Short answer: AI assistants cite pages they can reach, parse and trust. An AI citation audit checks those three layers:
- Reach: robots.txt lets GPTBot, OAI-SearchBot, PerplexityBot and Google's crawlers in, and the content is in the HTML (not only rendered by JavaScript).
- Parse: one clear H1, question-style H2s, direct answers, lists and tables, valid schema markup.
- Trust: a named author, visible dates, cited sources and consistent facts about your business.
You can run all of these checks on any URL with the free ZetsApp AI Citation Audit. It fetches the real page, robots.txt and llms.txt and scores what is actually there. For the wider strategy behind it, read our GEO guide.
Why AI citations need their own audit
Classic SEO audits focus on rankings: keywords, links, page speed. AI answers work differently. The assistant retrieves a handful of pages, extracts passages and quotes or links the ones that answer the question most cleanly. A page can rank on page one and still never be quoted, because its answer is buried, vague or blocked from the AI crawler.
Layer 1: Reach (checks 1-4)
1. AI crawlers are allowed in robots.txt
Check for Disallow rules under GPTBot, OAI-SearchBot, PerplexityBot, ClaudeBot and Google-Extended, and under User-agent: *. Many sites blocked these in 2023 and forgot. Note that Google-Extended controls use of your content for Gemini training and grounding; regular Google Search uses Googlebot.
2. The content is in the HTML
View the page source (not the inspector). If your article text is missing and only appears after JavaScript runs, many AI crawlers will see an empty page. Prerender or server-render content pages.
3. The page is indexable
No noindex, a self-referencing canonical, and a 200 status. A wrong canonical (for example, every page pointing to the homepage) silently removes pages from both search and AI retrieval.
4. An llms.txt file exists (optional)
/llms.txt is an emerging convention: a plain-text summary of your site with links to the most useful pages. It is cheap to add, though no major assistant has confirmed it as a ranking input.
Layer 2: Parse (checks 5-10)
5. One H1 that names the topic
Exactly one H1, in plain words, matching the question people ask.
6. Question-style H2s with answer-first paragraphs
Phrase section headings the way people ask assistants ("How much does X cost?"), then answer in the first 40-60 words under each one. That passage is what gets quoted.
7. Lists and tables for comparisons and steps
Steps, criteria, prices and comparisons are far easier to extract from a list or table than from a paragraph.
8. Valid JSON-LD schema
At minimum Organization and WebSite site-wide, plus the type that matches the page: Article or BlogPosting, Product or SoftwareApplication, Service, FAQPage, HowTo, BreadcrumbList. Invalid JSON is ignored entirely, so validate it.
9. Readable sentences
Aim for a Flesch reading ease of roughly 50-70. Long, hedged sentences are hard to quote accurately.
10. Semantic HTML
Main content inside <main> or <article>, navigation in <nav>, a lang attribute and alt text on images. It helps parsers separate content from chrome.
Layer 3: Trust (checks 11-15)
11. A named author with a bio
A visible byline linked to an author page, plus a Person node in schema. See our E-E-A-T checklist for what a credible author page includes.
12. Published and updated dates
Use <time datetime> and datePublished/dateModified. Assistants favour recent sources for anything time-sensitive.
13. Specific, checkable facts
Numbers, named standards, versions and dates. "Costs vary" is never quoted; "USD 50-300 per month for most small teams" might be.
14. Outbound citations to primary sources
Link the standard, the study or the official documentation behind key claims.
15. Consistent entity information
Your business name, address, products and founders should match across your site, Google Business Profile, LinkedIn and directories. Conflicting facts reduce confidence.
What to fix first
| Finding | Impact | Fix |
|---|---|---|
| AI crawler blocked | No citations at all from that assistant | Edit robots.txt |
| Content only rendered by JavaScript | Page looks empty to many crawlers | Prerender or SSR |
| Wrong canonical or noindex | Page excluded from retrieval | Self-referencing canonical |
| No question headings or direct answers | Nothing clean to quote | Rewrite section openings |
| No or invalid schema | Weaker page-type and entity signals | Add and validate JSON-LD |
| No author, dates or sources | Lower trust for competitive queries | Bylines, dates, citations |
How to measure AI citations
- Ask your target questions in ChatGPT, Perplexity, Gemini and Google AI Mode each month and record which sources are cited.
- Check referral traffic from
chatgpt.com,perplexity.aiandgemini.google.comin your analytics. - Track the pages that earn citations and copy their structure to others.
Frequently asked questions
Can I pay to be cited by ChatGPT?
No. Citations in AI answers are not sold. They come from retrieval of pages that are accessible, relevant and trustworthy.
Does schema markup guarantee AI citations?
No. Schema helps systems classify a page and its publisher, but the passage that answers the question still has to be clear and correct.
Should I block AI crawlers to protect my content?
It is a business decision. Blocking prevents use of your content, but it also removes you from AI answers and their referral traffic. Many businesses allow search-oriented bots and block only training crawlers.
How is this different from a normal SEO audit?
It focuses on passage-level answers, crawler access for AI bots and trust signals, rather than keyword rankings and backlinks. Both matter; this one covers the gaps classic audits miss.
Audit a page now
Run the free AI Citation Audit on your most important page, then fix the high-priority items. Pair it with the E-E-A-T Signal Checker for the trust layer.
