# What Does an LLM SEO Audit Reveal About Your Website?



**Short answer:** it reveals what Google Analytics can't, whether AI engines know you exist, whether they cite you, whether they recommend you, and why your competitors win the answers you should be winning. An LLM SEO audit doesn't measure rankings. It measures the buyer. And the first thing it usually reveals is a gap most businesses never imagined: AI engines can recognize your brand perfectly and still never recommend it.
			

## The 30-to-1 Gap: The Revelation That Starts Every Audit
		

In a study of 112 startups and 2,240 queries, ChatGPT got the names of brands right 99.4% of the time, but only suggested brands when searching for them by category 3.32% of the time. A ratio of 30:1 on the number of people known to those selected.


It's that gap that an LLM SEO audit exists to explain. Your brand might be known by the model but not the buyer – recognition and recommendation are two things, two signals, at two stages. This typically would not be discovered through a [traditional SEO audit](https://thechimeramarketing.com/search-engine-optimization/); you might have great rankings and poor recommendations. The purpose of the audit is to demonstrate the gap between the two and the particular reason for the gap.
			

## What an LLM SEO Audit Actually Measures
		

A proper audit runs in four stages, and each stage reveals something a keyword tool cannot detect.


### Stage 1: The prompt library


15-20 consumer-intent questions that reflect actual consumer questions and not keywords, but questions asked of AI engines. Don't use brand-name prompts; test category and comparison queries are ones where the buyer hasn't made up their mind yet. This is the set of queries you are evaluated against.


### Stage 2: The four-engine run


At least four engines: [ChatGPT, Perplexity, Gemini, and Google AI Overviews](https://thechimeramarketing.com/google-ai-overviews-are-stealing-your-traffic-heres-how-to-get-it-back/). Perplexity is particularly helpful in that it provides named citations with real links, allowing you to trace back and find out what sources AI has confidence in. Every engine is a different judge, and they each have their own preferences, and only 12% of cited domains are shared across engines, meaning one-engine audit is one-engine verdict.


### Stage 3: The four-level score


You will be rewarded with points at four levels: mention (your brand is mentioned), citation (you are mentioned as a source), recommendation (the engine recommends you as a source), or source absorption (your content's facts are included in the engine's answer, with or without attribution). For most businesses, this is the level they end up at, which is level one or two, and which is only known as the business, maybe mentioned, but rarely recommended. AI will only mention a couple of brands for a query. The shortlist is brief, and the audit indicates if you're on it.


### Stage 4: The five gap types


Point out gaps (the engine does not mention you), prompt gaps (the engine mentions categories of the whole question you do not appear in), source gaps (categories of the whole question where you never appear in the source citation), citation gaps (categories of the whole question you appear in, but never in the body of the source citation), and narrative gaps (categories of the whole question you appear in, but the description of you in the source is incorrect, obsolete, or incomplete). There are different fixes for each gap, and the value of an audit is knowing which gap you are facing; most businesses guess wrong.
			

## Revelation 1: Your Best-Ranking Pages Are the Least Cited
		

This is the discovery that contradicts most of the preconceptions. In fact, Semrush data from July 2025 shows that almost 90% of ChatGPT citations are from pages in Google's first 21 positions. The less popular your page is, the more likely it is to be cited.


For LLM citations, backlinks, keyword density, and internal linking barely make a dent in earning a top-3 Google ranking. So the study of 75,000 brands by Ahrefs discovered that brand web mentions have a higher correlation with AI visibility than backlinks, domain rating, and any other on-site aspect examined. This inversion can be seen directly on the audit: by looking at the number of times you're cited when you're ranked, and the page-by-page breakdown of where you're not matching up between ranks and citations. It can be the pages you've put the most effort into that the engines are leaving out, and it's not because of an answer; it's because it's a results page.
			

## Revelation 2: The Answer Is in the First 30%
		

From his analysis of 1.2 million verified ChatGPT citations, Kevin Indig discovered that 44.2% of all citations generated by LLMs are from the top 30% of content. LLMs search for the answer before determining whether to cite their source. The specificity of the answer trumps the depth of the topic each time.


It calculates each candidate page based on the following simple citation scorecard: answer-first formatting (the direct answer, not the conclusion, appears in the opening section), explicit H2 questions matching the way buyers ask them, named sources per section, and schema markup. A page with a score of 0-1 should be structurally reconstructed, not copy edited. For most businesses, the answer is hidden in 800 words of introduction and the engine, which reads the first 30%, moves on, never finds it.


The action step is to export 90 days of data from [Google Search Console](https://search.google.com/search-console/about) and narrow it down to the pages that have been seen more than 500 times, but haven't had a CTR above 3%. These are the top-visibility, low-click pages that AI answers are already stealing clicks from. They're the ones with a lot of traffic that are the highest leverage points for an audit, as you already have traffic and the answer is already getting stolen.
			

## Revelation 3: How the AI Describes You Is Often Wrong
		

An audit doesn't find anything missing; it finds something wrong. The entity check includes a simple question: What does AI say about you, your work, and your clientele? The answers are often wrong, out of date, or incomplete, and no one asked the engine to describe them, so the business didn't know.


This is the story hole, and this one is the most important. A model that incorrectly assumes you're a generalist will send you suggestions for the wrong keywords and exclude you from the right. A model working from a 3-year-old description will refer to services you no longer provide. The audit reveals the engine's perception of your business, and the remedy is entity work: uniform NAP, organized data, verification from independent sources, and content that clearly and repeatedly answers the question, "What do you do?
			

## Revelation 4: The Crawlers Might Not Even Be Let In
		

The audit also examines the plumbing: can the AI crawlers access your site at all? – before any of the content analysis issues. The answer is often no, and the business does not know. In July 2025, Cloudflare started listing known AI crawlers and began to block them on new domains by default. Googlebot, ClaudeBot, or PerplexityBot could be accidentally omitted by a robots.txt file designed for Googlebot. Googlebot can be served a full page while ChatGPT is served a blank page.


The technical layer dictates the ability of the engines to evaluate your content, as answer extractability, schema completeness, presence of llms.txt, and crawler access are all scored in the audit. Businesses that have a score of 0/20 for crawler access don't have a content problem. There is an issue with its door.
			

## Revelation 5: Citations Churn, Fast
		The final revelation is the one that changes how you think about the audit itself. Passionfruit measured 68% of cited sources churning month to month across 11.2 million citations. Two-thirds of what's cited this month won't be cited next month. The audit is a snapshot, not a permanent verdict, and that's actually good news, because it means the positions are contested, not owned. A competitor's citation today is your opportunity next month, if you fix the gap the audit found.
			

## What the Audit Does Not Reveal
		

The limits are required for honesty. The LLM SEO audit can't provide you with any particular deal attribution; it's not possible, yet. It can't guarantee a citation, since there's no one controlling the models' retrieval decision. But it can't replace the traditional SEO audit: whether rankings, links, and technical site health are the discovery layer will not change, and a GEO audit will assess if AI answers will recommend you, a different question with different answers.


What it can do is take the guesswork out of things and replace it with evidence. Every check results in a finding; every finding has a fix; every fix is next to its impact. Therefore, the audit is not the last step, but the first step.