How to Show Up in ChatGPT

"How to show up in ChatGPT" gets sold as a content trick. OpenAI's own publisher FAQ is blunter. Any public website can appear in ChatGPT search. For summaries and snippets, do not block OAI-SearchBot. Update robots.txt if you have to. Track referrals with utm_source=chatgpt.com. That is the official door. Everything else is whether the answer picks you.

Read the Publishers and Developers FAQ and the crawler overview before you buy a "ChatGPT SEO" SKU, because the vendor page tells you what the bot does and does not do, and most of the SKUs assume you have not read it.

Separate the three OpenAI agents

OAI-SearchBot is how sites get surfaced in ChatGPT search features. OpenAI recommends allowing it and allowing the published IP ranges. Robots.txt changes can take about 24 hours to apply, so a same-day "we allowed it and nothing happened" check is not a real test.

GPTBot is training of foundation models. You can allow Search and disallow training. They are independent. If both are allowed, OpenAI may reuse one crawl, but the decision to appear in Search does not require you to opt into training.

ChatGPT-User is a user-initiated fetch (a question in ChatGPT, Custom GPTs). It is not the Search opt-out switch. OpenAI says robots.txt may not apply because a user asked. It is not used to decide Search appearance. Use OAI-SearchBot for that.

If a disallowed page's URL still arrives via a third party, ChatGPT Atlas may show only the link and title. The FAQ says use noindex if you do not want that, and the crawler must be allowed to read that tag. A robots block that stops the fetch also stops the tag from being read, so it is not a clean hide.

CDN bot fights and WAF presets block these user-agents by accident more often than copy is "not GEO enough." A publisher who rewrites a page three times and still does not appear often has a 403 at the edge, not a content problem, and the rewrite was wasted effort on a page the bot never reached.

The accident is common because edge providers ship "block AI bots" presets that look like a feature and act like a wall. A team that turns that preset on to keep training crawlers out also keeps the search crawler out, and the two get conflated in the same checkbox. The fix is not more rewriting. The fix is a WAF rule that names the bot you want and the IPs you trust.

Make the page fetchable and quotable

Public. Not behind a wall the bot cannot pass. Important facts in text Google-style HTML, not only a canvas. Stable URLs. Internal links and a sitemap so the page is discoverable. OpenAI does not give you a Search Console for ChatGPT. There is no submit button, so the page has to be findable by the bot on its own crawl.

Google's generative AI optimization guide is still the best public writing on unique, non-commodity pages and why recycling everyone else's roundup is a bad long-term bet. It is a Google document. It does not rank you in ChatGPT. It does describe the kind of page that is worth citing anywhere: first-hand, specific, organized for a human.

Do not build an llms.txt hoping ChatGPT Search requires it. Google says it does not use those files for Search. OpenAI's publisher FAQ does not make one a prerequisite either.

The reason is that the file solves a problem the search engine does not have. A search crawler reads the page it fetched, not a separate manifest of pages you wish it read. Building one is a side quest that produces a file and no citation, and the time it took would have moved a real money URL closer to the answer.

Prove it after you allow the bot

Showing up means, on a frozen prompt, the answer named you or cited your URL. A single ChatGPT session is not a program. Answers move, and a one-off check is a snapshot, not a measure.

Promptwatch is the checker we rank for daily UI checks, page-level citations, and visitor analytics for chatgpt.com referrals. Explore is free with 10 ChatGPT prompts. Essential is $95/mo and tracks every supported LLM. Agent Analytics for ChatGPTBot and related fetches starts on Professional at $245/mo with 25M crawler logs.

Otterly.AI from $29/mo will count mentions. It will not show crawler errors. Profound Starter $99/mo annual is ChatGPT-only with 50 prompts. Fine as a thin ChatGPT desk, not a publish loop.

Order of operations

User-agent: OAI-SearchBot allowed. Confirm in logs against OpenAI's published ranges, not a spoofed UA.
Money URLs 200, no accidental noindex unless you want Atlas-only suppression.
One page that answers the buyer question in the first paragraph.
Ten prompts in Explore or Essential for two weeks.
If mentions never appear, check fetches before you rewrite.

That is how you show up. Allow the search bot, be the page worth retrieving, measure the answer. The rest is a slide.