How to Read Your Skavora Report, and What to Do About Each Finding

A Skavora report answers four questions: do AI assistants recommend your brand when buyers ask for what you sell (your Visibility Score), who gets recommended instead (Share of AI Voice), how prominently and how favourably you appear (average position and sentiment), and where the engines got their answers (the source breakdown). According to the G2 2025 Buyer Behavior Report, generative AI assistants are now the number-one influence on B2B supplier shortlists, ahead of review sites, supplier websites and salespeople (source), so these numbers describe a channel that already shapes who gets shortlisted. This guide explains every section of the report and, more usefully, the specific action each finding points to.

If you have not run a scan yet, start with how to measure your AI visibility step by step. This guide assumes you have a report in front of you.

What's in a Skavora report?

A Skavora report has eight sections: the Visibility Score, the branded versus non-branded coverage split, Share of AI Voice, average position, sentiment, the prompt-by-prompt table, the source wheel, and the outreach target list. The table below is the whole report in one view, what each section measures and what to do when it reads badly.

Each section below takes one row of that table and expands it: what the number means, and what to do about it.

What does the Visibility Score mean?

The Visibility Score is a 0–100 measure of how visible your brand was across 25 realistic buyer questions, run live through an AI engine with web grounding, meaning the engine searches the live web before it answers, rather than relying only on its training data. It is deliberately weighted toward non-branded coverage (appearing when buyers ask for what you sell counts for far more than appearing when they ask about you by name), so a modest score with strong non-branded presence beats a high score built on branded questions alone. The plain-English band under the number tells you how to read it at a glance.

Two honesty notes. First, the score is a snapshot: answers are generated live and vary from run to run, so treat the score as a reading, not a ranking. Second, if a scan degrades (rate limits, a failed scoring step), Skavora shows no score at all rather than an estimated one, and the report says exactly what was and wasn't measured. A number you can't trust is worse than no number.

What to do: don't act on the score itself. Act on the sections that produce it. The score tells you how much work there is; the rest of the report tells you which work.

What is the difference between branded and non-branded coverage?

Branded coverage is how often AI mentions you in answers to questions that name your brand ("is [your brand] any good?"). Non-branded coverage is how often AI mentions you when the buyer doesn't know you exist: "what's the best accounting software for a two-person building firm that invoices on site?" Roughly 80% of the questions in every Skavora scan are non-branded, by design, because the questions that win new customers are the ones where the buyer doesn't yet know who to ask for.

Appearing only in branded questions is not visibility. Your report will say so explicitly if that's your pattern. It means AI can describe you when prompted, but doesn't volunteer you: the buyers who reach you through an assistant already knew your name, and the ones who didn't were handed someone else's.

What to do about low non-branded coverage: open the prompt-by-prompt table and list the non-branded questions where you were absent. Each one is a content brief: publish a page that answers that question directly (question as the heading, complete answer in the first sentence) and earn mentions on the third-party sources the engine cited for it (covered below). This is the core of improving AI visibility, and the tactics with published evidence behind them are set out in how to get ChatGPT to recommend your brand.

What is AI share of voice?

AI share of voice, the metric Skavora reports as Share of AI Voice, is the percentage of all brand mentions in a set of AI answers that belong to your brand. If the engine named brands forty times across your 25 questions and eight of those mentions were you, your Share of AI Voice is 20%. Skavora computes it against the brands the AI actually named in your scan, not a pre-loaded competitor list, so the leaderboard shows the market as the engine sees it, including rivals you didn't think to enter.

Mentions concentrate hard: an Ahrefs analysis found the top 50 brands capture 28.9% of all AI mentions (source). If your category has an incumbent the engine reaches for first, the leaderboard names it.

What it means when a competitor dominates: the engine keeps finding them in its sources (review roundups, comparison pages, forum threads, editorial coverage) across many different questions. Whatever produced that footprint, the report makes it mappable. Read the answers where they appear and you don't, and note which sources were cited alongside their name. Their share of voice is a reading list of the coverage you need to earn.

What does average position tell you?

Average position is where your brand sits, on average, in the list of brands each answer names (first, third, seventh). Being named first in an AI answer is the shortlist; being named seventh is the also-rans. Skavora computes it only across non-branded answers where you actually appeared, so it measures prominence, not presence, and it needs at least three such answers before a figure is shown, because an average of one is not an average.

