Back to the blog

AI Visibility Audit: How to Run One and Act on It

AI Visibility Audit: How to Run One and Act on It

AI Visibility Audit: How to Run One and Act on It

An AI visibility audit shows whether ChatGPT, Gemini, Perplexity and Google’s AI answers mention your company, recommend it, cite it, and describe it correctly when buyers ask about your category. You run one by building a realistic prompt set, testing it across platforms, recording every result in a log, tracing the sources behind the answers, and turning the gaps into a plan. This guide walks through each step and shows what to do with the findings.

Storyline Pros
cal.com
Turn AI Visibility Gaps Into Action
Storyline Pros combines elite PR with GEOview AI Visibility Technology to help companies earn media placements, community authority, and AI search recommendations.
Book a strategy session

TL;DR

  • An AI visibility audit measures four things per prompt: mentions, recommendations, citations and answer accuracy. It is not a rankings report.
  • Build 20 to 40 prompts from real buyer language: category, use-case, comparison and alternative questions.
  • Run every prompt on each platform and log it. Results shift between runs, so look at patterns, not single answers.
  • The sources behind the answers matter more than the answers. Find which domains AI engines pull from, then earn your place there.
  • Repeat the audit on a fixed schedule so you can see whether your work moves the numbers.

Table of Contents

What an AI visibility audit measures

A traditional SEO audit asks whether your pages can be crawled, indexed and ranked. An AI visibility audit asks a different question: when a buyer asks an AI assistant who to trust, does your name come up?

That difference changes what you count. Ranking position is replaced by four signals:

  1. Mentions. Does the answer name your brand at all?
  2. Recommendations. Is your brand suggested as an option, and in what position among the others?
  3. Citations. Does the answer link to or draw on a page that you own or that is about you?
  4. Accuracy. Does the answer describe what you do, who you serve and how you are positioned correctly?

PartnerStack’s audit framework lists the same core measures: brand mentions, AI citations, message accuracy and share of voice across multiple AI platforms. Share of voice is your mentions as a portion of all brand mentions across your prompt set, so competitors enter the picture from the first pass.

One practical consequence: a page can rank well on Google and never appear in an AI answer, and the reverse also happens. Treat the two audits as separate checks, not one report with a new label.

For the bigger picture on how startups get recommended, see our guide to AI visibility for startups.

Step 1: Define the scope

Before you type a single prompt, decide three things.

Which platforms. Start with the assistants your buyers use. For most B2B tech buyers that means ChatGPT, Gemini, Perplexity and Google AI Overviews. Add Microsoft Copilot or Claude if your customers work in those environments. Keep the list short enough that you can repeat it every cycle.

Which topics. Pick the two to four categories where you want to be the obvious answer. A seed-stage fintech, for example, may care about one category, such as the problem it solves, and one adjacent use case. An audit that tries to cover everything produces a report nobody acts on.

Which competitors. List three to five companies that buyers compare you with. You will track them in every prompt. Include the ones you consider weaker than you; AI engines do not rank by who you think deserves the spot.

Step 2: Build a prompt set from real buyer language

The prompt set decides whether your audit is useful or decorative. Founders often test prompts written in their own product vocabulary, and those tell you little because buyers do not talk that way.

Two sources in the research agree on method. PartnerStack’s guide suggests pulling from customer support and sales material: go into Zendesk, tickets or customer emails and use the voice of the customer to reverse engineer prompts tied to real buying intent. Screpy’s guide organizes the set into four groups: category, use-case, comparison and alternative prompts, written in plain buyer language with realistic constraints such as company size, budget range, industry, technical skill level or location.

Here is how that looks for a fictional Series A security startup:

Prompt type Example prompt
Category What are the best tools for cloud access management?
Use case How can a 60-person fintech company meet SOC 2 access requirements without a dedicated security hire?
Comparison Compare [your brand] and [competitor] for a mid-sized engineering team.
Alternative What are alternatives to [competitor] for a startup with a small security team?

A good working size is 20 to 40 prompts. Fewer than that gives noisy results, and many more makes repeating the audit a burden. Weight the set toward the prompts closest to a buying decision: comparisons and alternatives, where a recommendation turns into a shortlist.

Also write unbranded prompts. Branded prompts test whether the model knows you exist. Unbranded ones test whether it recommends you, which is the commercial question.

