How do I get my content cited by ChatGPT and Perplexity? Our working checklist
Our checklist for getting content cited by ChatGPT and Perplexity: crawler access first, answer-first pages, outside sources, and a loop that measures citation rate.
We get this question from almost every new client, usually phrased as "what do we have to write?" Writing is the third step, not the first. The order matters, because we have seen strong articles sit uncited for months while a blocked crawler went unnoticed.
First, make sure the bots can fetch the page
ChatGPT search relies on OAI-SearchBot, and ChatGPT-User fetches pages when a person's request calls for it. GPTBot is a separate crawler tied to training. Perplexity uses PerplexityBot and Perplexity-User. Claude has ClaudeBot, Claude-User and Claude-SearchBot.
On every engagement we read the robots.txt, then the CDN and firewall rules, then the logs. Rules that look fine in robots.txt are sometimes overridden by bot protection at the edge. We use Promptwatch Agent Analytics for the logs: real-time, IP-verified crawler data, so fake user agents do not pollute the picture, plus error tracking with failure explanations. If a page throws errors for OAI-SearchBot, nothing else we do for that page matters.
Whether to allow training bots is a client decision. We lay out the trade-off and let them choose, but we make sure the search and user-triggered bots are not blocked by accident.
Second, find the questions behind the citations
Engines split a prompt into sub-queries. We look at query fan-outs per model in Promptwatch to see what each engine actually searched for, then map those sub-queries to pages. The map usually shows two things: topics where we have a page that is simply not answering the question directly, and topics where we have no page at all.
Third, write or rewrite pages to answer directly
The pages that get cited in our programs tend to share a few habits. They put the answer near the top, use headings that match how people ask, give concrete details with sources and get updated when facts change. We do not stuff keywords or add tricks. Google's own guidance on its AI features makes the same point: no special markup is required, and normal SEO fundamentals still apply.
For volume work, the Promptwatch Content Agent (Business plan and up) plans, writes and publishes against gaps on a schedule, with a review inbox or a hands-off mode with a quality check. Our clients with the review inbox on get a human edit on everything. We prefer that, particularly for any page that states prices, claims or regulated details. For existing pages, Content Refresh suggests diff-style updates, which is often a better use of time than a new article.
Fourth, earn sources that are not your own site
A lot of citations point to third parties: review sites, forums, YouTube, press. If those sources cover your competitors and ignore you, your own pages face an uphill climb. Citation analytics, plus the Reddit, YouTube and LinkedIn citation reports, show which outside domains the engines trust in your category. Offsite mentions show where your brand is named without a link. We turn those into an outreach and profile-building list.
Fifth, track citation rate, not just citation count
Promptwatch's crawl-to-citation path shows a citation rate per page: of the times a bot fetched it, how often did it end up cited. Citation Trends adds when a page was first cited, ramp-up days, peak rate and decay half-life. A page with a decaying rate gets refreshed. A page that is fetched and never cited gets rewritten.
We then connect it to outcomes with visitor analytics, which tracks AI-referred traffic and conversions from ChatGPT, Perplexity, Gemini, Claude and others. Promptwatch's own published case studies show what that can reveal: Crisp reported that AI-referred traffic converted at twice the rate of traditional channels. That is Crisp's result, not a promise for any other site.
What we say to clients about guarantees
We do not promise a specific page will be cited. Engines retrieve differently from run to run, and sources change. What we promise is a process: remove blockers, answer clearly, earn outside sources and report honestly on what moved.