Methodology
How the AI Visibility Score is computed.
Last updated 2026-05-29.
Score
What is AI Visibility?
AI Visibility measures how often your brand is named when AI assistants are asked the kinds of buyer-intent questions a real customer in your category would ask. It is not an SEO ranking, not a click-through prediction, and not a measure of online sentiment at large. It is one number that answers a specific question: when ChatGPT or Claude is asked the buyer-intent questions in your wizard, how often does your brand actually show up in the answer? The AI Visibility Score is a percentage between 0 and 100. A score of 35 means that across the analysed questions, your brand was mentioned in 35% of the AI responses. Lumialo computes the score on a fixed set of questions you and Lumialo agreed on at the start of the run, so the same questions can be re-asked in the future to track movement over time.
How the score is calculated
The score is a simple percentage: the share of analysed questions in which your brand was mentioned by the AI provider. There is no position weighting (rank 1 and rank 5 mentions count the same), and there is no sentiment weighting. Sentiment and position are captured in the underlying data and shown alongside each mention in the report and CSV export, but they are not folded into the headline number. We made this choice deliberately so the score stays easy to explain without methodology debate. Your AI Visibility Score is the share of analysed questions where your brand appeared, expressed as a whole-number percentage from 0–100 (rounded to the nearest integer). On the paid plan each question is asked multiple times across two AI providers (see Sample size below). A brand counts as "mentioned" for a question when at least two of those runs surface it — a deliberate stricter-than-one threshold so the headline number stays reproducible when a reader looks under the hood.
Providers
Which providers we query
Lumialo queries two large AI providers today: ChatGPT (OpenAI) and Claude (Anthropic). The free plan uses ChatGPT only — a single provider is enough to demonstrate the analysis end-to-end and produce a complete report on a smaller set of questions. The paid plan adds Claude so you get a second, independent reading of every question.
Versioning + change cadence
AI models change. The same question asked of "Claude" in March can return a different answer in June because Anthropic shipped a new model version in between. That is normal, and it is exactly why a reproducible visibility score has to record which version of each provider produced each answer. Lumialo records the exact AI model version that answered every question, and stores it alongside the answer in the underlying data. The CSV export shows which AI model version answered each question. If you compare two runs months apart and your score moved, the version log lets you tell whether the move was the AI changing, your brand presence changing, or both. Provider versions in use on the current run are visible in the report header and in the CSV.
Sample size
Queries per run
A free run analyses up to 10 questions; a paid run analyses up to 30. The questions are grouped into question groups you and Lumialo defined together in the wizard, so the set is specific to your business rather than a generic industry checklist. More questions give you a wider read on your category, which is why the paid plan opens up the larger set. Free is intentionally narrow-but-deep: a full report on a smaller question set, so you can judge the analysis end-to-end before paying. The 30-question paid ceiling exists to keep each run inside a predictable credit cost; it is not a statistical lower bound. Treat the numbers in your report as descriptive of how the AI answered those specific questions, not as a survey of the wider internet at large.
Runs per query (statistical confidence)
AI models vary — ask the same question twice and you can get two slightly different answers. To control for that, the paid plan asks each question multiple times on each provider and treats the brand as mentioned for a question when at least two of those runs surface it. The default on the paid plan is 3 runs per question per provider, which means a 30-question paid analysis can result in up to 180 AI calls (30 questions × 2 providers × 3 runs). The free plan uses 1 run per question on a single provider, which is enough to demonstrate the analysis but is more exposed to the fact that AI answers can vary day to day.
Evidence
Recommendations held policy
Each paid report aims to ship at least 5 recommendations, and every recommendation Lumialo ships has to pass six evidence checks before it appears in your report: 1. It names a specific query from your own analysis, by exact string. 2. It names a specific competitor that actually appeared in your data. 3. It names a specific gap or opportunity (not a generic "do better"). 4. It uses an actionable verb tied to that gap (not "consider", "explore", "think about", "optimise", "enhance", or "leverage"). 5. It carries an inline citation — the exact query string, a verbatim snippet, or a named competitor with position. 6. The full report contains at least 5 recommendations that pass checks 1 through 5. A "valid recommendation" on this page means one that passes all six. If, after our generator has retried, fewer than 5 valid recommendations land for your run, Lumialo holds the recommendations section, shows the "Limited recommendations this run" banner above your report, and refunds the credits that were not actually used to complete your run — see the Refunds section on the Terms page for the full policy. The rest of the report (scores, per-question competitor matrix, sentiment snippets, CSV) is delivered exactly as normal: holding recommendations is a quality check on one section of the deliverable, not a failure of the run.
How recommendations are validated
Each candidate recommendation is generated by an AI model and then passed through six checks — one per rule on this page — that read the recommendation's title, body, and evidence field and decide whether all six conditions hold. If any check rejects, the generator retries with a fresh draft and, if still failing, escalates to a stronger model. After three attempts a recommendation that still fails is dropped rather than weakened. This is why the report sometimes ships fewer than the target count and triggers the holding-recommendations path described above: Lumialo prefers five strong recommendations over ten weak ones, and prefers to refund unused credits over shipping recommendations a reader could not act on.