AI visibility audit for B2B SaaS
Find out what ChatGPT, Claude, Perplexity, Gemini and Google's AI say about you when a buyer asks who to use. Then get a ranked list of what to fix, in the order that actually matters.
Every "book a call" audit page on the internet describes the audit the same way: comprehensive, actionable, tailored. None of them show you the actual thing. This page does. It walks through the same report structure I use on every real engagement, and shows a finding formatted exactly the way a real one reads, priority, what's wrong, why it matters, the fix, and the effort it takes.
The AI-visibility section (part 5) checks the same factors controlled research has found predict whether an AI system cites a brand at all, rather than a competitor, machine-readable pricing, direct-answer content, and passage-level citability chief among them (SIGIR 2026, arXiv:2605.25517).
Every report has the same seven sections
No two audits find the same problems, but the shape of the report never changes. That consistency is deliberate, it's what makes the report something a developer or a writer can actually pick up and execute against, not just read.
The short version
Three to five sentences. What's working, what's not, and the single biggest opportunity. Written to be read even if nothing else is.
Top priorities, ranked by impact
Five to eight findings, ranked by what actually moves revenue, not by how many issues exist in a category. Each one gets what's wrong, why it matters, the fix, and an honest effort estimate.
Technical health snapshot
Crawl coverage, indexing status, Core Web Vitals, broken links, redirect issues. Numbers, not adjectives.
Content gaps
What buyers are actually searching that the site doesn't answer. What's thin or outdated. What's already working and should be doubled down on.
AI visibility (GEO)
The direct answer: does this brand show up when buyers ask ChatGPT, Claude, Perplexity, Gemini, or Google's AI who to use. What's blocking it. The fastest wins.
Authority and citations
Who links to you, who mentions you without linking, and where your competitors get talked about that you don't. Referring domains, the junk ones included, and whether the profile is growing or quietly shrinking.
The fix timeline
Every priority translated into a sequence: what to ship this week, what comes next, and what's genuinely a later-stage move. Concrete, not a pitch.
Where I test your visibility
Five engines, checked by hand. No API sampling, no dashboard averaging. You get the raw results with dates, so you can run the same test yourself later and see whether anything moved.
| Engine | What gets recorded |
|---|---|
| ChatGPT | Whether you're named, how you're described, and which source it leaned on |
| Claude | The same, plus how it handles a direct comparison against a competitor |
| Perplexity | Named sources, since Perplexity shows its citations and most others don't |
| Gemini | Whether you appear, and how Google's own index feeds the answer |
| Google AI Overviews and AI Mode | Whether the overview replaces the click you used to get |
How I pick the prompts you get tested against
The prompts decide the result, so guessing at them makes the whole audit worthless. I build the list in four groups, using your positioning and your actual search data rather than what sounds plausible.
Category prompts
How somebody asks for your product without knowing your name. "Best tool for onboarding a distributed team." These are the ones you most want to appear in, and most often don't.
Comparison prompts
You against the two or three names a buyer would shortlist alongside you. This is where a model repeating an outdated or wrong description of your product does real damage.
Problem first prompts
The symptom before the solution. Somebody describes what is going wrong before they know what category fixes it, which is how a surprising share of buying journeys now start.
Branded prompts
Your own name. Models get pricing, positioning and headcount wrong about companies with perfectly clear websites more often than you would expect.
The full prompt list goes to you before I run anything, and you can add to it. If you already know the question you're afraid of, put it in.
What the report actually looks like
Six pages from a real, 14 page report, reproduced here with a fictional company, RouteStack, so the format is honest without exposing a real client's site. This is roughly the front half. The rest goes deeper into every remaining priority, the full content gap map, and the complete fix timeline.
Page 1 — the report opens with the short version and a breakdown of what got crawled, not a table of contents.
RouteStack has a real, uncontested positioning angle, route optimization built specifically for regional freight brokers, but almost none of it reaches a search engine or an AI answer engine right now. The six comparison pages that should be winning "RouteStack vs [Competitor]" searches share one identical meta description, and the pricing page carries no structured data at all.
