What a GEO Audit Should Actually Contain

What the prompt list should look like

A GEO audit is only as good as the prompts it runs. The deliverable should start with 15 to 30 real buyer questions in your category, not your brand name typed into ChatGPT once. That means shortlisting questions, comparison questions, and problem aware questions a buyer types before they know which vendor names to search for. If the agency handed you a prompt list built from guesswork instead of your actual sales conversations, everything built on top of it points at the wrong things.

• Shortlisting prompts: "best [category] agency or tool for [use case]"
• Comparison prompts: "[competitor] vs [you]", "[you] alternatives"
• Problem aware prompts: written before the buyer knows which vendor names to search for
• Branded prompts: your name alone, to catch factual errors, not to prove visibility

Retrieval and citation, reported separately

Retrieval and citation are different events, and a real audit reports them as two numbers, not one blended visibility score. Retrieved means a model's search layer pulled a page into its context window while building an answer. Cited means the model actually named the brand or linked the page in the text a buyer reads. A page can be retrieved on most prompts in a category and cited on only a handful of them, and that gap, not the retrieval count, is where the real work lives.

If the report you get back shows one number labeled "AI visibility" with no split between the two, ask for the split before you accept the findings. For a version you can run yourself between paid audits, see how to measure AI search visibility and the fuller walkthrough in what an AI visibility audit should show you.

A source map, not a percentage

The audit should list, prompt by prompt, which specific URLs each engine cited, not just a count. That map turns a score into a plan: if a competitor's comparison page keeps showing up and yours does not, you know exactly which page to build or fix, and in what format. In practice, independent comparison pages and community discussions earn citations more consistently than a brand's own homepage, because models weigh corroboration from sources they did not write themselves. Ask for citations broken down by page type, listicle, comparison, review, forum, homepage, so you can see which format is winning the questions that matter to you.

Competitive share of voice, including names you didn't ask about

A real audit shows how often you appear across the same prompts as named and unnamed competitors, not just the one rival you asked about at kickoff. Tracked-prompt tools can supplement that picture, but they measure a sample of a market, not the whole field: the engines will name agencies and vendors nobody put on your competitor list, and a useful audit surfaces them instead of only tracking the ones you already know. If your audit only compares you to two competitors you named going in, you are seeing a narrower field than the one your buyers actually see.

A corroboration gap analysis

The audit should name specifically where third-party corroboration is thin: which claims about you live only on your own site, which categories have no independent source echoing them, and which competitors already have that corroboration built. This is the part most vendor audits skip, because it takes real research across press, reviews, forums, and directories rather than running your domain through a scoring tool. Managing that picture over time, correcting stale facts and building the missing corroboration, is what reputation management actually means in an AI search context.

Fixes tied to findings, not a generic roadmap

Every finding in the audit should map to a specific task, this fact is wrong on this cited page, request a correction here, not a line like "improve your online presence." A prioritized list, ranked by which gaps affect the most buyer prompts, is worth more than a deck of screenshots. Speed matters more than it used to: in the same G2 research, 69 percent of B2B software buyers had already chosen a different vendor than they originally planned based on what an AI chatbot told them, and 33 percent bought from a vendor they had never heard of before (G2). A finding that sits in a deck for a quarter is a finding that missed that buyer. Run the free AI visibility grader for a first pass, and treat it as the baseline a paid audit should improve on, not replace.

How often to re-run it

An audit is a snapshot the day it runs, and the market and the models keep moving under it. Re-run the same prompt list quarterly at minimum, and treat any audit older than two quarters as describing a market that has already shifted. If you want a live read on where your buyers' AI answers stand today, book a strategy session and we will run it with you.

The audit checklist

Keep this next to any audit you receive, from Corbelix or anyone else. A deliverable missing more than one or two rows below is not a finished audit yet.