Most pages that fail to get cited by ChatGPT fail on something mechanical, not on quality. The page is blocked, or empty when fetched, or the answer is buried, or the one fact the question needs is missing. This checklist catches those. Run it on the page you want cited for a specific question, top to bottom. Each item has a test that takes a minute and a fix that usually takes less than an hour.
Key takeaways
Work in order: access, rendering, match, answer, facts, trust, freshness. A page that fails an early item cannot pass the later ones no matter how well written it is. Most sites fail two or three items, and they are almost always in the first half of the list.
Part 1: Can ChatGPT reach the page?
1. Robots.txt allows OAI-SearchBot and ChatGPT-User.
OpenAI documents three bots on its crawler page: GPTBot for training, OAI-SearchBot for the search index behind ChatGPT search, and ChatGPT-User for pages fetched when a user asks. Blocking GPTBot is a legitimate choice. Blocking the other two removes you from answers.
Test: the AI bots checker reads your robots.txt for each bot and then fetches your homepage as each one.
Fix: add explicit Allow: / rules for OAI-SearchBot and ChatGPT-User. Our robots.txt guide has a file you can copy.
2. The page returns 200 to a bot, not a challenge. Bot protection, geo-blocking and "verify you are human" screens return a page, but not yours. Test: the same checker shows the status code and the first lines of what each bot received. Fix: allow-list the documented user agents and IP ranges in your CDN or firewall.
3. No login, cookie wall or age gate in front of the content. An interstitial that a human clicks through is, to a fetcher, the whole page. Test: open the URL in a private window. If the first thing you see is not the content, neither does the bot. Fix: serve the content and put the interstitial in an overlay, or remove it from informational pages.
Part 2: Does the content arrive as text?
4. The main content is in the HTML, not rendered by JavaScript. Fetchers used for citations take the HTML response. Content that a script inserts afterwards is not there. Test: the citability checker fetches the page as a bot and shows the text it got, with a score. Fix: server-side render the page, or pre-render it. Most frameworks have a setting for this.
5. The answer is not inside tabs, accordions or "read more". Hidden panels are often in the HTML, which is fine, but some builders load them on click, which is not. Test: view the page source and search for a sentence from inside the panel. Fix: keep the direct answer in the open, and let the panels hold detail.
6. Text is text, not images. Prices in a graphic, opening hours in a PNG, a features grid built from icons: all invisible. Test: select the content with your mouse. If it does not highlight, it is an image. Fix: write it as words. Keep the graphic if you like, and add the sentence next to it.
Part 3: Does the page match the question?
7. The customer's question, in its plain form, is in the title and the first paragraph. ChatGPT rewrites questions into short literal searches. "Best invoicing app for freelancers" or "freelancer invoicing software" should appear in the title tag, the H1 and the opening lines of the page you want cited for it. Test: read the title tag. Does it contain the question as a stranger would type it? Fix: one page per question. Retitle the page around the question, not the brand.
8. The answer is in the first screen. Long introductions push the answer past what gets read. Lead with it. Test: load the page on a phone. Is the answer visible without scrolling? Fix: move the answer up. The story of why you wrote the page can go below.
9. The page answers one question completely. Pages that touch ten questions lightly lose to pages that finish one. Test: could you summarise the page in a single sentence starting with the question? Fix: split multi-topic pages. Link between them.
Part 4: Are the facts there?
10. Every fact the question needs is a sentence. Price, location, who it is for, what is included, what is excluded, how to start. "Plans from $12 a month, no card needed for the 14-day trial" can be quoted. A pricing table with hover tooltips cannot. Test: list the five facts a customer would want, then find each one as a complete sentence on the page. Fix: our post on the facts AI can't find has the page-by-page list.
11. Names are consistent. The business name, product name and location are written the same way on this page, the homepage, your Google profile and your review listings. Assistants match names across sources; variants split you into strangers. Test: search your site for each variant you have ever used. Fix: pick one and use it everywhere, including the schema.
12. Structured data describes the page. JSON-LD for Organization or LocalBusiness, plus Product, Service, FAQPage or Article as appropriate. It does not get you cited on its own, but it confirms what the text says and is how search engines, which feed step one, understand the entity. Test: the schema checker lists what the page declares and what is missing. Fix: add the missing type with the same facts as the text. Schema that disagrees with the page is worse than none.
Part 5: Will it be trusted and kept fresh?
13. Author, date and organisation are visible. A named author, a "last updated" date, and a link to an about page. Cited pages overwhelmingly have all three, because the search step favours them. Test: look at the top or bottom of the page. Fix: add a byline block. Update the date only when you change the content.
14. The page has been updated in the last year and links out to sources. Freshness matters for anything with prices or rankings. Outbound links to sources make the page look like reference material rather than an advert. Test: the date on the page, and a count of outbound links. Fix: review the page quarterly. Cite where your numbers come from.
Running the list
Put the fourteen items in a spreadsheet with one column per page you want cited. Score each cell pass or fail. Fix in order, top to bottom, because an early failure hides the later ones: a page that fails item 4 shows nothing to test for items 7 through 14.
Then re-check the question in ChatGPT with search on, in a fresh chat, a week later. Citations do not update the moment you publish. They update when the search index catches your change and the model happens to read your page for that question. In our experience that is one to six weeks depending on how often your site is crawled.
If you would rather have the checks run for you, a free scan runs the technical items on your site, asks ChatGPT twelve of your customers' questions, and lists the pages it cited so you know which URLs to run this list on next.
