How Astiva Measures AI Brand Visibility: Full Methodology
Last updated: May 2026
Every number Astiva reports is defined, computed, and refreshed on a published cadence. This page documents the full pipeline — query sampling, mention extraction, brand normalization, sentiment scoring, and freshness — so that every claim in the product is traceable back to a method. Nothing is inferred, rounded, or smoothed.
What is AISO and why does measurement matter?
AISO (AI Search Optimization) is the practice of monitoring, analyzing, and improving how a brand appears in AI-generated answers across platforms like ChatGPT, Claude, Gemini, and Perplexity. Unlike traditional SEO, where rankings are observable and reproducible, AI platform responses are probabilistic, personalized, and vary by query phrasing, region, and model version. Rigorous, daily measurement is the only way to build a reliable picture of AI brand presence.
The 7 AISO Metrics Astiva Tracks
Each metric has a precise definition, a reproducible compute formula, and a fixed set of reporting windows. All 7 metrics are reported per platform and aggregated across platforms at 24-hour, 7-day, and 30-day windows.
Visibility Percentage
The share of tracked prompts in which the brand is mentioned at least once.
Formula: (prompts with at least one brand mention) / (total prompts run) x 100.
Reporting windows: 24h, 7d, 30d per platform and aggregated across platforms.
Share of Voice
The proportion of mentions in a tracked competitive set that belong to the brand.
Formula: SoV = brand mentions / (brand mentions + all competitor mentions in the same prompt run) x 100.
Computed over the same prompt set that covers both the brand and all tracked competitors, ensuring apples-to-apples comparison. Reporting windows: 24h, 7d, 30d.
Average Position
Where the brand appears inside each AI response: first, middle, or later.
Each mention is tagged with its ordinal position in the response. Averages are weighted so that first-named mentions count more than later mentions. Reporting windows: 24h, 7d, 30d.
Brand Sentiment
Tone of each mention scored positive, neutral, or negative at the sentence level.
Sentiment is scored per mention, then aggregated as the weighted average across all mentions in the platform-window. Reporting windows: 24h, 7d, 30d.
First Mention Rate
The share of AI responses in which the brand is named before any tracked competitor.
Formula: First Mention Rate = (responses where brand is named first) / (eligible responses that include the brand or at least one competitor).
A direct measure of top-of-mind awareness. Reporting windows: 24h, 7d, 30d.
Mention Frequency
The total count of brand mentions across all responses in the window, de-duplicated at the response level.
Multiple mentions within a single response are counted individually; responses with zero mentions are excluded from the per-response frequency average. Reporting windows: 24h, 7d, 30d.
Sentiment Volatility
Week-over-week variance in brand sentiment across AI platforms.
Computed as the standard deviation of daily sentiment scores within the reporting window. Rising volatility with a stable average sentiment is an early signal that perception is polarising across AI platforms — often a leading indicator of a reputational shift before the average score moves. Reporting windows: 7d, 30d.
The 5-Step Measurement Pipeline
Every tracked prompt passes through the same five pipeline steps every day across every AI platform on the customer plan. No step is skipped, sampled, or cached across sessions.
Query Sampling
Astiva builds a platform-specific query set across informational, commercial, and transactional intent buckets. Queries are rotated across runs to reduce AI cache bias — the tendency for repeated identical queries to receive identical, cached responses rather than fresh model outputs. Customers can add custom prompts and bulk-upload their own query libraries. Customer-defined prompts receive the same daily cadence as Astiva-generated prompts.
Mention Extraction
Each AI response is parsed for: brand mentions (direct name, common abbreviations, and semantic equivalents); citation position (first mentioned, middle, or later); sentiment at the sentence level; citation source URLs where the platform exposes them (Perplexity, Google AI Mode); and the competitive set named alongside the brand in the same response. All extracted data is timestamped and stored.
Brand Normalization
The normalization engine resolves brand mentions across casing variants (Astiva / ASTIVA / astiva), spacing and hyphenation variations, common misspellings, and contextually equivalent references. Cross-validated against a human-reviewed ground-truth evaluation set sampled across industries, category archetypes, and naming patterns. The normalization accuracy figure on this page corresponds to the most recent evaluation run against that set.
