Your AI Visibility Audit Is Measuring the Wrong Layer

Most AI visibility audits treat the problem as one score to improve. They are not. AI visibility is a stack of at least three distinct failure layers, each with different symptoms, different fixes, and different owners inside your org. Running all seven audit steps when your actual problem is confined to one layer wastes your week and misattributes the failure. I have watched teams optimize extractability for months when their content was never being crawled in the first place. Here is the diagnostic I use to identify the binding constraint before spending time on a full audit.

AI Visibility Is Three Layers, Not One Metric

Duane Forrester wrote the clearest articulation of this in June 2026: "The mistake is treating AI visibility as a single problem when it isn't. There are three structurally different layers between your brand and the answer a user receives, each with its own failure modes, its own fixes, and increasingly its own organizational owner."

Those three layers:

1. Retrieval — Can AI engines access and parse your content? This is crawlability, robots.txt policy, render method, and chunk-friendliness. If you fail here, nothing downstream matters.
2. Entity and knowledge — Does the model know what your brand IS? This is entity resolution: whether the model can distinguish your company from competitors, correctly attribute your claims, and connect your brand to the right category. Microsoft Research has documented that standard retrieval-augmented generation struggles to "connect the dots" across chunks — it retrieves relevant text but cannot reason about entity relationships without structured knowledge.
3. Answer positioning — When the model retrieves and recognizes you, are you the recommendation, the citation, or the footnote? This is where content structure, authority signals, and competitive positioning determine whether you appear as the answer or as background context.

Maria Dykstra's measurement framework makes the downstream cost of this confusion concrete: "GA4 measures clicks. AI decisions happen before the click. You are measuring the last 20% of the path to purchase." Traditional analytics only see the answer-positioning layer — by the time a buyer clicks through from an AI response, retrieval and entity resolution already decided which brands made the shortlist.

The 10-Minute Layer Diagnostic

Before running a full 50-prompt audit, spend 10 minutes identifying which layer is actually broken. Here is the sequence I run:

Layer 1: Retrieval check (3 minutes)

Check your server logs for GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, and Applebot activity in the past 30 days. If you see zero hits from these crawlers on your key content pages, your problem is retrieval. Check robots.txt for blocks, check your CDN/WAF for bot rate limits, and verify that key content renders in crawlable HTML rather than client-side JavaScript.

If AI crawlers ARE hitting your pages regularly, your retrieval layer is likely functional. Move to Layer 2.

Layer 2: Entity resolution check (4 minutes)

Ask ChatGPT, Perplexity, and Gemini: "What is [your company name]?" and "What does [your company name] do?" Compare the answers to reality. If the model confuses your company with a competitor, describes a product you do not offer, or gives vague non-answers, your entity layer is broken. The model cannot recommend what it cannot identify.

AnswerMentions' audit methodology separates recommendations from citations from casual mentions for exactly this reason. A brand with broken entity resolution might get mentioned (the model saw your name somewhere) but will never get recommended (the model does not understand you well enough to stake its answer on you).

If the model describes your company accurately, move to Layer 3.

Layer 3: Answer positioning check (3 minutes)

Run five buyer-intent queries in your category — the "best [category] tools" and "how to solve [problem]" queries your customers actually ask. If you appear in answers but only as a list item rather than the recommended solution, your issue is answer positioning. The fix is content structure, citation architecture, and competitive authority — not more pages.

Toolsolved's 2026 analysis found that "sites AI engines crawl and cite generate 3.2x more human traffic than those they ignore." But that statistic only applies when all three layers are functional. Optimizing Layer 3 with Layer 1 broken is like rewriting your sales deck while your website is down.

Why the Entity Layer Is Where Most Operators Go Blind

In my experience working with B2B brands on AI visibility, retrieval is the easiest layer to diagnose and fix — it is binary. Either crawlers access your content or they do not. Answer positioning is the most-discussed layer because it is the most measurable with existing tools.

Entity resolution is the layer that produces the most misdiagnosis because its symptoms mimic the other two. A brand with poor entity resolution will often show zero citations and conclude "we are not being retrieved" when the actual problem is that the model retrieves their content but cannot attribute it correctly, so it cites a competitor instead.

Astiva AI's research found that content with strong E-E-A-T signals — particularly Experience and Trustworthiness — receives 5.2 times more AI citations than content without them. But E-E-A-T is not a content-formatting exercise. Experience requires verifiable first-hand evidence. Trustworthiness requires independent corroboration. Both are entity-layer properties: they depend on the model recognizing WHO made the claim and WHETHER that entity is credible.

This is where Machine Relations operates — building the independent corroboration and entity clarity that lets AI models resolve your brand correctly. But you should only invest there after confirming that retrieval works and that entity resolution is actually your binding constraint.

The Execution Move

Run the 10-minute diagnostic above before your next full audit cycle. If you discover your problem is Layer 1 (retrieval), the fix is technical: robots.txt, WAF rules, SSR. If Layer 2 (entity), the fix is earned media and structured data — third-party publications that describe your brand consistently. If Layer 3 (answer positioning), the fix is content architecture — answer-first structure, comparison tables, and source density.

Stop running all three fixes simultaneously. Diagnose first.

FAQ

How do I know which layer my AI visibility problem is on?

Run the 10-minute diagnostic above in sequence. Layer 1 shows in server logs (no crawler hits = retrieval failure). Layer 2 shows in direct entity queries (model misidentifies your brand = entity failure). Layer 3 shows in buyer-intent queries (you appear but are not recommended = positioning failure). Fix the earliest layer that fails.

Can I skip the diagnostic and just run a full audit?

You can, but you will spend equal effort on all layers when one layer likely accounts for 80% of your gap. Maria Dykstra's research found that traditional metrics cover only 20% of the AI decision path. A full audit without layer diagnosis tells you WHAT is broken but not WHERE in the stack to prioritize — which means your team optimizes everywhere at once instead of fixing the binding constraint first.