If a buyer asks ChatGPT about the service you sell, your public pages should give it something useful to work with: what you do, who you help and evidence the buyer can check.
Start by establishing whether the intended search system can reach that information. Then improve the answers a prospective customer needs. Access opens the possibility of being used; it does not decide which source an answer selects.
This guide helps a service-business owner give the web team a clear access check. For help applying the finding, book a free Clarity Call.
Make three separate choices
“Allow AI” is too broad an instruction for the person managing your site. OpenAI documents different agents for different purposes.
| Your business decision | The relevant agent and check |
|---|---|
| We want our public service information available to ChatGPT search | Review OAI-SearchBot access to those public URLs, including robots rules and hosting restrictions. |
| We want to choose whether content may be used for model training | Review GPTBot separately. This preference is independent of search crawler access. |
| We want to understand a visit triggered by someone using ChatGPT | ChatGPT-User handles certain user-triggered visits. It is not the automatic search crawler or the control determining search inclusion. |
A business can permit search crawling while declining potential training use. Check the full existing configuration rather than pasting a generic robots recipe over rules protecting other parts of the site. OpenAI also notes that robots rules may not apply to user-triggered visits. See the current OpenAI crawler documentation.
Start with information intended to be public
Choose one priority service page and a relevant supporting page. Confirm that both are meant to be accessible without a login. Include the exact URLs in the request to your web team.
Ask for four things:
- The robots rules that apply to
OAI-SearchBotat each URL, including any broader site rule. - The page response and whether its important service information is available in the retrieved content.
- Relevant hosting or security evidence of blocked, challenged or rate-limited requests.
- If a crawler request is identified, how its identity was checked against OpenAI’s published IP information.
A copied user-agent name can be sent by anyone. A test using that name is useful for diagnosing how a rule behaves, but it is not proof that OpenAI visited.
Do not use this process to expose client portals, private documents or other restricted information. The task is to make the chosen public information available under the policy you intended.
Know what each result establishes
Match the next action to what the check actually showed.
| What you observed | What you can conclude |
|---|---|
| The robots rule permits the search crawler | The published rule permits that access. Retrieval still needs separate evidence. |
| The page opens in a browser | That browser could load it. A crawler may encounter a different response. |
| A request with a crawler-like name succeeds | That test request succeeded. The name alone does not verify the caller. |
| A verified crawler request retrieves the page | That crawler retrieved that URL at the recorded time. It does not prove the page was later cited. |
| A sampled ChatGPT answer links to the page | The link appeared for that recorded question and context. Another answer may differ. |
| Analytics records a referral visit | A visit was recorded under the available attribution. Enquiry quality and delivery need their own records. |
Missing logs leave retrieval unconfirmed. One answer that omits your business leaves the reason unresolved. Both observations can still tell you which evidence to collect next.
OpenAI’s publisher FAQ also describes circumstances in which a page title and link may appear through other discovery sources even when search crawling is blocked. Do not treat a robots block as a promise that every reference disappears.
If a request is blocked, keep the repair specific
Give the hosting or security team the URL, time, response and verified request details. A forbidden response, a rate limit and a human-verification challenge need different investigations.
Ask which rule caused the response and whether it conflicts with the intended public access policy. The appropriate change may be a narrow rule correction followed by a new test. Turning off the site’s firewall or all bot protection is not the default remedy.
For a confirmed technical issue, our technical SEO work sits inside an agreed implementation scope. Establish the issue and responsibility before commissioning changes.
If Cloudflare manages the affected hostname, use the Cloudflare crawler-access worksheet to compare the current policies, verify request identity and trace a block to its rule. A search-access setting and a successful request are separate pieces of evidence.
Once access is checked, help the buyer evaluate you
Our AI SEO service page illustrates the content job: it explains who the work suits, how implementation works, what affects scope and how to begin. Its supporting cost guide answers a different buying question. This is an example of how we organize information, not a claim that this arrangement secured a ChatGPT citation.
Use the existing guides for the next decision:
- Structure a service page when the offer, fit or process is unclear.
- Connect an article to its next step when the reader has an answer but no relevant way forward.
- Read recorded ChatGPT referrals when you want to understand the visits analytics actually recorded.
- Keep a monthly answer sample when you want comparable records of questions, answers and cited sources.
These are different jobs. A reader who is ready to talk can go directly to the call; there is no required sequence of articles.
Send one clear request to your team
Copy this into your existing task system:
For these public URLs, confirm the intended ChatGPT search access and our separate training preference. Record the rules checked, dated response evidence and any verified crawler retrieval. If a block is found, identify its cause and propose the smallest appropriate correction. Name the responsible person and the check to repeat after approval.
Keep the result beside the request: confirmed access evidence, a specific failure, or a named check still outstanding. Set a review date for the team without promising when a crawler will visit or an answer will cite the page.
If you want us to help decide what the finding means for your business, book a free Clarity Call. The free Outside View is available if you prefer a written starting point.