AI Visibility Report
How to launch a run, interpret metrics, and turn insights into actions.
AI Visibility Report
AI Visibility Report shows how often your brand appears in generative AI answers, in what context it is mentioned, and how it compares to competitors.
Core principle: AI answers are stochastic. The same question can produce different outputs, so the report uses repeated sampling and answer-level metrics instead of a single fixed rank.
When to use it
- Before a campaign, to capture baseline visibility.
- After content, PR, or product updates, to measure impact.
- For regular competitor benchmarking by market and language.
- For client-facing AI SEO / GEO reporting.
What to prepare
- Brand: target brand to analyze.
- Locale: market/language.
- Base intent: natural user-style prompt.
- Competitors: ideally 3-10 direct competitors.
- LLM sources: providers/models to include.
- Sample profile: speed vs depth mode.
Tip: write Base intent as a real user question, not as a keyword list.
How to launch a run
- Open AI Visibility.
- Fill in Launch a run.
- Choose profile:
- Fast: quick diagnostic check.
- Balanced: default trade-off.
- Deep: maximum depth, longer runtime. - Click Run report.
- Track progress in Run history, then open run detail.
Typical runtime is around 10-20 minutes depending on profile and selected sources.
How to read the run detail page
KPI block
- Brand presence: share of sampled AI answers where the target brand appeared at least once.
- Share of voice: share of sampled answers that mentioned any tracked brand or competitor and also included the target brand.
- Target mentions: raw extracted mention events. This count can exceed sample count when one answer repeats a brand.
- Confidence: internal data-quality score for collection, parsing, and extraction.
- Volatility: variance across sampled outputs.
- Sample coverage: how complete and parseable the sample set is.
Brand presence and Share of Voice are calculated by sampled answers, so percentages cannot exceed 100%. Mention counts are evidence volume, not probabilities.
Provider summary
Shows per-source contribution:
- answer-level presence,
- sample-based SoV,
- target samples,
- raw target mention volume.
Use it to identify where your brand is strongest or weakest.
Competitors table
Shows mention volume, sample-based competitive share, average confidence, and also newly discovered competitors from extraction.
Use this to prioritize real market competition in AI answers, not just your initial competitor list.
Sentiment and sources
- Sentiment: tone around target-brand mentions.
- Top cited sources: domains AI responses cite most often. Where the model returned the underlying links, you can open them as clickable URLs from a full answer.
If visibility rises but sentiment drops, this is a reputation issue, not a pure growth signal.
Source links only appear for web-search-backed models (for example AI Overview or search-grounded models). Purely generative models answer from training data and usually return no sources, so their answers show no links. This is expected behaviour, not missing data.
Verbatims and full answers
Representative snippets let you quickly validate quality and context without reading every sample. Each example shows a short 3-line preview; click Show full answer to expand the complete model response inline, together with the source links that answer cited. To review every answer at once, use the XLSX or CSV export.
How to read the metrics
AI services answer the same question differently every time. The report therefore never relies on a single answer: each provider gets a series of questions, and the metrics are computed over the whole sample of collected answers. Every percentage in the report is a share of that sample — an estimate, not an exact value.
Brand presence
The share of answers that mention your brand, out of all collected answers. Presence of 43% means the brand appeared in 43 out of 100 answers.
How to read it: above 50% — stable visibility; 20–50% — the AI knows the brand but does not always recall it; below 20% — the brand is nearly invisible to AI in this topic.
Share of Voice
Some AI answers name no brands at all. Among the answers that name at least one tracked brand (yours or a competitor), Share of Voice is the percentage of answers that include your brand. In plain terms: when the AI names market players, how often you are one of them.
How to read it: only in comparison with the competitors in the same report — their shares are computed over the same answers. If your share is below ~15%, a typical AI answer offers the user competitors without you.
Note: brand shares do not add up to 100% — a single answer can mention several brands at once.
Mentions
How many times the brand is named in total across all answers; one answer can mention a brand more than once. Presence answers "in how many answers you appear"; mentions answer "how much you are talked about".
Sentiment
Each mention of your brand is rated positive, neutral, or negative. The percentages are shares of all brand mentions (not of all answers).
How to read it: compare with the competitor average in the same report. A negative share above ~30% is worth investigating: AI retells what it reads, so look for the origin of the negativity in the Sources section and in the answer examples.
Brand role
Not just whether the brand is mentioned, but how: the AI recommends it, compares it to others, merely lists it, or warns against it.
How to read it: roles are not equal — the goal of visibility work is to move mentions from "listed" to "recommended". The "warns against" role matters even at a small share: open the actual answers and see what the AI warns about.
Where you are not found (gap analysis)
Questions to the AI are grouped by user intent: "what to choose", "what can replace a specific brand", "where to buy", and so on. For each group the report shows the share of answers where you are present.
How to read it: look for the skew. Strong groups are where the AI already knows you; weak ones are the stages of the customer journey where you do not exist for the AI. A weak group with a large number of answers is the most concrete growth point in the report.
Competitors
Brands ordered by the number of answers that mention them. The list is not limited to the competitors you specified: if the AI regularly names a brand missing from your list, it appears in the table as discovered — that is a finding too.
Sources
The sites the AI cited while answering questions on your topic. These platforms shape what the AI knows and says about your category.
How to read it: this is a map of the platforms where your content influences AI answers. If the top of the list is dominated by platforms that say nothing about you, that is a concrete direction to work on.
Data quality
Service-level reliability indicators: how many answers were collected out of the planned amount (coverage), how many of them were parsed, how confidently mentions were extracted (confidence), and how stable the result is across question series (volatility).
How to read it: if coverage and parsing are above ~90% and volatility is low, the headline numbers can be trusted. With high volatility, treat the percentages as a range: the next run may produce a noticeably different value.
How to interpret warnings
Warnings are analytical signals, not just technical errors.
Common cases:
- No/low samples: widen sources or simplify base intent.
- Target not detected: verify brand spelling, locale, and prompt framing.
- Low confidence: rerun with Balanced or Deep, then compare.
- New competitors found: include them explicitly in next run.
Recommended workflow
- Run baseline on priority locales.
- Implement content/reputation updates.
- Rerun with comparable settings.
- Compare presence, sample-based SoV, sentiment, and source mix.
- Update content plan, FAQ, comparisons, and PR priorities.
Important limitations
- Results are probabilistic, not deterministic ranks.
- One run is directional; trend over multiple runs is more reliable.
- Different AI sources can produce different market pictures.
Export and sharing
When a run is completed, export from run detail:
- PDF — formatted report for client updates and internal reviews.
- XLSX — full per-answer data: every sampled answer (complete text) plus a dedicated Sources sheet with the cited URLs.
- CSV — the same answers and sources as a single flat table.