Part 3: How to Audit Any Page for AI Citability

Q: How do I know if my changes are working?

Track three things: GA4 referral traffic from chatgpt.com, perplexity.ai, and claude.ai; monthly manual query audits in ChatGPT and Perplexity; and your Google Search Console data for ranking improvements. Part 2 of this series covers the full tracking routine.

This post is Part 3 of four in a series about getting cited by AI search engines like ChatGPT and Perplexity.

In Part 1, we covered the content traits that get you cited — answer capsules, link-free formatting, original data. In Part 2, we covered how to track whether those changes are working. This post completes the series: how to audit a specific page and fix what is holding it back.

For background on what GEO is and why it matters, see the Generative Engine Optimization guide.

The problem with most AI visibility advice: It tells you what to do in general. It does not tell you what is wrong with this specific page and what to fix first. That is what an AI citability audit solves.

We built a dedicated AI Citability report into Brass-SEO because we needed it ourselves. After publishing Parts 1 and 2, the obvious next question was: how do we evaluate our own pages against these criteria without manually checking every heading and paragraph? The answer was to build the audit into the tool.

As with Parts 1 and 2, this post uses the answer capsule technique throughout. Every H2 is followed by a direct, self-contained answer before the supporting detail.

What an AI Citability Audit Checks
Answer Capsules Are the Highest-Impact Fix
Brand Intent Is the Factor Most People Miss
Why FAQ Sections Matter More Than You Think
Original Data Is Your Competitive Moat
Expert Quotes and Citations Boost Visibility
Structure Determines What Gets Extracted
What llms.txt Does and Whether You Need It
How to Prioritize Your Fixes
Frequently Asked Questions

What an AI Citability Audit Checks

An AI citability audit evaluates whether a page is structured so that language models can extract, attribute, and cite its content. It checks ten factors across content structure, data quality, and discoverability.

The ten factors are not arbitrary. Each one comes from published research or from patterns observed in pages that actually receive AI referral traffic. The answer capsule data comes from Adam Gnuse's study of 15 domains and 7,500 ChatGPT referral sessions. The tracking framework comes from Paul DeMott's research on LLM visibility measurement. And the quotability signals come from Princeton's GEO (Generative Engine Optimization) study, which found that adding expert quotations improved AI visibility by up to 41%.

Here is what a citability audit evaluates:

Answer capsules — Does each H2 have a self-contained opening sentence?
Link-free capsules — Are those openings free of links?
Original data — Does the page contain proprietary numbers or unique findings?
Information gain — Does the page say something an LLM cannot assemble from other sources?
Brand intent clarity — Could an LLM extract a clear statement of who you are and what you do?
FAQ section — Is there a dedicated Q&A section?
Content front-loading — Are key answers in the first 30% of the text?
Expert quotes and citations — Does the content cite sources or quote authorities?
Heading hierarchy — Is the H1-H2-H3 structure clean and logical?
Scope clarity — Are the page's claims bounded and quote-safe, or do unscoped superlatives make them risky for an LLM to cite?

If you are a Brass-SEO customer, the Brass-SEO AI Citability button on your dashboard runs this full audit on any page in seconds — it fetches the page, checks your Google rankings and AI referral traffic, evaluates all ten factors, and gives you prioritized fixes with specific rewrite suggestions.

Answer Capsules Are the Highest-Impact Fix

Answer capsules appeared in 72.4% of blog posts cited by ChatGPT. Adding them to your H2 headings is the single highest-impact structural change you can make.

As we covered in Part 1, an answer capsule is a self-contained sentence of 120-150 characters placed immediately after a heading. It directly answers the question the heading implies. No links, no qualifiers, no "click here to learn more."

What makes this the highest-impact fix is the combination of effect size and ease of implementation. You do not need to rewrite your entire post. You need to rewrite the first sentence after each H2. That takes about 10 minutes per page.

Here is the difference:

Before (no capsule):

How to Improve Page Speed

There are many factors that go into page speed, and if you have been reading about Core Web Vitals, you know that Google has been emphasizing this metric...

After (with capsule):

How to Improve Page Speed

Compress images, defer non-critical JavaScript, and enable browser caching to reduce load times by 40-60%.

The second version gives an LLM exactly what it needs: a complete, quotable answer in one place. The first version requires the model to read three more paragraphs before it finds anything citable.

Brand Intent Is the Factor Most People Miss

Brand intent clarity means your page explicitly states what you want an AI to say about you — and most pages do not do this.

This is the factor we added to the audit based on our own experience, not from published research. When we asked ChatGPT about various small business websites, the AI responses were often vague or generic — "a company that offers services" — because the pages themselves never made a clear positioning statement.

