How to Show Up in ChatGPT

In our directory, tools that only rewrite blogs to "show up in ChatGPT" rank below tools that prove the fetch. OpenAI already wrote the fetch rule. Rank the stack in that order, then freeze prompts so you can tell whether the answer used you.

"Show up" is sloppy language. Named in a roundup is not the same as a source link. A session from utm_source=chatgpt.com is a referral. Those three KPIs get mashed into one slide. We keep them separate. A team that reports "we showed up in ChatGPT" without saying which of the three it measured is reporting nothing in particular.

The three KPIs are worth separating out because they answer different questions and they get fixed by different teams. Being named in a roundup is a brand-recognition signal, and it moves slowly with PR and content. Earning a source link is a content and crawl problem, and it moves with what you publish and whether the crawler can read it. A referral session is an analytics and attribution problem, and it moves with your tracking setup. Mashing them into one number hides which one changed, and it lets a team claim a win when only the noisiest of the three moved.

What OpenAI ranked for you

The Publishers and Developers FAQ is blunt. Any public site 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 after that is whether the answer picks you.

Crawler docs split the agents. SearchBot is search surfacing. Allow it, and allow the published IP ranges. Robots.txt changes can take about 24 hours. The 24-hour figure is the reason a "we blocked it yesterday and still appear today" ticket is usually a patience problem, not a config problem.

GPTBot is training of foundation models. Independent switch. You can allow Search and disallow training. If both are allowed, OpenAI may reuse one crawl. That reuse is their operational note, not a ranking factor you can buy. A common mistake is to treat GPTBot as the search lever. It is not. SearchBot is.

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 SearchBot for that.

Atlas caveat from the FAQ: a blocked page can still show as title plus link if the URL arrived another way. Use noindex if you want that gone, and the crawler must be allowed to read the tag. Disallowing the page and then wondering why the tag never applied is a self-own. The order matters: allow the crawler to read the page, then put noindex on it, then block if you want suppression beyond Atlas.

The order matters because each step depends on the one before it. A crawler cannot honor a noindex tag on a page it is not allowed to read. If you disallow the crawler in robots.txt and also expect the noindex tag to take effect, you have asked for two things that contradict each other, and the disallow wins, which means the tag never applies. The correct sequence is to allow the crawl, let the crawler read the noindex, and only then block the path if you want the page gone from Atlas too. Getting this backwards is one of the more common support tickets.

We rank a CDN allow-list that uses published IPs above a content calendar that never checked 403s. WAF presets that block "GPT" strings often block Search by accident. A string match on "GPT" is the wrong tool for a bot policy; the published IP file is the right one.

What we do not rank as a ChatGPT ranking factor

A magic FAQ schema. An llms.txt Google already said Search does not use. Instant alerts. A $29 Promptwatch plan (that number is Otterly, not Promptwatch).

Google's optimization guide is useful homework on unique, non-commodity pages. 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. Recycling everyone else's roundup is a bad long-term bet on any engine.

There is no ChatGPT Search Console. There is no submit button. Public HTML, stable URLs, internal links, a sitemap so the page is discoverable. The claim in text, not only on a canvas the bot never sees. A page that exists only as a rendered widget is a page the crawler may never read.

Software we put after the bot is allowed

Promptwatch sits first in the rankings because appearance is a time series. Explore has 10 ChatGPT prompts for free. Essential at $95/mo adds citation analytics and referred conversions. Crawler logs start on Professional at $245/mo or an agency plan. Content Agents can draft the missing page when the CMS is Webflow or Framer. WordPress publishing is not live.

The reason a tracker comes after the bot is allowed is sequencing. If you block SearchBot, no tracker will show you in the answer, because you are not in the answer. If you allow SearchBot and never measure, you are guessing. The tracker turns the guess into a stored check you can replay.

Otterly Lite is the cheap mention check from $29/mo. It will not show crawler errors. Profound Starter is ChatGPT-only, 50 prompts, annual $99. Fine as a thin ChatGPT desk, not a publish loop. Peec AI last: screenshots, three models on Starter, no fetch log on our listing.

Paid Promptwatch checks are daily. Do not write "instant alerts" into a comparison table. Daily is the cadence; instant is a different product story we will not invent.

A ranked checklist

robots.txt allows OAI-SearchBot.
WAF allows SearchBot IPs from OpenAI's published file, not a regex on the string "GPT".
Canonical money URL returns 200 with the claim in HTML. No accidental noindex unless you want Atlas-only suppression.
Freeze 20 purchase prompts, including vs-queries. Run them daily on a paid plan for two weeks.
If you are named without a link, that is PR, not a citation. Different ticket.
If mentions never appear, move to Professional or an agency plan and read fetches before you rewrite.

The checklist is ordered for a reason. Step 1 is the door. Step 2 keeps the door open under a WAF. Step 3 makes sure the page behind the door is the one you want cited. Steps 4 through 6 are the measurement layer that tells you whether the door worked. Skipping to step 6 without steps 1 and 2 is how teams rewrite content for a problem that was a robots.txt line.

Show up means the stored answer includes you. OpenAI opened the crawl. The tracker tells you whether the answer used it. Product: promptwatch.com.