Optimize for ChatGPT Search: A Technical Checklist

ChatGPT search works in stages, and only some of them are yours.

1. Access. A crawler has to be allowed to fetch your page and actually reach it. This is index-time inclusion, and you control it.
2. Retrieval. For a given question, ChatGPT decides whether to search at all. If it does, OpenAI says it "rewrites your query into one or more targeted queries" and sends them to its search providers. You can't see those queries.
3. Generation. The model picks which retrieved pages to use and cite. OpenAI's full ranking disclosure is that it uses "multiple factors" and "placement is not guaranteed."

There's also a fourth channel that sits outside all of this: what the model already knows with search off. That knowledge is frozen at a training cutoff. As of October 2026, OpenAI documents a February 16, 2026 cutoff for the GPT-5.6 family and August 31, 2025 for the Instant path. Nothing you publish this quarter changes an answer given with search off.

The practical conclusion is simple. Stage 1 is the only stage with documented, publisher-controlled levers. A critical survey of 45 GEO studies (arXiv, July 2026) found no technique with a stable, cross-platform causal effect on discoverability. So this checklist starts with access, and mostly stays there. For the detail behind each stage, see our walkthrough of OpenAI's help page on searching the web with ChatGPT.

OAI-SearchBot is the crawler OpenAI uses to surface sites in ChatGPT's search features. OpenAI's crawler page is direct: sites opted out of OAI-SearchBot "will not be shown in ChatGPT search answers, though can still appear as navigational links." A navigational link is a plain pointer to your site, not a cited source inside an answer.

OpenAI runs three separate agents, and its documentation says "each setting is independent of the others":

• OAI-SearchBot: ChatGPT search inclusion.
• GPTBot: training data for OpenAI's foundation models. Nothing else.
• ChatGPT-User: fetches pages when a user asks ChatGPT to. OpenAI says robots.txt rules "may not apply" to it, and it isn't used to decide what appears in search.

A robots.txt that keeps search access while opting out of training looks like this:

User-agent: OAI-SearchBot
Allow: /

User-agent: GPTBot
Disallow: /

You don't have to block GPTBot. That's a separate policy choice about training, covered in GPTBot vs OAI-SearchBot. What matters here is that OAI-SearchBot isn't caught by a blanket rule. The most common real-world error is a User-agent: * disallow, or a copied "block all AI bots" list, that silently removes ChatGPT search eligibility along with training.

This is the half most sites never test. OpenAI's help centre states that to be eligible you must allow OAI-SearchBot "and confirm that the website host or content delivery network allows traffic from OpenAI's published searchbot IP addresses." The word is and. A permissive robots.txt does nothing if your firewall returns a 403.

How this fails in practice:

• Bot management or a WAF flags the crawler. OpenAI's own guidance notes these systems "can mistakenly block legitimate crawlers, often returning 403 Forbidden errors."
• A JavaScript challenge or CAPTCHA intercepts the request. An automated crawler can't solve it, so it gets an HTML challenge page instead of your content.
• A CDN default changes under you. From September 15, 2026, Cloudflare blocks its Training and Agent crawler categories by default on ad-displaying pages for new domains, new sites on existing accounts and existing free-tier customers, while Search stays allowed by default. Check which category each OpenAI agent falls into in your own dashboard rather than assuming.
• IPv6-only edges. As of August 2026, OpenAI's published searchbot.json listed IPv4 prefixes only. An origin or bot-facing path reachable only over IPv6 can't be reached by those ranges.

OpenAI publishes its ranges as JSON at openai.com/searchbot.json, but it also warns against relying on IP allowlisting alone. Validate on the user-agent plus your CDN's verified-bot programme, and use the IP list as a cross-check. Our page on OpenAI crawlers, robots.txt and firewalls goes deeper on configuration. To see the result directly, the AI Crawler Checker requests a page as each AI crawler and shows which ones your firewall turns away.

