By Satish K, Co-Founder and CEO, Astiva AI · Updated April 2026
Define the competitor set. Pick 5-10 competitors your buyers already compare you against. Include the 2-3 category leaders even if you believe you do not compete with them. AI often names them regardless.
Build the prompt library. Target three intent buckets: informational ("what is X"), commercial ("best X for Y"), and transactional ("X vs Y", "X pricing"). Aim for 30-100 prompts. Rotate daily to reduce cache bias.
Compute Visibility % and Share of Voice. Visibility % = prompts where the brand was named / total prompts run, per platform. Share of Voice = brand mentions / (brand mentions + total competitor mentions).
Compute Average Position and First Mention Rate. Each mention has an ordinal position. Average Position is the mean. First Mention Rate = responses where the brand is named first / eligible responses.
Score Sentiment and Sentiment Volatility. Classify every mention as positive / neutral / negative. Sentiment Volatility = standard deviation of daily sentiment over a rolling 14-day window. It catches narrative shifts before the average moves.
Freeze the report window and version the methodology. Name the window. Document the prompt library version, competitor set, and normalization rules. Each follow-up audit compares to this baseline.
Why this matters
Dashboards without published formulas do not survive procurement review. Every AISO claim you make to an executive or a client will be challenged; a defensible audit answers every challenge by pointing at the formula, the reporting window, and the raw data behind it.
Common questions
How many prompts do I need for a statistically meaningful audit?
30 prompts per intent bucket run daily for 30 days gives approximately 900 response observations per bucket, enough to surface stable Visibility % trends. Below 100 total prompts, weekly noise can drown the signal. Astiva's Growth plan supports 100 tracked prompts and rotates them daily to reduce cache bias across all 10 monitored AI platforms.
Do I need to track every AI surface?
Match the platform list to where your buyers actually are. For most B2B categories, that is ChatGPT + Perplexity + Claude + Gemini at minimum. Add Google AI Overview if you care about Google search intent collapsing into AI answers. Add Grok and Meta AI if your buyers are active on X or Facebook. You do not need to track every surface — you need to track every surface your buyers use.
How does Astiva define brand normalization?
Astiva's normalization engine resolves brand mentions across casing, spacing, hyphenation, common misspellings, and brand aliases, cross-validated against a human-labeled ground truth evaluation set sampled across industries and naming patterns. The accuracy rate reflects the most recent evaluation run and is published on the methodology page. See the methodology page for the full normalization specification and error bounds.