The fix is straightforward. Somewhere in the first 30% of your page — ideally in the first paragraph or in a prominent callout — state clearly:

Who you are
What you do
Why it matters
What makes you different

This is not marketing copy. It is a quotable fact. An LLM that encounters "Brass-SEO is an AI-powered SEO analysis tool that gives small businesses access to the same data insights that agencies charge thousands for" has something concrete to cite. An LLM that encounters "We help businesses grow" does not.

When auditing your pages, ask yourself: if ChatGPT summarized this page in one sentence, would it say what you want it to say? If the answer is no, the brand intent is unclear.

Why FAQ Sections Matter More Than You Think

Q&A is the highest-cited content format in AI-generated responses. A dedicated FAQ section gives LLMs pre-formatted question-answer pairs that are trivially easy to extract and cite.

This is not speculation. Research on how generative engines select content to cite consistently shows that question-answer formatting outperforms dense paragraphs, narrative prose, and even bulleted lists. The reason is mechanical: an LLM responding to a user's question looks for passages that directly answer questions. A FAQ section is literally that.

For SEO practitioners, this is doubly valuable. A properly structured FAQ section with the heading ## Frequently Asked Questions also generates FAQ rich snippets in Google search results through FAQPage schema markup. You get both traditional SEO value and AI citability from the same content.

If your page does not have a FAQ section, adding one is usually the second-highest-impact fix after answer capsules. Start with 3-5 questions that your target audience actually asks. Use the exact phrasing your audience uses — not corporate jargon.

Original Data Is Your Competitive Moat

Original data appeared in 52.2% of blog posts cited by ChatGPT. It is the factor that competitors cannot easily replicate.

As we discussed in Part 1, original data does not mean you need to commission formal research. It means publishing information that only you can provide. Your own test results. Your own benchmarks. Your own customer outcomes.

We did this in our AI model evaluation post, where we published the actual cost, speed, and accuracy results from testing 10 AI models in our production application. No other site has those numbers because no other site ran those tests. When an LLM needs to cite model comparison data, our page is one of very few options.

The strongest configuration found in the research was answer capsules combined with original data — present in 34.3% of cited posts. If you have proprietary data, put it in a capsule. Make it easy for the LLM to extract your unique finding in one sentence.

Expert Quotes and Citations Boost Visibility

Princeton's GEO research found that adding expert quotations improved AI visibility by up to 41% — the single largest effect of any optimization technique tested.

This finding comes from the foundational Generative Engine Optimization paper, which tested nine different optimization methods. Adding quotations from experts outperformed every other technique, including adding statistics (+34%), citing sources (+30%), and improving fluency (+15-30%). Notably, keyword stuffing — a standard SEO tactic — showed "little to no improvement" for AI citations.

The practical application is simple: when you make a claim, back it up with a quote from a recognized authority. When you cite a statistic, name the source. When you describe a process, reference who developed it. This gives LLMs the credibility signals they need to confidently cite your content.

This post demonstrates the technique. We cite Adam Gnuse, Paul DeMott, and the Princeton GEO researchers by name, with links to their original work. We do not say "research shows" — we say who researched it and where it was published.

Structure Determines What Gets Extracted

44% of all LLM citations come from the first 30% of a page's text. Front-loading your key answers is a structural decision that directly affects citation rates.

This means the most important information on your page should not be buried after a long introduction, a personal anecdote, or three paragraphs of context-setting. The answer should come first. The context should come after.

A citability audit checks several structural factors:

Heading hierarchy — H1 → H2 → H3 without skipping levels. LLMs use headings as semantic landmarks to understand page structure.
Content front-loading — Is the key answer in the opening paragraphs, or buried below the fold?
Lists and formatting — Structured content with headings and lists performs almost as well as Q&A format, while dense unbroken paragraphs perform worst.

This is consistent with what works in traditional SEO too. Google's featured snippets operate on the same principle: they extract concise, direct answers from pages with clear structure. Optimizing for AI citability and optimizing for featured snippets are largely the same work.

For a wider treatment of shaping content so language models can lift and attribute it, see Copper Sun's guide to content structure for AI citation, which approaches the same structural principles from the content-production side.

What llms.txt Does and Whether You Need It

llms.txt is a proposed open standard that helps AI systems navigate your website by providing a curated markdown map of your most important content. It is a site-level signal worth knowing about, not one of the ten page-level factors the audit scores above.

The specification was proposed by Jeremy Howard of Answer.AI in 2024. You place a /llms.txt file at your site's root — similar to robots.txt or sitemap.xml — containing a markdown-formatted list of your key pages with brief descriptions. The idea is to give LLMs a concise overview of what your site offers, since their context windows are too small to process an entire website.

