How to rank in ChatGPT: replace the ranking myth with tests
A practical B2B guide to ChatGPT visibility: crawler policy, source strategy, prompt testing, citation analysis, and honest reporting.
There is no single, public ChatGPT ranking that works like a fixed search results page. Responses depend on the user’s prompt and the product behavior available in that conversation. Some responses use web search and show sources; others may answer without a visible web citation. The sensible objective is therefore measured visibility for a defined set of buyer prompts, not a universal rank.
You can improve the source conditions you control: access policy, useful pages, factual consistency, evidence, and internal discovery. You can then run controlled tests. You cannot guarantee that ChatGPT will retrieve, mention, recommend, or cite your company.
Define what “ranking in ChatGPT” means for your team
Before doing any optimization, decide what outcome counts. Teams often mix four different observations:
- Mention: the response names the brand without a source link.
- Citation: the response links to an owned page as a source.
- Recommendation: the brand appears as an option for a stated use case.
- Position: the brand appears first, second, or later in one sampled response.
These are not interchangeable. A mention can be inaccurate. A citation can support a narrow fact rather than an endorsement. First position in one generated list is not a stable platform-wide rank.
Write a measurement statement such as: “For 40 fixed US-English prompts, measure brand mentions, owned-domain citations, factual accuracy, and qualified referral visits during a four-week test.” This defines the denominator and stops a visibility tool’s proprietary score from becoming the strategy.
Leaf’s guide to measuring AI search visibility explains how to keep these outcomes separate.
Review OpenAI crawler controls deliberately
OpenAI publishes user-agent and crawler information in its official bot documentation. The documentation distinguishes crawlers used for different purposes and explains robots.txt controls. Because names and behavior can change, use that source rather than copying an old user-agent list from a marketing post.
Decide access policy with SEO, legal, security, and content owners. Check production robots.txt, CDN rules, bot management, authentication, and server logs where available. A permissive robots rule does not prove the crawler fetched a page. A server log showing a request does not prove the content was selected in a response.
Also remember that crawler access is not the only possible source path. Search products can draw on third-party web sources and other retrieval systems. Do not describe an OpenAI crawler change as a guaranteed route into answers.
Publish canonical sources, not “ChatGPT content”
The strongest source strategy is useful regardless of platform. Give important company and product claims one maintained home. Link related pages to that source. Remove contradictions across pricing, documentation, comparison pages, press releases, and partner listings.
High-value B2B sources include:
- technical documentation with version and review dates;
- security and compliance scope explained precisely;
- pricing mechanics and exclusions;
- transparent benchmark methods and samples;
- implementation requirements and failure modes;
- comparisons with explicit evaluation criteria;
- original data whose collection method can be inspected.
Generic listicles written to mention “ChatGPT SEO” repeatedly do not create unique evidence. Google’s people-first content guidance is useful here even though it is Google documentation: publish substantial, sourced information for readers rather than content made mainly to manipulate discovery. It does not describe ChatGPT’s ranking system.
For source-level improvements, see Leaf’s guide to getting cited by ChatGPT.
Make key passages quotable without removing nuance
A source-worthy page should answer a specific question directly, then support the answer. The passage must remain accurate when a reader sees only a small excerpt.
Weak:
Our innovative platform is an enterprise-ready solution for modern teams.
Useful:
The platform supports SAML SSO on the Enterprise plan. SCIM provisioning is available for Okta and Microsoft Entra ID; other identity providers require manual provisioning. This scope was reviewed in August 2026.
The second version defines the capability, plan, limitation, and date. It is better for buyers and less likely to be misunderstood. That does not make it a guaranteed citation.
Use descriptive headings and put caveats beside the claim. Label hypothetical examples. Link technical assertions to maintained documentation and external assertions to primary evidence. If a concise summary would become misleading when separated from a long footnote, rewrite the summary.
Build a controlled prompt panel
A useful panel reflects the buying journey rather than a random keyword export. Include five groups:
- Category discovery: “What types of platforms help with…?”
- Use-case fit: “Which tools support this workflow under these constraints?”
- Comparison: “Compare approaches A and B for a mid-market team.”
- Implementation and risk: “What are the security or migration requirements?”
- Branded facts: “Does Company X support capability Y?”
Use neutral wording for discovery tests. Keep branded factual prompts separate because they answer a different question. Save the exact prompt, date, product or mode, account conditions if known, full answer, all visible sources, mention state, and errors. Repeat high-value prompts to expose variability.
Do not ask leading prompts such as “Why is our company the best?” and report the result as market visibility. Do not silently edit prompts between baseline and retest.
Diagnose gaps with a source comparison table
When another domain is cited, inspect the actual page rather than guessing at a ranking factor.
| Observation | Likely investigation | Appropriate response |
|---|---|---|
| Competitor documentation answers the capability question precisely | Owned documentation may be vague or fragmented | Create one maintained capability source and link to it |
| An industry body is cited for a definition | Primary authority is appropriate | Cite the authority too; do not imitate it with an unsupported definition |
| A directory lists outdated pricing | Cross-source inconsistency | Correct the owned pricing source and request legitimate directory updates |
| Your page is blocked by bot policy | Access defect, if policy is unintended | Review and change policy with the responsible owners |
| Your brand is omitted from a broad recommendation | Relevance, evidence, or sample variability | Investigate; do not infer a penalty from one response |
| Your page is cited but the answer is wrong | Extraction or synthesis problem | Clarify the source passage and record the error on retest |
This table turns a vague “rank us higher” request into specific source work.
Report findings without fake precision
For every reporting period, show the prompt count, repetitions, dates, product or mode, and market. Report raw counts alongside percentages: “5 owned citations across 30 prompts” is more transparent than “16.7 visibility.” Preserve answer examples and URLs for auditability.
Track referral traffic if it is identifiable, but do not assume every mention creates a visit. Connect visits to qualified actions and pipeline where your analytics permit. Keep a clear separation between observed correlation and causation: if citation coverage rises after a documentation release, the timing is evidence for further testing, not proof of a universal factor.
Use a 30-day implementation sequence
In week one, define the prompt panel and crawler policy. In week two, reconcile priority product facts and identify canonical source gaps. In week three, improve a small set of pages with direct, evidenced answers. In week four, rerun the unchanged panel and review both citations and accuracy.
Expand only after the source changes prove useful to buyers and the test reveals a repeatable gap. This approach is less exciting than selling a ChatGPT rank tracker, but it produces assets you control and evidence you can defend.
The central rule is simple: optimize sources, not mythology. ChatGPT controls its responses. Your team controls whether the web contains a clear, credible, current account of what you do.