How to Audit Your B2B Brand's Visibility in ChatGPT, Claude, Gemini, and Perplexity
Last updated
A B2B AI visibility audit runs a defined query matrix — buyer questions, persona variants, buying stages — across ChatGPT, Claude, Gemini, and Perplexity, then scores each answer for whether your brand is named, positioned accurately, and cited. Resonate Labs turns that score into a prioritized fix list.
Map the buyer conversations that matter
Every audit starts with the conversations your buyers are actually having with AI. Pulling a generic list of keywords misses the point — B2B buyers ask messy, scenario-shaped questions like "what's the best HRIS for a 250-person services firm with multi-state payroll?" not "best HRIS." Before running a single query, document the buyer personas you sell to, the two or three problems each persona brings to the conversation, and the deal-stage at which AI gets consulted.
A useful default split is five personas, three stages each: problem framing, vendor shortlist, and trust-check on a named vendor. That yields roughly 15 conversation types per persona — a workable base for the query matrix in the next step. The map is not a keyword list. It's a set of buyer intents expressed as natural questions, because that's how LLMs see them.
Run the query matrix on every major surface
Once the map is in place, run each query on every major AI surface and record the raw answer. At minimum: ChatGPT (both with and without web browsing on), Claude, Gemini, and Perplexity. Each platform has different training cutoffs, retrieval behaviors, and citation styles — a brand that's strong in Perplexity can be invisible in Claude, and the difference is rarely a content problem. It's usually a structured-data or authority-signal problem on specific pages.
Run each query three times across separate sessions to control for temperature variance. Save the full answer text, any URLs cited, the timestamp, and which platform returned it. The goal isn't a single snapshot — it's a baseline you can re-run monthly to detect movement.
Score visibility across four dimensions
For each answer, apply a four-dimension score: named (is your brand mentioned), positioned (is the description accurate — strengths, ICP, differentiators), cited (does the answer link to a page you control), and compared (does the AI name you alongside the right competitors, or against vendors you've already out-matured). Each dimension gets a yes/no. A brand that's named 80% of the time but positioned correctly only 30% of the time has a narrative problem, not a visibility problem — and the fix is very different.
Roll the four scores up to a persona × stage grid. A 5-persona × 3-stage grid gives you 15 cells, each scored 0–4. Patterns emerge fast: usually two or three cells are the bottleneck for the entire pipeline.
Which AI surfaces matter at which buying stage
Which AI surface matters most depends on what stage the buyer is in. The table below is the default Resonate Labs audit frame — tune it to your category if your buyers skew technical or executive.
| Buying stage | Primary surface | Secondary | What it's answering |
|---|---|---|---|
| Problem framing | ChatGPT | Gemini | "What should I be thinking about when my team has this problem?" |
| Category learning | Perplexity | ChatGPT (browsing) | "What are the categories of solutions, and how do I evaluate them?" |
| Vendor shortlist | Perplexity | ChatGPT | "Who are the top 3–5 vendors for my profile?" |
| Vendor trust-check | Claude | Gemini | "What do customers say about this specific vendor?" |
| Negotiation prep | ChatGPT | Claude | "What pricing and terms are reasonable for this category?" |
Find the citation gaps that actually move deals
A citation gap is the cell in the scoring grid where AI answers the buyer's question well but doesn't mention you. The instinct is to assume this is a content quantity problem — "we need to write more." Usually it isn't. Most gaps are content-attribution problems: the information exists on your site but isn't structured, titled, or cross-linked in a way that AI platforms can confidently lift from.
The highest-leverage fixes cluster into three types. Technical: missing <main>, <article>, schema.org blocks, or author attribution on the exact pages buyers' queries would land on. Page-level: H1 and meta descriptions that don't state the claim AI needs to cite. Topical: no single page that owns the buyer's specific question — your answer is spread across a pillar page, two blog posts, and a comparison PDF nothing indexes.
Frequently asked questions
How is a GEO audit different from an SEO audit?
An SEO audit asks "does Google rank this page." A GEO audit asks "does an AI platform name us in an answer." The surfaces are different — Google indexes pages; LLMs ingest and summarize content. The fixes are often in the same file (schema, structured content, clear claims) but prioritized differently.
How often should we re-run the audit?
Monthly, at minimum, for the first six months after you start making fixes. AI platforms re-crawl and re-index at different cadences, and you want to detect movement in weeks, not quarters. After six months, quarterly is usually sufficient.
Can AI answers be influenced, or is this basically waiting for AI to notice you?
They can be influenced, deliberately and measurably. The mechanisms are different from SEO: schema.org markup, canonical author attribution, clear declarative claims, and citation-worthy formatting all move the needle. Passive waiting is the most expensive strategy.
Do we need separate audits for each AI platform?
One audit with per-platform scoring is enough. The query matrix and buyer map don't change; only the platform-specific results do. A single consolidated view makes prioritization easier.