English · Lire en francais
The seven checks that decide whether an answer engine can cite your page
An answer engine can only cite a page it can fetch, parse and trust. Seven structural checks cover almost all of it: valid JSON-LD, quotable title and H1, a named author and date, real numbers, a reachable sitemap and llms.txt, an answer in the first paragraph, and a robots.txt that does not block AI crawlers.
Most pages that fail to get cited do not fail on quality. They fail on plumbing. The model never got a clean, attributable, parseable version of the page, so it quoted somebody else who did.
This is the checklist we run, in the order we run it. It is the same list our GEO Site Auditor automates, published in full because a checklist you can read is more useful to an agency than a black box that returns a score.
1. schema.org JSON-LD, present and valid
Look for Article, FAQPage or Organization in a <script type="application/ld+json"> block. Not microdata, not RDFa, not a plugin that emits three conflicting blocks.
The failure mode nobody catches: two plugins each inject their own Article node with a different headline, and the page now makes two claims about itself. Validate the output, not the plugin settings.
2. A title, meta and H1 written to be quoted
Written to be quoted is different from written to rank. A title that reads well as the first line of a generated answer is a title that states the claim. Elevator maintenance: what the 2026 rules change for public buildings can be lifted whole. Home | Acme Group cannot.
Rule of thumb for client work: if you cannot imagine the sentence appearing verbatim in an answer, rewrite it.
3. E-E-A-T signals, meaning a named author and a visible date
Not a category page byline. A human or organisation name, and a publish date the crawler can find in the markup, not only in a sidebar widget rendered by JavaScript.
An undated page is a page an engine has no reason to prefer over a dated competitor, and no way to age out when it goes stale.
4. Verifiable stats rather than vague claims
Most companies is unciteable. A sentence of the shape of the N pages we audited for this client in June 2026, X blocked GPTBot is citeable, because the number carries a scope, a date and a source. Engines tend to surface the sentence that carries the number, so the number is the hook. Use your own measurements, never a round figure borrowed from a blog post.
This is also the check that most often improves the writing on its own. Removing every unsourced claim from a client page tends to shorten it by a third.
5. llms.txt and sitemap.xml, reachable at the domain root
Both at the root, both returning 200, both listing what actually exists. A sitemap that lists redirects and 404s is worse than no sitemap, because it spends crawl budget proving your site is unreliable.
6. An answer in the first paragraph, under 60 words
The engine reads the top of the page first and often stops there. If the first paragraph is a warm-up about how the industry has changed in recent years, that is what gets read and discarded.
Put the answer first. Everything below it is evidence for the reader who wants more.
7. robots.txt that does not block the AI crawlers
The most common single cause of zero citations, and the least visible. A security plugin, a CDN preset or a well-meaning developer adds a broad disallow, and the page becomes invisible to the engines while ranking perfectly well on classic search.
Check for these user agents by name:
| Crawler | Belongs to |
|---|---|
GPTBot, ChatGPT-User | OpenAI |
PerplexityBot | Perplexity |
ClaudeBot, Claude-Web | Anthropic |
Google-Extended | Google, controls AI training and grounding separately from Googlebot |
Blocking Google-Extended while keeping Googlebot is a legitimate choice. Doing it by accident is not.
How to sell this without pretending it is magic
Three of these seven are usually already fine on a well-built site. The value of the audit is not finding seven problems, it is proving in twenty minutes which of the seven are real, and handing the client a document that says so.
That is the whole deliverable: a page-by-page pass or fail, the exact line to change, and a date. Charge for the judgement, not for the scan.
What we run this with
Our own extension, the GEO Site Auditor, automates the seven checks and exports a white-label Markdown or PDF report you can hand to a client under your own name. Free for five audits a month, 29 dollars once for unlimited. No subscription, no account, nothing leaves the browser.
The checklist above works with or without it. It is the checklist that matters.