Why doesn't AI cite you for this question?
Pick one question a buyer asks ChatGPT or Claude. We read your site the way an AI crawler does and tell you why you're not the answer: a crawl problem, a missing page, or a page that's there but weak. Then the one change to make.
Try:
The loop
- Crawl problem? The page exists but robots.txt blocks AI search crawlers, it's marked noindex, it returns an error, or its text only appears after JavaScript runs. Fix access first: until crawlers can read the page, they can't quote it. If your homepage refused our crawler or didn't load, we say we couldn't read your site rather than guess what AI crawlers get.
- Missing page? Nothing on the site mentions most of the question's key terms. Publish one page that answers it in the first sentence, under a heading that asks it.
- Weak page? A page covers the topic, but no heading names the question, the first sentence isn't a direct answer, or it leaves out the specific (a price, a time, a yes or no). Improve that page rather than writing a new one.
- Answered on-site? Then the gap is off-site. Work the sources assistants already cite, sorted into what you own, profiles you can claim, mentions you have to earn, and wrong facts to contest.
Run it for each question you lose, one at a time, or let the question map propose your likely questions and check them all at once. Add a competitor's domain in vs to see what their page gives an assistant that yours doesn't. The same check is earncitations diagnose on the command line (--vs for the comparison), the MCP tool diagnose_question, and GET /api/diagnose (docs).
Questions
Why diagnose before writing?
Because the fix depends on the cause. If the page exists but AI crawlers can't read it, a new article won't help. If a page covers the topic but never answers the question plainly, improving it beats writing another. Only a truly missing answer needs a new page, and if your page already answers, the work is off-site.
How do you pick the page?
We read your homepage, robots.txt and sitemap, pick up to six pages whose address or link text match the question (pricing, FAQ and help pages get a look too, and your docs, developer or help subdomain if your homepage links to it, after checking its own robots.txt), and score each on how many of the question's key terms its text mentions, whether a heading or FAQ entry names the question, and its title. Synonyms count: a page that says "pricing" answers "how much does it cost".
What counts as a plain answer?
A heading or FAQ entry that names the question, a short first sentence under it, and the specific thing a buyer needs: a price stated as text for cost questions, a time for "when" questions, a number for "how many", a comparison table for "vs", and a yes or no for "does it" questions.
What about "what's the best X for Y?"
Assistants can answer those as a short ranked list, each entry "Product — best for [use case]" with a few reasons (as in the ChatGPT answer Jam shared about Akamai). So we check two things: that your homepage title, main heading or opening text says you are an X, and that some sentence says you're for Y ("built for", "designed for", "helps"… within a few words of it). If you are an X but never say who for, the fix is one true sentence, not a new page. If nothing calls you an X, the doctor says so: if you aren't one, assistants are right to leave you out.
Why compare with a competitor?
When an assistant names a competitor for your question, the useful question is what their page gives it that yours doesn't. Add their domain and we read both sites the same way (one after the other) and list the differences: a page that answers plainly, a heading that names the question, a price stated as text, a short sentence an assistant can lift. If the two pages are level, the difference is off-site (who cites them) or in the product itself, and no rewrite will close it. Comparing competitors this way follows Jam (spreadjam.com).
Does this ask ChatGPT or Claude?
No. It reads your own pages the way an AI crawler does (server HTML, no JavaScript, honouring robots.txt) and makes no AI calls, so it is free and gives the same answer every time. To see who the assistants cite today, ask them the question yourself; the off-site list tells you what to do about each source.
Where does the method come from?
The loop (diagnose why you lose a question, then fix the cause, then work the sources AI already trusts as own, claim, earn or contest) follows the AEO workflow Granola published, as summarised by Jam (spreadjam.com). This page automates the on-site half of it.