Passing both gates gets OAI-SearchBot to your page. It still has to find your content there. Vercel's crawler study found that ChatGPT's crawlers fetch JavaScript files (11.50% of requests) but "don't execute them. They can't read client-side rendered content." Independent re-testing in June 2026 confirmed that GPTBot, OAI-SearchBot and ChatGPT-User still don't render JavaScript.

If your pricing table, product names or key claims load after hydration, OpenAI's crawlers may see an empty shell. Check it this way:

1. Fetch the page with JavaScript disabled (for example, curl or your browser's view-source).
2. Render the same page in a normal browser.
3. Compare the visible text. Anything present only in the rendered version is at risk.

Pay special attention to content behind tabs, accordions, "load more" buttons and infinite scroll that pull text in with a fetch call. The fix is server-side rendering, static generation or prerendering for the content that matters. You don't need to server-render everything, just the facts you'd want quoted.

That last point is the one positive on-page signal in the evidence. Stating a specific fact (a price, a number, a one-line definition) on a page ChatGPT can fetch gives it a sentence it can quote. It doesn't make selection more likely. It means there's something correct to select.

Most "optimize for ChatGPT" advice blurs this line. Here it is plainly.

One thing sits in between. When ChatGPT cites other sites for your category instead of yours, being named accurately on those sites is associated with appearing in the answer. It's an association, not a promised effect, but it's a real route.

Remember too that logged-in answers vary. OpenAI says ChatGPT may use saved memories when rewriting a search query, so one person's test isn't a measurement. For how fresh the fetched copy is, see whether ChatGPT reads your live site or a saved copy.

Work from the cheapest, most certain checks to the slower ones.

1. Read your robots.txt. Confirm OAI-SearchBot is allowed on the pages you want in ChatGPT search, and that no wildcard rule overrides it. Decide on GPTBot separately.
2. Test reachability as the crawler. Request a few representative URLs with the OAI-SearchBot user-agent. A 200 with your real content is a pass. A 403, 429 or challenge page is a fail, even if robots.txt says allow. Confirm /robots.txt itself is reachable too.
3. Review CDN and bot-management settings. Look at AI crawler categories, verified-bot handling and any rules added since September 2026.
4. Diff raw HTML against the rendered page. List every price, product name and core claim that exists only after JavaScript runs, and move it into the server response.
5. Remove accidental suppression. Search templates for noindex, nosnippet, max-snippet:0 and data-nosnippet applied wider than intended.
6. Make your own site consistent. If your pricing page, about page and footer state different facts, an engine can quote any of them. Pick one version and use it everywhere.
7. Then look at what ChatGPT says. The free checker on this site shows whether ChatGPT names you for up to three buyer questions. Treat it as a snapshot, not a trend.

Steps 1 to 5 are binary. Once they pass, they stay passed until someone changes a config, so recheck after migrations, CDN changes and redesigns. The same sequence applies to other engines with their own bots; our Perplexity checklist covers PerplexityBot.

• Allowing GPTBot to "get found by ChatGPT." GPTBot governs training only. OpenAI asserts no search benefit for it anywhere.
• Blocking all AI bots in one rule. It feels tidy and removes ChatGPT search eligibility along with training. Name each agent.
• Treating robots.txt as the whole answer. It's necessary, not sufficient. The firewall decides whether the crawler gets in.
• Assuming ranking in Bing carries over. OpenAI says ChatGPT "sometimes" partners with other search providers, with no measured effect. In OtterlyAI's 2026 test, two sites ranking in Bing drew effectively zero ChatGPT citations while ChatGPT's crawlers kept visiting.
• Adding schema, FAQ markup or an llms.txt and waiting. None has credible evidence as a ChatGPT lever. Existing markup does no harm, but it isn't the fix.
• Rewriting pages in an "AI-friendly" style. C-SEO Bench (NeurIPS 2025) found most content-rewriting methods ineffective and often harmful to ranking within the model's context.
• Hiding text or instructions for AI crawlers. Prompt injection and cloaking exist, and research shows some engines can be manipulated by them. It's illegitimate, it puts your reputation at risk, and it's not a strategy.
• Reading one logged-in answer as the truth. Retrieval is discretionary and memory shapes queries. Test logged out, more than once.