The term "AI visibility audit" gets thrown around loosely. In practice, a real audit has a specific methodology, measurable outputs, and deliverables a client can act on. This is what that looks like in full.
Why methodology matters here
AI search is probabilistic. The same prompt can produce slightly different answers across sessions, models, and time. That variability means a casual audit ("I asked ChatGPT about your brand and here's what it said") isn't particularly reliable or repeatable.
A good AI visibility audit controls for this by running each prompt multiple times, across multiple platforms, and recording consistent patterns rather than single data points. The goal is a directional baseline, not a one-off snapshot. That distinction matters when you're trying to show progress over time.
Step 1: Visibility baseline
Start by running the client's target queries across ChatGPT, Perplexity, and Gemini. These should be queries that a realistic potential customer would actually ask: category queries, comparison queries, recommendation queries. Not brand name lookups.
For a project management tool, that might be: "what project management tool is best for agencies," "compare Asana and Monday," or "tools for managing client projects." For a Webflow agency, it might be "best Webflow agencies for SaaS," or "who builds Webflow sites for startups."
Record two things for each query: does the brand appear at all, and if so, in which position (first mention, second, mentioned briefly, etc.). This gives you the raw visibility data.
The Visibility Score comes from this step: the percentage of target queries where the brand appears at least once.
Step 2: Position analysis
Appearing in an AI answer and being the first recommendation are very different outcomes. A brand that appears in 70% of target queries but only as a third or fourth mention has weaker effective visibility than one that appears in 40% of queries as the primary recommendation.
For each query where the brand appears, note whether it's:
- The primary recommendation (named first, explained in most detail)
- A secondary option (mentioned as an alternative)
- A brief footnote (named in passing, minimal description)
This gives you an average citation position metric and a more honest picture of where the brand actually stands.
Step 3: Sentiment analysis
AI responses aren't just about whether you appear. They also say things about you, and those things can be accurate, outdated, incomplete, or actively unhelpful.
For each appearance, record whether the description is positive (accurate, favorable, specific), neutral (mentioned without meaningful characterization), or concerning (associated with limitations, outdated information, or framed in a way the brand wouldn't want).
Sentiment matters because a brand that appears in 80% of queries but is consistently described as "an older option" or "suitable for basic needs" has an AI visibility problem even with high raw visibility. The fix is different. It's usually a content problem, not a technical one.
Step 4: Competitor mapping
This step answers a question clients always care about: who else is in the answer, and how much of the AI's attention are they getting?
For each query, record which competitors or alternatives appear alongside the client. Then calculate Competitive Share: out of all brand mentions across all tested queries, what percentage went to the client versus competitors?
A client with 15% competitive share in a category where one competitor has 60% has a concrete, quantifiable gap. That framing tends to make the business case for the work faster than abstract talk about AI search trends.
Step 5: Technical readiness
This is the part of the audit that doesn't involve running prompts at all. It's an audit of the site itself, specifically whether it's set up to be properly understood and cited by AI systems.
Check each of the following:
Bot accessibility: can AI crawlers actually access the site? Check robots.txt for any rules that block major AI crawlers (Googlebot, GPTBot, ClaudeBot, PerplexityBot).
Sitemap: is there a valid sitemap.xml submitted to GSC, and does it include all key pages?
llms.txt: is there an llms.txt file that tells AI systems what the site contains and how to use it? Most sites don't have one. It's a straightforward addition with meaningful signal value.
Schema markup: does the site have Organization schema on the homepage? FAQPage schema on key pages? Article schema on blog content? These are the structured data signals that AI models rely on most.
Heading hierarchy: do page headings follow a logical H1 to H2 to H3 structure, or are they scrambled? Scrambled heading structures make content harder for models to parse.
Readability and page crawlability: are key pages actually indexable? Is the content human-readable without JavaScript rendering, or is it locked behind dynamic rendering that crawlers may not see?
AEO Copilot's Technical Scanner checks all of this automatically. It's the fastest way to run this step at scale without manually inspecting every page.
Step 6: Content gap analysis
The final step looks at which queries the brand should appear in but doesn't. These are the content gaps.
Take the full query list. For every query where the brand is absent, ask: does a page on this site answer this question clearly and directly? If the answer is no, that's a content gap. If the answer is yes but the page buries the answer in long prose, that's a structure gap.
Content gaps become the basis for a content roadmap. Structure gaps become a list of page optimization tasks. Both are immediate, actionable outputs.
For specific content changes that move visibility — restructuring pages, writing for AI extraction, building external citations — see How to show up on ChatGPT and other LLMs.
KPIs to track
After a thorough audit, you should have numbers for:
- Visibility Score (% of target queries with at least one brand mention)
- Average citation position (first, second, footnote, averaged across appearances)
- Sentiment breakdown (% positive, neutral, concerning)
- Competitive Share (% of all AI mentions that went to this brand)
- Technical readiness score (how many of the technical checks pass)
These are the baseline numbers. Every subsequent monitoring report updates them and shows direction of change. For a full breakdown of what each metric means and how to find LLM traffic in your analytics: AEO metrics that actually matter.
What the deliverable looks like
A client-ready AI visibility audit has four parts:
A 2-page executive summary. Non-technical. Visibility Score, main findings, top 3 priority actions. This is what goes in front of a CEO or CMO. Run a free audit first to generate the Visibility Score and technical baseline before you structure the full deliverable.
A prioritized technical checklist. Every technical gap found, ordered by ease of fix and estimated impact. This is what your dev or content team works from.
A competitive visibility snapshot. A table showing the brand versus top 3 competitors across the tested queries. Clear, visual, not ambiguous.
A 90-day action plan. Specific tasks, sequenced by phase. Technical fixes first (fastest impact), then content optimization, then ongoing monitoring setup.
The audit is the deliverable that earns the retainer. Clients who see concrete gaps, a clear competitor picture, and a specific plan to close the distance are the ones who stay engaged. See How to sell AI visibility as a service for how to structure the pitch and scope around this methodology.
Making it repeatable
The audits worth selling are the ones you can run again in 90 days and show measurable change. Manual prompt-testing doesn't scale to that cadence.
AEO Copilot runs the prompt monitoring automatically, surfaces the Visibility Trend, and flags changes in Sentiment or Competitive Share week over week. The methodology stays the same. The time it takes to run it shrinks significantly.
That repeatability is what turns a one-time audit into a long-term retainer. For more on building that revenue model, see How agencies sell AI visibility.
You can also start with a quick site check using the /website-audit before running the full manual methodology. For how agencies operationalize this with a dedicated tool, see AI visibility tool for agencies.