Current reality: No major LLM provider (OpenAI, Google, Anthropic) has officially confirmed they follow llms.txt when crawling websites. Adoption is growing — Mintlify rolled it out across thousands of documentation sites in late 2024, and Anthropic's own docs site supports it — but it is not yet a confirmed ranking factor for AI citations.

Our recommendation: If you have a content-heavy site with 20+ pages, creating an llms.txt file is a low-effort, no-downside optimization. It takes 15 minutes and costs nothing. It will not hurt your citations if LLMs ignore it, and it may help if they start paying attention to it. For a site with 5 pages, it is probably not worth the time.

Brass-SEO treats llms.txt as a general site-level recommendation, separate from the ten page-level factors the citability audit scores.

How to Prioritize Your Fixes

Start with the two changes that research shows have the largest impact: add answer capsules to your H2 headings, then add a FAQ section at the bottom of the page.

Not all ten citability factors are equally important, and not all pages need all ten. Here is a practical prioritization:

Fix first (highest impact):

Answer capsules on H2 headings — 72.4% of cited posts have them
FAQ section — highest-cited content format for LLMs
Brand intent statement — especially on your homepage and service pages

Fix next (significant impact): 4. Original data — if you have it, put it in a capsule 5. Information gain — say something readers can't get from ten other pages 6. Expert quotes and citations — up to 41% visibility improvement 7. Content front-loading — move your key answer to the first 30% 8. Scope clarity — bound overclaims so an LLM can quote you without being wrong

Fix when convenient (moderate impact): 9. Link-free capsules — remove links from the first sentence after H2s 10. Clean heading hierarchy — no skipping levels

Adding an llms.txt file is a separate, site-level tactic — low effort and low priority, and not one of the ten scored factors.

The Brass-SEO AI Citability report in Brass-SEO ranks these fixes by impact for your specific page, so you do not need to guess. It also checks whether the page is already receiving AI referral traffic from ChatGPT, Perplexity, or Claude — and whether it ranks on Google's first page, since 62% of brands on page one appear in ChatGPT answers.

If you are not a Brass-SEO customer, you can still run this audit manually. Check each H2 for a capsule, look for original data, verify your heading hierarchy, and confirm a FAQ section exists. It takes 15-20 minutes per page — longer than an automated report, but still worthwhile for your most important content.

To measure your AI visibility before and after running the audit, see How to Test Your Brand's AI Search Visibility.

For the audience-development side of AI citation visibility (the Preferred Sources button that drives 2x CTR on cited results in AI Mode), see Google Preferred Sources: How to Add the Button.

Frequently Asked Questions

What is an AI citability audit?

An AI citability audit evaluates whether a page is structured so that language models like ChatGPT and Perplexity can extract, attribute, and cite its content. It checks ten factors including answer capsules, original data, brand intent clarity, FAQ presence, scope clarity, and content structure.

Which factor should I fix first?

Answer capsules on H2 headings. Research shows 72.4% of ChatGPT-cited blog posts use this pattern, and adding them takes about 10 minutes per page. Add a FAQ section second — Q&A is the highest-cited format in AI responses.

Does fixing these factors also help with Google rankings?

Yes. Answer capsules improve your chances of winning Google's featured snippets. FAQ sections generate FAQ rich snippets through schema markup. Clean heading hierarchy helps Google understand your page structure. Most AI citability improvements are also traditional SEO improvements.

How do I know if my changes are working?

Track three things: GA4 referral traffic from chatgpt.com, perplexity.ai, and claude.ai; monthly manual query audits in ChatGPT and Perplexity; and your Google Search Console data for ranking improvements. Part 2 of this series covers the full tracking routine.

Do I need to audit every page on my site?

No. Start with your top 10 pages by impressions — these are already ranking for relevant queries and have the best chance of being picked up by LLMs. Use your Google Search Console data to identify which pages those are.

What is the difference between GEO and traditional SEO?

Generative Engine Optimization (GEO) focuses on structuring content for AI citation, while traditional SEO focuses on ranking in search engine results pages. There is significant overlap — 62% of brands on Google's page one appear in ChatGPT answers. The main GEO-specific techniques are answer capsules, content front-loading, and explicit brand positioning.

This is Part 3 of our LLM Visibility series. Part 1 covers content traits. Part 2 covers tracking. Part 4 is a before-and-after case study.

The research cited in this series comes from Adam Gnuse (answer capsule data), Paul DeMott (LLM tracking framework), and Princeton's GEO study (quotation and citation impact data). The AI Citability audit described in this post is a feature of Brass-SEO. For how it fits alongside traditional SEO tools, see What Brass-SEO Is (and What It Is Not).