Start here
Make your pages accessible to search, answer a specific buyer question, and inspect the sources returned for that question. Access makes a page eligible to be found; it does not guarantee a citation.
A ChatGPT answer can mention your brand without linking to your website, or cite your website without recommending your product. Decide which outcome matters before treating both as one visibility score. A pricing correction, a product shortlist and a sourced definition require different checks.
Search access and training access are different
OpenAI documents three relevant agents. OAI-SearchBot supports search discovery, GPTBot collects potential training material, and ChatGPT-User handles certain user-initiated visits. Their purposes and controls differ. Blocking training access is not the same decision as opting out of search.
| Agent | Purpose | Decision |
|---|---|---|
| OAI-SearchBot | Search crawling | Allow it if you want your pages considered for search results. |
| GPTBot | Potential model training | Choose separately from search access. |
| ChatGPT-User | User-requested page visits | Inspect user-fetch behavior separately from automatic crawling. |
Check your CDN and firewall as well as robots.txt. A permitted user agent can still receive a challenge page. Verify requests against the provider’s published IP ranges before trusting a crawler name in a log. Keep authenticated and private pages protected.
Inspect the question before changing the page
- 1.Choose a question that could change a buying decision. Include the customer’s task and constraints, without inserting your brand into every question.
- 2.Run it in the ChatGPT product and record the date, account context, search state, full answer and visible sources. Repeat the check rather than saving only the result you like.
- 3.Open the cited pages. Note whether they are vendor pages, comparisons, documentation or discussions. Record what information they supply that your page lacks.
- 4.Check whether your brand is absent, mentioned inaccurately, or named for the wrong use case. Each calls for a different response.
Worked example, not a measured result
Suppose you sell scheduling software. “Best scheduling software for a five-person salon with deposits” is a more useful discovery prompt than “Is Acme good?” If the answer compares cancellation rules and deposits, a page explaining those policies gives a reader something concrete to assess.
Write the page the buyer needs
Put the direct answer near the relevant heading. State eligibility, limits, price conditions and tradeoffs in language a buyer can check. Use a comparison table when it makes differences clearer. Link to evidence for specifications and claims. These are practical content recommendations, not a published formula for ChatGPT’s ranking system.
For the salon example, show which plan supports deposits, what happens after a cancellation and whether an existing booking can be changed. A paragraph saying the product “transforms appointments” does not resolve any of those questions. Do not invent a policy, statistic or customer quote to make the page sound more specific.
Check third-party sources without planting reviews
When the answer relies on a third-party comparison, review whether its description of you is current. Supply corrections through the publisher’s normal process and disclose commercial relationships. A paid placement does not guarantee a recommendation. A source link in an AI answer also does not tell you how many people saw or clicked it.
Measure the same surface again
After publishing, confirm the page is accessible and that the change is actually visible. Reuse the original prompts and comparable settings. Keep missing or failed answers separate from answers that omit your brand. A new citation is worth investigating, but one appearance is not a stable trend.
API results and the consumer ChatGPT experience should be labeled separately. They can differ in model, tools, search behavior and account context. Do not combine an API baseline with a browser follow-up and present the difference as the effect of your content edit.
Put the findings to work
August connects visibility tracking with a reviewable next step: inspect the answers and citations, prepare a draft, publish through your own workflow, then check what changed. Copy the content or create a GitHub draft PR; you control what goes live. A follow-up comparison is evidence to examine, not proof that one edit caused a gain.
Get started with August. Compare the current plans: Starter includes 25 tracked prompts across one brand workspace; Pro includes 75 across three, with Google AI Overviews and AI Mode alongside the four core engines. Prompt allowances are shared across the account.
Related reading
Use the GEO introduction for the broader method and the Claude guide for its separate crawler controls.
Questions, answered.
Can I allow search while blocking training?
OpenAI provides separate controls for OAI-SearchBot and GPTBot. Review the bot documentation and your firewall together.
Does allowing the crawler guarantee a citation?
No. It addresses access. Retrieval, source selection and the wording of the answer are separate steps.
How quickly should a change work?
There is no dependable universal timeline. Verify publication, keep the measurement settings comparable, and watch repeated results rather than changing the page after every run.
Find out what AI tells your customers.
Book a call and we’ll show you exactly what AI says about you.