Your Page Ranks on Google. ChatGPT Still Ignores It. Here's the Checklist I Use
Your page ranks on Google. ChatGPT still ignores it. Here's the checklist I use A page can sit near the top of Google and still get zero citations from ChatGPT, Perplexity, or Doubao. I ran into this gap while working
Your page ranks on Google. ChatGPT still ignores it. Here's the checklist I use
A page can sit near the top of Google and still get zero citations from ChatGPT, Perplexity, or Doubao.
I ran into this gap while working through a real question: why does good SEO not turn into AI visibility? The short answer is that ranking and being cited are different jobs. Ranking is about being found and ordered. Being cited is about being usable as evidence in an answer.
That split is worth checking in order, not all at once. If you skip ahead to "write more content," you can spend weeks fixing the wrong stage.
1. Can the AI crawler even get in?
Start here because it is the easiest stage to prove, and the easiest to get wrong.
Check robots.txt first, but read it carefully. OpenAI splits its crawlers by purpose:
-
OAI-SearchBotis tied to ChatGPT search display. -
GPTBotis tied to model training. -
ChatGPT-Usercan fetch a page when a user asks for it.
Blocking GPTBot does not block ChatGPT Search. I have seen teams block training crawls for policy reasons and then assume they also blocked search. They did not, or the reverse: they thought they allowed everything, but one rule blocked the search bot.
Then check the server, not just the file. A clean robots.txt does not help if your CDN, WAF, or firewall returns 403, 429, or 5xx to those crawlers, or throws a human check at them. Also check whether the main text only appears after JavaScript runs, and whether a bad noindex, canonical, or redirect is in the way.
One more caution from server logs: seeing a user-agent string in a log does not prove the real official crawler visited. That string can be faked. When it matters, check the platform's published IP ranges or verification method. And a successful crawl is not the same as being indexed, or being picked for an answer.
2. Does your page answer the question AI users actually ask?
This is where most ranking pages fail.
Google's query fan-out is a useful mental model here. For a complex question, an AI search product may split it into several related queries and pull different pieces of information. A page that ranks for one broad keyword may match none of those sub-queries.
Take the example from the original write-up: a Beijing air-source heat pump installer page ranks well for Beijing air-source heating. But an AI user asks something narrower:
A 120-square-meter rural self-built house in Beijing, using air-source heating in winter β how much will it cost per month? Which local installers are reliable?
That question needs a cost calculation, insulation conditions, local temperatures, a real install case, and a service area. A page that explains the benefits of heat pumps does not provide that evidence, even if it ranks.
SEO asks: can a search system find you and rank you? AI visibility also asks: can your page supply a direct, checkable piece of the answer?
3. Does the page give evidence, or just adjectives?
"Professional," "reliable," and "affordable" do not help an AI answer a question. Numbers, conditions, and limits do.
What tends to be citable is specific: a real case, a parameter, a measurement condition, a price structure, a coverage area, an applicable scope. That is not a promised scoring formula from any platform. It is just what an answer needs in order to use a source without guessing.
If your page is mostly marketing copy, do not expect a citation to fix it. Rewrite one section around a question a real user would ask, and give the answer with its conditions attached.
4. Mentioned, cited, and recommended are three different results
This distinction saves a lot of false conclusions.
An AI system can mention a brand, cite a specific URL, and recommend a business β or do only one of those. It can cite a dealer site to explain how heat pumps work, and still not recommend that dealer, because recommending needs more: a verified business, brand authorization, service area, and ability to deliver.
So when you test, do not record a single yes/no. For a fixed set of 20 to 30 real user questions, run each one more than once and record separately:
- Was the brand mentioned?
- Was the site cited, and which URL?
- Was the business recommended, and in what position?
- Who else was cited, and for what?
One run with no citation proves very little. The same question, worded differently or asked on a different day, can return a different set of sources. AI does not always go online, and it does not always show every source it used.
5. Submitting a sitemap helps. It does not finish the job.
For sites working with Chinese search ecosystems, there is basic work worth doing: verify the site in Baidu, Toutiao, Shenma, Sogou, and 360 webmaster tools, submit a sitemap, and watch crawl errors and index status. That helps the related search system find the content and gives you something you can re-check.
But do not turn that into a promise. Submitting a sitemap does not mean a URL will be crawled, indexed, or cited. And for Kimi and DeepSeek, I have not found a public URL submission portal for ordinary sites, so do not force them into the same table.
The order that makes sense to me is: first make sure you can be found, then give the page something worth citing, then test with repeatable questions and watch whether mention, citation, and recommendation move separately.
The full version of this checklist, with the crawler IDs, the webmaster entry points, and the five-layer diagnostic I use for a client site, is in this five-stage GEO troubleshooting framework.
Disclosure: this draft was written with AI assistance and edited by hand. The framework, the heat-pump example, and the crawler details come from the linked post, published 2026-10-20.
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes β full credit and traffic to the original publisher.