What Does AI Say About Your Brand? Run This 60-Minute Audit
By Guillaume Rufenacht, RouterGrowth · published aug 2026
Before paying anyone for AI visibility, spend an hour measuring it. A spreadsheet and three chat windows produce a defensible baseline: what assistants say about you, who they recommend instead, and which sources they're reading. Here's the exact procedure, free to steal.
Build the question panel (15 minutes)
Write 15–20 questions in four groups, phrased the way a buyer talks to an assistant, not the way keywords look. Category: "what's the best [category] for [your ICP]?". Comparison: "[competitor] vs [you]", "[you] alternatives". Problem: "how do I fix [the problem you solve]?" (these catch buyers who don't know the category exists). Direct: "is [you] legit?", "what does [you] cost?". Freeze the panel; the value is in running the same questions every month.
Run it (20 minutes)
Same panel, three surfaces: ChatGPT with search enabled, Perplexity, and Google's AI Overviews or AI Mode. Fresh chats each time, so your history doesn't contaminate the answers. Record two things per question per assistant, verbatim: the answer's brand mentions, and every cited source. One row per question per surface. Resist cleaning up the answers; the wording ("popular", "budget option", "users report issues with") is data.
Read the sources, not just the mentions (15 minutes)
The cited domains are the map of your category's evidence layer, and it's usually short: Reddit plus a handful of comparison and review sites doing most of the work (why Reddit dominates is documented in our Reddit SEO guide). Note the specific threads and pages that produce your competitors' mentions. That short list of URLs is where your next quarter of visibility work actually lives; everything else is commentary.
Score competitors on the same panel (10 minutes)
For each competitor: mention rate across the panel, sentiment of the mention, and which sources carry them. Then compute the only summary number worth tracking: share of voice, your mentions divided by all brand mentions on the panel. That's your baseline. It will be humbling or reassuring, and either way it's now a number instead of a feeling.
Act on the three gap types
Absent (you're not mentioned): build mentions in the sources that carry your category; the playbook is in the recommendation guide. Wrong (an assistant states something false): trace the citation it shows, correct it at the source where possible, and publish the accurate fact on your own citable pages and llms.txt; assistants follow their sources. Fragile (mentioned, but carried by one old thread): broaden the base before it drifts. Then re-run monthly, same panel: answers move (Semrush documented citation weightings swinging massively within two months), and the trend is the product.
Frequently asked questions
How often should we re-run the audit?
Monthly on the frozen panel, plus after launches, renames or pricing changes. Drift between runs is normal; platforms retune constantly. The trend across months is the signal.
Do we need a monitoring tool for this?
Not to start; the spreadsheet is the method. Tools earn their keep at scale: many brands, many markets, daily cadence. Our AEO cost guide covers when that spend makes sense.
What if an assistant says something false about us?
Trace the source it cites, fix it at the source where you can (an outdated thread deserves an honest reply, an old article a correction request), and state the accurate fact plainly on your own pages. Assistants repeat their evidence; change the evidence.
What does the done-for-you version add?
Run the procedure above and you own the baseline; that is genuinely the hard 80%, and if you have the hour, take it. Our <a href='/ai-search-audit'>AI Search Audit</a> ($1,500, ten business days) adds the parts an hour cannot buy: a 20-question panel mined from real search and community data rather than guessed, all three surfaces run inside one window so the numbers are comparable, every cited source ranked by whether you can realistically win it, your site checked for whether assistants can read and quote it at all, the search openings you already half-own, and a two-page memo naming the one channel to fund first. You also keep the frozen panel and the raw CSVs, so the second month is a re-run rather than a fresh opinion.
The AI Search Audit runs a frozen question panel across ChatGPT, Perplexity and AI Overviews, maps the sources they cite, finds your search openings and names the channel to fund first. Ten business days, and the fee comes off your first retainer invoice if you go on to hire us.