Hold position lightly: SparkToro research found less than a 1-in-100 chance that ChatGPT gives the same list of brands in any two responses to the same query (source). Answers are probabilistic, and positions wobble run to run.

What to do: a consistently late position with decent coverage usually means the engine's sources mention you in passing, one entry on a long list, rather than prominently. The fix is depth, not breadth: aim to be the substantive recommendation on the sources AI already cites in your category, not the tenth bullet.

What does the sentiment reading mean?

Sentiment measures how the AI describes your brand in the answers where you appear: favourably, neutrally, or with caveats. A brand can have solid coverage and still lose the recommendation because every mention arrives with a "however": dated pricing, a recurring complaint, a comparison it keeps losing.

What to do about weak sentiment: read the actual answer text in the prompt-by-prompt table and identify the recurring caveat. It is usually traceable to source material. Stale pricing pages, an unanswered thread of complaints on a review site, an old comparison article. Correcting the source is what changes the description: update your own pages, respond where responses are possible, and earn fresher coverage that reflects the current product. Recency appears to matter: an AirOps study found 95% of ChatGPT citations come from content published or updated within the last 10 months (source), so newer, accurate material at least has a chance of being the version an engine reads. That is a pattern in what gets cited, not a guarantee that new coverage displaces old.

How do you read the prompt-by-prompt table?

The prompt-by-prompt table is the evidence behind every headline number: all 25 questions, the live answer each one produced, every brand named, and every source cited. The headline metrics tell you that you have a problem; this table tells you which problem, question by question.

Read it as three lists. Questions where you appeared prominently are your strengths; note which sources were cited so you can protect them. Questions where you appeared late or with caveats are prominence and sentiment work. Questions where you were absent entirely are the priority list, because each absence is a question your buyers ask being answered without you, with the brands that filled your slot and the exact sources that put them there recorded alongside.

Take one absence and follow it through. Suppose "best accounting software for a two-person building firm that invoices on site" named three competitors and cited two review roundups and a trade-forum thread. That single row hands you the entire play: a page to write (you, answering that question directly), two roundups to pitch for inclusion, and a forum where your category's buyers are already comparing notes. Repeat for each absent row and you have a concrete plan, not a vibe.

Want to see your own table before reading further? Run a free AI visibility check: 25 live buyer questions, the full scored report, no signup.

What does the "where AI gets its answers" wheel show?

The source wheel classifies every citation from your scan into types: retailers, marketplaces, review and comparison sites, editorial and press, reference sites, Reddit, YouTube, forums, blogs, industry sites, competitor sites, and your own. It shows whose material the engine drew on when it answered your buyers' questions, the closest thing AI visibility has to a map.

The shape of the wheel should shape your strategy. "Best of" list posts account for around 43.83% of ChatGPT citations for commercial queries, in an analysis by Glen Allsopp published by Ahrefs (source), so a wheel dominated by review and comparison sites is a common result, and a specific instruction: if those roundups don't include you, work on your own site is unlikely to close the gap on its own, because your site isn't what the engine cited for those answers. A wheel heavy on Reddit and forums points to buyer conversations as the material in play; one heavy on editorial points to press coverage.

What to do when review sites dominate and you're absent from them: that combination, the most common bad news in a report, means the outreach play, below. For the full taxonomy and how each source type behaves, see where AI engines get their answers.

What is the outreach target list?

The outreach target list is the source wheel made actionable: the specific sites the engine cited in your scan, ranked by influence over your category's answers. It converts "earn more coverage" from advice into an address book: these exact pages and publishers are what the engine drew on when it answered your scan.

Work it top down. For review and comparison sites: get listed, complete your profile, and pursue inclusion in the specific roundup articles that were cited. For editorial and industry sites: pitch something genuinely useful (data, a strong viewpoint, a customer story). For forums, Reddit and YouTube: be present honestly, because buyers and engines both read them. Breadth across source types tracks with visibility: an Evertune study of 75,000 brands found brands present on four or more non-affiliated forums are 2.8x more likely to appear in ChatGPT responses, and a Search Engine Land study of 800+ websites found multi-platform brand mentions correlate with AI visibility at r=0.87 (source). Both findings are correlations (widely mentioned brands are often simply the bigger ones), so treat breadth as a sensible bet, not a proven lever.

One rule keeps the list honest: earn the coverage, never fake it. Astroturfed reviews and planted forum posts are both an integrity problem and a fragile tactic: sources that vet contributions are precisely the ones engines keep citing.

Why isn't ChatGPT recommending my brand?

