on category | Both | Citation analysis shows whether these appear for you |
| 13 | Volume of coverage over time | Observed, slow | Model-only | Brand mention count across the indexed web |
| 14 | Factual accuracy of what is already written about you | Observed (affects what is said, not whether) | Both | Compare answers with confirmed facts |
| 15 | Structured data (Organization, Product, FAQ) | Claimed as a direct factor; useful for clarity | Search mode | Rich Results test |
| 16 | llms.txt | Claimed; no evidence of use by OpenAI | — | None available |
| 17 | Keyword density, "optimizing for ChatGPT keywords" | Claimed; contradicts how generation works | — | — |
| 18 | Google ranking position | Claimed; indirect at best | Search mode | Compare your Google rank with your ChatGPT rate on the same prompts |
The documented factors (1 to 4)
OpenAI publishes the names and purposes of its crawlers and how to control them in robots.txt. GPTBot collects training data. OAI-SearchBot builds the index ChatGPT search draws from. ChatGPT-User fetches a page when a user's prompt points at it. Each respects robots.txt. The current names are in OpenAI's bot documentation.
These are access factors. Allowing them does not make you preferred. Blocking them makes you absent. The most common failure we see is not a deliberate block but a CDN or security preset that blocks "AI bots" as a category, applied by someone who never looked at the list. The second most common is content that exists only after client-side rendering, which a search crawler may not execute.
The observed factors (5 to 14)
These come from reading saved answers. The pattern is consistent enough across categories to act on, and each one has a check you can run without any tool.
Presence on cited sources (5) is the strongest observed factor in search mode. When ChatGPT cites five pages and names four brands, the brands are almost always on those pages. If you want to be named for a prompt, be on the pages ChatGPT already uses for it. AI citations explained shows how to find those pages.
Consistency (6) and the plain statement (7) govern how the model describes you. A brand described five different ways gets a muddled description or none. This is what our brand authority in AI answers service works on.
Direct answers (8), comparison content (9) and specific facts (10) govern quotability. The model lifts passages that stand alone. A question heading followed by a two-sentence answer is the most liftable unit on the web. Start from what prompts buyers ask AI. Being on the pages that compare GEO agencies, or whatever your category’s comparison pages are, is factor 9 in practice.
Freshness (11) matters only in search mode, and only for facts that change. Review and community presence (12) depends on category. For software, G2 and Reddit dominate citations. For local services, directories and maps data do. For B2B services, trade publications and lists. Check your own citations before assuming.
Coverage volume (13) is the slow factor behind model-only answers. It cannot be bought quickly. It is why established brands appear in model-only answers while new ones appear only in search mode. Accuracy (14) does not change whether you are named, but it decides whether being named helps. Reputation monitoring tracks it.
The claimed factors (15 to 18)
Structured data (15) is worth doing for clarity and for Google, and it may help retrieval classify a page. We have no evidence it changes whether ChatGPT names a brand, and we do not claim it does.
llms.txt (16) is a proposed convention for telling language models where your key content is. OpenAI has not said that ChatGPT reads it. Publishing one costs nothing and shows intent. Counting on it is a mistake. The proposal is at llmstxt.org, and our llms.txt and structured data guide shows what one looks like.
Keyword density (17) is advice carried over from early SEO. Generation does not count keywords. Stuffing them makes your pages less quotable.
Google position (18) correlates with ChatGPT visibility only because both reward clear, well-covered brands. ChatGPT search uses OpenAI's index and partners, not Google's results. Brands with strong Google rankings and weak ChatGPT presence are common. The reverse is rarer but exists.
How do the factors differ between ChatGPT, Gemini and Claude?
The "sources it leans on" row is an observation from our tracked prompts, not a published rule.
| Factor group | ChatGPT | Gemini | Claude |
|---|---|---|---|
| Crawler control | GPTBot, OAI-SearchBot, ChatGPT-User | Googlebot (search), Google-Extended (training and grounding) | ClaudeBot, Claude-SearchBot, Claude-User |
| Search grounding | Optional search mode | Grounds in Google Search; Maps and Shopping data for local and product prompts | Optional web search |
| Sources it leans on | Depends on category; review sites, lists and Reddit are common | Google index; Reddit, YouTube and Wikipedia prominent | Depends on category; tends toward documentation and reference pages |
| Where to read more | This page | How to rank in Gemini (below) | How to get found on Claude (below) |
A checklist you can run this week
Eight items. The full GEO audit checklist has 26.
- Read robots.txt and CDN bot settings. Confirm GPTBot, OAI-SearchBot and ChatGPT-User are allowed.
- View source on your three most important pages. Confirm the main text is in the HTML.
- Write the one-sentence statement. Put it in the first paragraph of the homepage and every service page. Copy it to every profile.
- Run 20 buyer prompts with search on. List every cited domain. Mark which ones mention you.
- For each prompt, name the page on your site that answers it directly, or note that none does.
- Find the existing "best X" and "X vs Y" pages in your category. Note whether you are on them.
- Date-check every page with a price or spec.
- Compare the saved answers with your confirmed facts. Log every error with its probable source.
FAQ
Questions and answers
Does ChatGPT have ranking factors like Google?
Not in the same sense. It has inputs that make a brand more or less likely to be written into an answer. Four are documented by OpenAI. The rest are observed or merely claimed.
Does ChatGPT favour big brands?
Model-only answers reflect coverage volume, which favours established brands. Search-mode answers can name a new brand that is on the pages it read. Being new is a disadvantage in one mode and not in the other.
What are ChatGPT's local ranking factors?
For local prompts, search-mode answers lean on directory and review pages for the area. Consistent name, address, category and hours across those sources, plus recent reviews, is the observable pattern. We have not tracked local prompts at scale and say so.
Is schema markup a ChatGPT ranking factor?
Not demonstrably. It is good practice for clarity and for Google.
Does llms.txt help with ChatGPT?
There is no evidence OpenAI reads it. It is cheap to add and harmless.
How is this different from the how-to-rank guide?
That guide tells you what to do, in order. This one explains which factors have evidence behind them, so you can decide what not to do.