Step 3: Test the prompts and log the results

Run each prompt on each platform you chose. Use a clean session where you can, with no prior conversation, so earlier context does not shape the answer. Log everything in a spreadsheet with one row per prompt per platform.

Useful columns:

  • Date and platform
  • Prompt and prompt type
  • Brand mentioned (yes or no)
  • Position among recommended options
  • Competitors named
  • Sources cited or linked
  • Accuracy of the description (correct, partly wrong, wrong)
  • Sentiment (positive, neutral, negative)
  • Notes on wording, such as how the model describes your category

Keep the column names fixed from the first cycle. You will compare against them later, and changing the structure mid-way breaks the trend line.

Read the limits of the data

AI answers are not stable, and any honest audit says so up front. Microsoft Clarity notes that responses change with model version, session context, retrieval updates and regional or platform differences, so two identical prompts sent minutes apart can produce different citations. The same post points out that simulated prompts cannot capture how real users phrase questions, and that prompt simulators see only the final answer, not the retrieval process that chose the sources.

Three habits handle this:

  1. Run each prompt more than once. Record the share of runs where you appear, such as 3 of 5, rather than a single yes or no.
  2. Report ranges and trends. One audit is a baseline. The direction across several audits is the finding.
  3. Pair the audit with real data. Look at referral traffic from AI platforms in your analytics and ask new customers how they found you.

Step 4: Find the sources shaping the answers

The answer is the symptom. The sources are the cause. If Perplexity names three competitors and not you, the useful question is which pages it used to build that answer.

Search Engine Land’s off-page audit method gives a workable process: pull the top 40 domains by how often engines retrieve them, review the queries engines fan out to, reclassify the source categories by hand, add domain authority and traffic for context, and flag competitors whose sites appear in a large share of chats.

When you sort those domains, you will usually see the same groups:

  • Editorial and trade media that review or list companies in your category
  • Listicles and “best of” roundups, which AI engines lean on for recommendations
  • Community threads, such as Reddit discussions where practitioners compare tools
  • Podcasts and interview transcripts where founders explain what they do
  • Your own site and documentation
  • Review and directory sites

This is category analysis in practice. You are mapping where your category’s story is told, who tells it, and where your company is absent from that story.

The same Search Engine Land piece offers a useful diagnostic. If you are weak only in Google AI Overviews, the likely cause is rankings; if weak only in ChatGPT, thin entity presence and mentions across the web; if weak only in Perplexity, citation sourcing and freshness; and if weak on one topic across every engine, a content gap. Use it to decide where to spend first instead of trying to fix everything at once.

For how mentions on third-party pages turn into citations, read our piece on brand mentions and AI citations. For the Google side specifically, see how to get cited in Google AI Overviews.

Step 5: Score the gaps and rank the fixes

Now turn the log into a short list of problems. Most audits for early-stage companies surface the same four.

You are absent. The model does not mention you in unbranded prompts. The cause is almost always a thin footprint in the sources it trusts. The fix is earned coverage in those sources, not more posts on your own blog.

You are present but misdescribed. The model gets your category, customer or capability wrong. This is a narrative problem: public sources describe you inconsistently, so the model blends them. The fix is a clear, repeated positioning across your site, press and community presence.

Competitors own the comparison prompts. They appear in listicles and reports that you do not. The fix is to earn inclusion in those lists or to publish the original data that others cite.

Your pages are not cited. The model uses third-party material about your category and ignores your site. The fix is evidence pages: specific, structured pages that state facts a model can quote. Our guide to AI search optimization explains the approach.

Rank each gap by two factors: how close the prompt sits to a purchase, and how much effort the fix takes. A comparison prompt where you are missing outranks a broad awareness prompt, and a wrong fact on your company description outranks both because it misleads the buyer directly.

Step 5: Score the gaps and rank the fixes

How often to repeat the audit

No official body sets an audit schedule, so choose one that fits your pace of change. A reasonable rule: run the full audit quarterly, and run a light check of your 10 most important prompts monthly. Add an out-of-cycle run after a funding announcement, a product launch, a rebrand or a major press hit, because those events change what public sources say about you.

Because answers vary, compare like with like. Same prompts, same platforms, same columns, same number of runs. When you add new prompts, keep the old ones so the trend stays readable.

Track a small set of numbers over time:

  • Mention rate across the unbranded prompt set
  • Share of voice against your named competitors
  • Number of distinct third-party domains citing or describing you
  • Share of answers that describe you accurately
  • AI referral traffic and the pipeline it touches

Only the last one ties to revenue, which is why it belongs on the same page as the audit scores.

