How Skavora measures AI visibility

Skavora checks AI visibility by running 25 research-derived buyer questions live through AI assistants (Gemini on the free check, plus ChatGPT, Claude, Google AI Overviews and Google AI Mode on the £9.99 report) and scoring how often your brand is named, where it appears in the answer, and which brands are recommended instead. Every question is answered with grounding switched on, meaning the engine searches the live web as it answers rather than replying from training data alone. According to the G2 2025 Buyer Behavior Report, "generative AI chatbots" are now the number-one influence on B2B supplier shortlists, ahead of review sites, supplier websites and salespeople (source). If AI assistants are building shortlists, your brand is either part of the answer or it isn't, and most brands have no way of knowing which.

This page sets out the full method: how the questions are generated, how the scan runs, what the report contains, and the rules we follow to keep the measurement honest. There is nothing hidden in it. Anyone could audit this process, and that is the point.

How does a Skavora scan work?

A Skavora scan runs in three steps: tell us about your brand, let the scan research your market and run 25 live buyer questions, then read your scored report. No signup and no card details for the free check. If you want the background on what AI visibility is and why it now sits alongside SEO, the explainer covers it in full. The steps below cover how Skavora measures it.

Step 1: Tell us who you are

You enter four things (your brand name, your website, your category and your country) plus, optionally, up to three competitors. That is all Skavora ever asks for. There is no account creation, no connection to your analytics, and no access to anything you own. The scan sees exactly the same answers a buyer would get asking the same questions, which is precisely what makes the measurement meaningful.

Step 2: Skavora researches your market and writes 25 buyer questions

Before a single question is asked, Skavora researches your brand live: what you sell, who buys it, and how those buyers actually decide. From that research it generates 25 realistic buyer questions. We call them research-derived buyer questions, meaning conversational, multi-constraint questions built from live research into how your buyers decide, never keyword strings and never templates stamped out from the category name alone.

A real buyer doesn't ask an assistant "best CRM software UK". They ask something like "best CRM for a 10-person agency that bills hourly and already uses Xero". Skavora's questions are written to match how people genuinely talk to AI assistants, because those are the questions where a buyer who has never heard of you decides who to consider.

Around 80% of the questions never mention your brand at all. This is deliberate. Non-branded coverage, how often you are named in answers to questions that never mention you, is the real visibility signal, because the questions that win new customers are the ones where the buyer doesn't yet know who to ask for. The remaining ~20% name your brand, which tests what AI says when someone has already heard of you. The split is enforced in code, not left to chance, and there's a full explanation of what AI visibility is, and why non-branded questions decide it if you want the reasoning in depth.

Step 3: The questions run live, and your report is scored

Every question runs live through AI with web search grounding switched on: live grounded answers, meaning real responses generated at scan time with access to the current web, never cached results. The distinction matters: an answer produced from a model's training data alone is a snapshot of what that model absorbed before its cut-off, while a grounded answer reflects what the engine can find on the web at the moment it replies. From each answer, Skavora captures every brand named and every source cited, then scores the lot: your 0–100 Visibility Score, your Share of AI Voice against the brands AI actually named, your average position when you appear, sentiment, and the full breakdown of which sites the answers drew on.

Live measurement matters because AI answers and traditional search results have quietly diverged. In a 15,000-prompt study, Ahrefs found that only 12% of URLs cited by ChatGPT, Gemini and Copilot appear in Google's top 10 for the same query (source). Google rankings are a weak proxy for AI visibility: the two overlap far less than most teams assume, so the only way to know where you stand in AI answers is to ask the engines directly and record what comes back.

What is an AI visibility score?

An AI visibility score is a 0–100 measure of how often, and how prominently, AI assistants name your brand across a set of realistic buyer questions. Skavora's score is weighted toward non-branded coverage: being named when the buyer already typed your name counts for much less than being named when they simply described their problem. A brand that only ever appears in branded answers has recognition, not visibility, and the report says so explicitly rather than letting a flattering number stand.

Alongside the score, the report computes Share of AI Voice: your share of all brand mentions across the scan, measured against the brands the AI actually named, not a list you supplied. If the assistants named nine competitors and you twice, that ratio is your real competitive position in AI answers, and it is calculated by exact brand and domain matching, one counting rule for the whole report, so the headline number and the competitor leaderboard always agree.

What's in your report?

The free report is the complete scored picture from one engine, Google's Gemini with live web grounding, across all 25 questions. It contains:

• Your 0–100 Visibility Score, weighted toward non-branded coverage, so the headline number reflects the questions that actually win new customers, not just recognition of your name.
• The branded / non-branded coverage split, so you can see at a glance whether AI knows your brand, or merely acknowledges it when prompted.
• A competitor leaderboard with Share of AI Voice, so you know exactly which brands the engines recommend in your place, including ones you may not think of as competitors.
• Sentiment and average position when you're named, so you know not just whether you appear, but how favourably and how prominently.
• Prompt-by-prompt answers, so you can read exactly what the AI said, word for word, for every one of the 25 questions.
• The cited-source breakdown, ranked by influence, so you know which review sites, retailers, forums and publications the answers actually drew on, and therefore exactly where to earn coverage next. Every citation is classified into a source taxonomy (retailers, review and comparison sites, editorial, Reddit, YouTube, competitor sites, your own site and more) and there's a full piece on where AI assistants get their answers if you want to see how that ecosystem works.

The free check answers: does AI recommend you today, on one engine? The complete AI Visibility Audit, £9.99, answers: how does that hold up across Gemini, ChatGPT, Claude, Google AI Overviews and Google AI Mode, with a cross-engine benchmark, a prompt-by-engine heatmap, per-engine source ecosystems with measured overlap, follow-up Q&A on your own report, and custom question sets, where you can edit the generated questions or write your own (up to 25) and re-run the scan. Once you have a report in front of you, the user guide walks through every section of the report and what to do with each finding.

How do we keep the measurement honest?

Skavora's measurement is honest by construction: the prompts are neutral, the answering engine is never told which brand is being measured, and a degraded scan never shows a score. These rules are product invariants, enforced in code, not policy documents, and each one exists because the alternative would quietly bias the number you see.

The questions are neutral by construction. No question ever names a platform, and no question ever names any brand other than your own (and only in the ~20% branded set). A prompt that mentions a competitor or steers the assistant toward a particular site would contaminate the answer before it was written.

The answering engine is never told which brand is being measured. The AI answers the buyer question cold, exactly as it would for a real user. It cannot flatter you, because it doesn't know you're asking.

The ~80/20 non-branded split is enforced in code. It cannot drift toward the branded questions that produce more comfortable numbers.

A degraded scan shows no score. If rate limits or a scoring failure mean part of the scan didn't complete, the report states exactly what was and wasn't measured, and the score is withheld rather than estimated. Skavora never pads or fabricates a number to fill a gap.

A scan is a snapshot, not a permanent verdict. AI answers genuinely vary: SparkToro's research found less than a 1-in-100 chance that ChatGPT gives the same list of brands in two responses to the same query (source), and the Semrush AI Visibility Index found 40–60% of AI-cited sources change from month to month (source). Skavora reports what the engines said at scan time, and re-scanning after you've earned new coverage is how you measure movement, not a workaround for a flaw.

We don't promise outcomes. Skavora measures your visibility and shows you where the answers come from, so you can direct your effort at the sources that matter. It does not, and honestly cannot, control what any AI engine says. Any tool that guarantees you a place in AI answers is promising something no one controls.