Page 3 — every priority finding follows the same four-part structure.
Page 6 — content gaps get mapped against real search demand, not guessed at.
The two highest-intent gaps, "freight broker routing tools" and "reduce empty miles trucking," sit closer to the actual buyer language than "route optimization software" does, and neither has anywhere for a search engine or an AI system to send someone. Both are also phrased almost exactly like real questions buyers type into ChatGPT.
Page 8 — a second priority, this one on the GEO side rather than classic SEO.
Product, no Offer, no FAQPage. Pricing exists as plain text inside a pricing table component, readable by a person, not machine-extractable.Product and Offer schema to the pricing page with actual tier names and starting prices, plus FAQPage schema on the six questions already answered in the page's accordion.Page 9 — the AI visibility section starts with the rolled-up numbers across all buyer questions tested, then the worked example behind them.
| Engine | RouteStack mentioned | Cited instead |
|---|---|---|
| ChatGPT | Not mentioned | Samsara, Onfleet |
| Claude | Not mentioned | Samsara, Motive |
| Perplexity | Mentioned | Cited from a G2 comparison thread, not RouteStack's own site |
| Gemini | Not mentioned | Samsara, Onfleet |
| Google AI Overview | Not mentioned | Samsara, Project44 |
Page 13 — the report closes with a sequenced fix timeline, not a pitch for more work.
What you actually get, end to end
-
A written report, all seven sections, specific to your site
Not a generic template with your logo swapped in. Every finding references your actual pages, your actual crawl data, and (where relevant) how you show up in ChatGPT, Claude, Perplexity, Gemini, and Google's AI right now.
-
Your link and mention profile
Referring domains, unlinked brand mentions, and a list of specific places your competitors get cited that you don't. Places, not advice.
-
The raw prompt results
Every question I asked, every answer I got back, dated. Yours to keep and re-run later, whether or not we work together again.
-
A walkthrough call
We go through the prioritized fix list together, in the order it actually matters, with effort estimates so you know what's quick and what's structural.
-
Yours to keep, either way
Hand it to any developer or writer whether or not we ever work together again. If you move to a monthly engagement afterward, the audit fee credits fully toward month one.
How it works
One week, start to finish. If something is going to take longer, I tell you on day three rather than on day seven.
You send access
Search Console, analytics, and the site. If Search Console isn't set up I'll do it, it takes about ten minutes.
Prompt list to you
Four groups, drawn from your positioning and your search data. You add anything I missed before I run a single query.
The actual work
Prompts run against five engines, site crawled, link and mention profile pulled, findings written up.
Report lands
You read it before we speak, so the call is questions rather than a presentation.
Walkthrough call
About 45 minutes, whenever suits. You ask the awkward questions, I answer them.
What it costs
USD 1,500, fixed. No hourly, no scope creep, no "starting from."
That's a founding client rate for the first five companies, and it goes up after that. I'd rather be straightforward about why: I'm building the case studies that let me charge properly, and you're getting the same work for less while I do it.
Who this is for, and who it isn't
Worth your money if
- You sell B2B software and your buyers research before they ever talk to sales.
- You already rank reasonably on Google and can't work out why that isn't translating into AI answers.
- Somebody on your team can actually ship the fixes, or you have a developer who can.
Not worth it if
- Your site is under a couple of months old. There isn't enough data yet and I'd be guessing.
- You need somebody to do the fixes as well as find them. That's a different engagement, ask me about it instead.
- You want a number to go up next week. Most of these fixes take a few weeks to show anything.
What does an AI visibility audit actually check?
What is generative engine optimization (GEO)?
What format is the audit report delivered in?
Is the audit finding format shown here real?
Do I get to keep the report if I don't continue working together?
How is this different from a normal SEO audit?
Why not just use one of the free AI visibility tools?
How do you decide which prompts to test?
Do you cover backlinks?
What do you need from me to start?
How long until anything changes?
Do you make the changes or just find them?
What if the audit finds nothing badly wrong?
Now you've seen the format. Here's what it looks like on your actual site. One week, fixed price, yours to keep either way.