ChatGPT usually doesn't recommend a brand because the sources it reads when answering that category's questions don't mention the brand, or mention it too thinly to survive into a shortlist. AI assistants assemble answers from live web sources, and those sources are not the ones traditional SEO optimises for: an Ahrefs study of 15,000 prompts found only 12% of URLs cited by ChatGPT, Gemini and Copilot appear in Google's top 10 for the same query (source). You can rank well on Google and still be invisible to AI, because the engines are reading review roundups, forums and comparison pages where you may have no presence at all.

The fix is therefore not one fix but the sequence this guide describes: find the questions where you're absent, publish direct answers, and earn mentions on the sources the engines actually cite. To see the problem measured rather than guessed at, run a scan: Skavora's free check runs on Google's Gemini with live web grounding, and the £7.99 report runs the same questions across Gemini, ChatGPT, Claude and Google AI Overviews, so you can see whether your ChatGPT gap is specific to that engine or common to all four.

How do you improve AI visibility?

You improve AI visibility by measuring where you stand, closing the gaps the measurement reveals, on your own site and on the sources AI trusts, and re-scanning to check the needle moved. If you're new to what AI visibility is, start there; the steps below assume a report in hand.

Step 1: Measure by running a scan and reading your absences

Run the scan, then work the prompt-by-prompt table as described above. Output of this step: a list of non-branded questions you were absent from, the brands that filled your slot, and the sources that were cited.

• List every absent non-branded question
• Note the brands named in your place
• Note the cited sources per question
• Record your score, Share of AI Voice and average position as the baseline

Step 2: Publish direct answers to the questions you lost

For each absent question, create or upgrade a page that answers it the way an engine can lift it: the question as the heading, a complete standalone answer in the first sentence, specifics rather than generalities. This is answering real buyer questions, not stuffing keywords: the questions in your report are worded the way buyers actually speak, so write for exactly those.

• One question, one page or section, no catch-all pages
• Answer completely in the first one to two sentences
• Keep pricing, features and comparisons current
• Show a real, honest "last updated" date and keep earning it

Step 3: Earn coverage on the sources AI already cites

Take the outreach target list top down: the roundups to be included in, the review profiles to complete, the communities to participate in honestly. Prioritise the source types that dominate your wheel: that's where influence over your category's answers is concentrated.

• Pursue inclusion on the specific cited review and comparison pages
• Complete and maintain profiles on cited review platforms
• Pitch cited editorial and industry sites with something worth publishing
• Participate genuinely where buyers discuss your category

Step 4: Fix what shapes sentiment

Where the report shows caveated or unfavourable descriptions, trace each recurring caveat to its source and correct it, your own stale pages first, then responses and fresh coverage elsewhere. You cannot delete the old material, but you can make sure accurate, current material exists for an engine to find.

• Identify the recurring caveat in the answer text
• Update your own pages that feed it: pricing, features, comparisons
• Respond on third-party sources where responses are possible
• Earn fresher coverage that reflects the current product

Step 5: Re-scan and compare

After the work ships and has had time to be seen (weeks, not days), run the scan again and compare against your baseline: score, non-branded coverage, Share of AI Voice, position, and the source wheel. Expect noise between runs; look for direction across runs, not identical numbers.

• Compare every headline metric against the Step 1 baseline
• Check whether previously absent questions now name you
• Note which cited sources changed since the last scan
• Set the new numbers as the baseline for the next cycle

AI visibility is a loop, not a one-off audit

AI visibility has to be re-measured, because the ground moves: the Semrush AI Visibility Index, tracking 2,500 prompts, found 40–60% of AI-cited sources change from month to month (source). A source that dominated your category's answers in July can fade by September, and a gap you closed can reopen. The working rhythm is a loop: measure, see which sources AI trusts in your category, earn presence on those sources, re-scan.

The compressed version of this whole guide:

• Non-branded coverage is the number that matters; branded-only presence is not visibility
• Every absent question in the prompt table is a content brief plus an outreach target
• A dominant competitor's share of voice is a map of the coverage you need to earn
• The source wheel tells you where to earn it; the outreach list names the addresses
• Sentiment problems usually trace to source material: correct the material, not the report
• Re-scan after the work ships; judge direction, not single runs

Skavora doesn't control AI outputs and won't promise you a place in any answer. No honest tool can. What a scan gives you is the thing effort needs most: a target. Run your free AI visibility check and start the loop.