Our perspective: audit the narrative, then earn the sources

We built Storyline Pros around one observation: AI engines repeat what trusted sources say about a company. They do not reward the company that publishes the most pages about itself.

That is why we treat an AI visibility audit as a narrative audit. The question is not only whether a model mentions you. It is what story the web tells about you, who tells it, and whether that story is consistent. Narrative engineering is the work of fixing that story at the source: tier-one media, podcast features, community authority and the lists and reports that models cite.

Our GEOview AI Visibility Technology exists to measure this and to track the impact of the work, so a placement is tied to a change in how AI search describes you. Our GEO strategy runs across a 6-channel ecosystem, because one channel rarely moves an answer on its own.

We also think the model of payment matters. Many agencies sell retainers and report activity. We bill per delivered placement, podcast appearance or syndicated piece, backed by a performance-based guarantee. If an audit shows the gap, the next step should be output you can see, not a deck.

Nik

What to do after the audit

The audit pays off only when it changes what you do next. A practical 90-day order of work looks like this, and it is a sequencing suggestion, not a commitment period:

  1. Fix factual errors first. Correct your own site, profiles and directory listings so the basic facts match everywhere.
  2. Publish evidence pages for the two or three prompts that matter most, written so that a model can lift a clear, sourced statement.
  3. Target the source gaps. From your domain list, choose the publications, lists, communities and podcasts that already shape answers in your category, and pursue coverage there.
  4. Create something worth citing. Original data, a category report or a practical benchmark gives journalists and list authors a reason to reference you.
  5. Re-run the audit on the same prompts and compare.

If you want to see how measurement and earned placements fit together, our article on ChatGPT brand visibility shows the measure-first approach, and the GEO playbook on citation share goes deeper on the strategy.

Get an audit that ends in placements

If you are a seed to Series B founder, or an executive at a growth-stage or public tech company, you can run the steps above yourself. The harder part is the work after the audit: getting the right publications, podcasts, lists and communities to cover you.

That is where we come in. Storyline Pros combines category analysis with GEOview AI Visibility Technology to show where your category’s story is told and where you are missing from it. We then earn the placements that close the gap, billed per delivered result and backed by our performance-based guarantee.

Storyline Pros

Ready to see where you stand? Book a strategy session and we will walk through your category, your competitors and the sources AI engines use. If you are weighing agencies, our AI visibility agency guide explains how to compare them, and our page on getting mentioned by ChatGPT shows how we approach it.

Common audit mistakes

  • Testing only branded prompts. You will see yourself and conclude all is well.
  • Using your internal vocabulary. Buyers describe the problem, not your feature names.
  • Treating one run as the truth. Responses vary, so record frequency.
  • Skipping the sources. Without them you know the score but not the cause.
  • Ignoring accuracy. A wrong description in front of a buyer can cost more than silence.
  • Auditing once. A baseline with no follow-up cannot show whether your work is moving anything.

FAQ

How long does an AI visibility audit take?

A basic version, covering 10 to 20 prompts across three platforms, fits in an afternoon. A full audit with 30 or more prompts, multiple runs per prompt and a source analysis takes longer, mostly because of the logging and the review of cited domains.

Is an AI visibility audit the same as an SEO audit?

No. An SEO audit checks whether search engines can crawl, index and rank your pages. An AI visibility audit checks whether AI assistants mention, recommend and accurately describe your brand, and which sources they rely on. The two overlap, since strong rankings can help in Google AI Overviews, but they answer different questions.

Which AI platforms should I include?

Start with ChatGPT, Gemini, Perplexity and Google AI Overviews, then add any assistant your own buyers say they use. Log results for each platform separately, because weakness on one platform points to a different cause than weakness on all of them.

How many prompts do I need?

Twenty to forty prompts is a workable range for most startups. Spread them across category, use-case, comparison and alternative questions, and weight them toward the ones closest to a purchase decision.

What should I do if AI describes my company incorrectly?

First correct the facts on your own site and profiles. Then look at the third-party sources the answer drew on and update or earn better coverage there, since the model repeats what those sources say. Re-run the same prompts afterward to confirm the description has changed.

Sources

Ready to make the market pay attention?

We take on a small number of companies each quarter.
Book a strategy session to explore if it's a fit.