How do you make your content appear in ChatGPT? What we do for clients
Our working method for getting client content into ChatGPT answers: bot access, fan-out coverage, quotable pages, off-site mentions, and a weekly measurement loop.
Clients ask this in the first call, usually in the form "how do we get ChatGPT to mention us?" The honest answer is that nobody controls the output. What we control is whether the model can reach the content, whether the content answers the questions it asks, and whether we can see any of it happening. Here is the order we work in.
Start with access, not content
Before we write anything, we check whether ChatGPT's bots can read the site. OpenAI uses GPTBot, OAI-SearchBot and ChatGPT-User, and each does a different job. We have seen sites with excellent pages and a bot-protection rule that quietly returned errors to all of them. Nothing else we do matters until that is fixed.
We look at robots.txt, CDN and firewall rules, and whether the key text exists in the server-rendered HTML. Then we confirm it with real log data. We run client programs on Promptwatch for this reason: its Agent Analytics shows IP-verified AI crawler logs, so spoofed user agents are filtered out, and it follows the path from crawl to citation with a citation rate per page. If a page is fetched often and never cited, that is a content problem. If it is never fetched, that is an access problem. Those need different fixes, and without logs people guess.
Logs come into Promptwatch through the integrations the client already has: Cloudflare, AWS CloudFront, Fastly, Vercel, Netlify, Akamai, Google Cloud CDN, a custom HTTP route, or the WordPress plugin. Crawler logs start on the Professional plan, so on Essential we use the free Agentic Readiness Check as a first pass.
Map the questions behind the question
ChatGPT does not search for your prompt verbatim. It splits it into sub-queries, the query fan-outs, and assembles the answer from whatever pages match those. A client who wants to appear for "best payroll software for small teams" needs pages for the sub-questions too: pricing, setup time, integrations, who it is not for.
We pull the fan-outs per model in Promptwatch, and we use the GEO Inspector Chrome extension in workshops because it shows fetched versus cited sources on a live answer, running locally in the browser. That moment, when a client sees ChatGPT fetch ten pages and cite two, changes the conversation about what to publish.
Then we run content gap analysis. Promptwatch scores coverage and lists the prompts where competitors are named and the client is not. That list becomes the editorial calendar.
Write pages a model can quote
We keep the editorial rules plain:
- The direct answer goes in the first lines under the heading.
- Headings use the words buyers use.
- Claims are specific and dated, and anything that changes (price, limits) gets a review date.
- Comparison questions get a table.
- Every page has a named author and a clear owner.
For existing pages, we use Content Refresh, which proposes diff-style updates rather than a full rewrite. For new pages, the Content Creation tool works from a brief, and we edit hard. Publishing goes to WordPress, Webflow or Framer. One note for Framer clients: there is no per-item publish there, going live redeploys the whole site, so we batch changes.
Get named somewhere other than your own site
Models read more than your domain. Forums, review pages, videos and professional networks all feed answers. In Promptwatch we look at offsite mentions (the brand named in third-party pages without a link), and the Reddit, YouTube and LinkedIn citation reports. The point is to find out which outside sources the model already trusts for a client's prompts, then decide whether outreach, a community presence or a partnership is worth the effort. We do not buy mentions. If a source would not hold up to a human reader, we skip it.
Measure weekly, adjust monthly
We set up a prompt panel by topic and persona, usually a few dozen prompts per client, tagged by intent. Promptwatch runs them daily per model, country and language and reports Visibility Score, Share of Voice, Sentiment and Position in Answer. We review movement weekly and change content monthly, because answers vary run to run and a single day tells you little.
The Action Board helps here. Its Discovery agent reviews the data on a schedule and suggests up to five evidence-backed actions, which we triage with the client. Agent Chat answers ad hoc questions over the live data, and we use scheduled chats for the weekly briefing.
What we tell clients it cannot do
Tracking does not make ChatGPT recommend anyone. It shows where you stand and which levers moved. We also do not promise timelines. Some pages get cited soon after a fix, others never do, and the Citation Trends view (first cited, ramp-up days, peak rate, decay half-life per page) is useful precisely because the pattern differs from page to page.
If you are starting from zero, the order is simple: confirm bots can fetch you, map the fan-outs, close the biggest gaps with quotable pages, find the outside sources that matter, and measure on a schedule. It is slow, unglamorous work, and Promptwatch is where we do the measuring.