Sentiment Scoring
Sentiment is scored at the mention level, not at the response level. A response can contain both a positive brand mention and a negative competitor mention; those are scored independently. Mention-level scores are then aggregated per platform and per time window. Sentiment Volatility tracks week-over-week standard deviation so that emerging PR issues surface before the average score shifts.
Freshness Cadence
Tracked prompts run on a daily automated schedule across every platform included in the customer plan. Dashboard data refreshes within a 24-hour rolling window so customers always see current results without manual intervention. Alerts are delivered within 24 hours when visibility, sentiment, or competitor presence shifts meaningfully beyond the configured threshold. Historical data is retained per plan: 30 days (Starter), 180 days (Growth), 365 days (Pro), up to 3 years (Enterprise).
What Does "Accuracy" Mean at Astiva?
Accuracy at Astiva refers specifically to brand normalization accuracy — the rate at which the normalization engine correctly resolves a brand mention in an AI response to the intended brand entity.
It is measured against a human-reviewed evaluation set sampled across industries and naming patterns. The evaluation set is versioned, and the current accuracy number on the product corresponds to the most recent evaluation run. When the methodology, model, or evaluation set changes, the date on this page changes with it.
Accuracy does not refer to sentiment classification, position estimation, or any aggregate metric — those have their own error bounds, documented per metric above.
Accuracy Reporting Summary
Definition: Brand normalization only — not sentiment, not position, not any aggregate.
Evaluation set: Human-reviewed, cross-industry sample across naming patterns and casing variants.
Cadence: Re-run when the normalization engine or evaluation set changes, with the date updated on this page.
Current window: See the Last Updated date at the top of this page.
How Does Astiva Handle Multi-Platform Aggregation?
Metrics are computed independently per platform first, then aggregated. Aggregation weights are proportional to the number of prompts run per platform in the window, which is consistent across all platforms in a given customer plan. This means that a platform with twice the prompt volume carries twice the weight in aggregated metrics. Platform-level breakdowns are always available alongside aggregated numbers so customers can audit the composition of any aggregate figure.
Why Rotate Queries?
AI platforms cache responses to recently-seen queries. Running the same prompt repeatedly at short intervals increases the probability of receiving a cached response rather than a fresh inference. Astiva rotates prompts within a query pool for each tracked topic, ensuring that each daily run receives a fresh model inference. Query rotation methodology and pool sizes are available on request for Enterprise customers subject to NDA.
Methodology Change Log
Any material change to a metric definition, formula, normalization model, or evaluation set is documented in the methodology change log with the effective date. Customers with historical data spanning a methodology change receive an annotation in their dashboard on the date of the change so that trend comparisons account for the definitional shift rather than attributing it to a real visibility change.
Frequently Asked Questions — Methodology
How does Astiva measure AI brand visibility?
Astiva fires real buyer-intent queries across all tracked AI platforms every day and parses each response for brand mentions, position, sentiment, and the competitive set named. Results are scored against 7 AISO metrics and reported at 24h, 7d, and 30d windows per platform.
What does accuracy mean at Astiva?
Accuracy refers to brand normalization accuracy: the rate at which the normalization engine correctly resolves a brand mention to the intended brand entity. Measured against a human-reviewed evaluation set. Does not refer to sentiment classification or position estimation, which have separate documented error bounds.
How often does Astiva update AI visibility data?
Tracked prompts run on a daily automated schedule. Dashboard data refreshes within a 24-hour rolling window. Alerts are delivered within 24 hours when visibility, sentiment, or competitor presence shifts meaningfully beyond configured thresholds.
How is Share of Voice calculated?
Share of Voice = brand mentions / (brand mentions + competitor mentions). Computed over the same prompt set that covers both brand and all tracked competitors in the same run, ensuring fair comparison on identical query surfaces. Available at 24h, 7d, and 30d windows per platform.
What is Sentiment Volatility and why does it matter?
Sentiment Volatility is the standard deviation of daily sentiment scores within a rolling window. A rising volatility number with no change in average sentiment is an early signal that perception is polarising across AI platforms — often a leading indicator of a reputational shift before the average score moves. PR and brand teams use Sentiment Volatility as their primary early-warning metric for emerging narrative issues.
Methodology is the Product
Start a free 14-day trial and track your brand across 10 AI platforms using the same pipeline documented above. Every number in your dashboard maps back to a formula on this page. Cancel before day 14 and pay nothing.