Measuring AI visibility: what actually matters.

Measurement has always been half the SEO discipline. AI search is no different, except that the measurement layer is newer and the metrics vendors have not yet converged on definitions. This post is opinionated about which metrics are load-bearing, which mislead, and what an ecommerce team should actually put on a wall. The short version: measure per-SKU, per-engine, per-query-intent, then roll up only after. Everything else is a derivative.

The metric primitive: per-SKU, per-engine, per-query-intent

Start with the atomic question: for this product, on this engine, when buyers search with this intent, how often is the product named, cited, and characterized accurately? That's the primitive. Every useful rollup — category visibility, catalog score, trend lines — derives from this granularity. Measurement tools that don't store at this level can only produce aggregate metrics, which are the ones most likely to mislead.

eCommerce Insights stores per-SKU-per-engine-per-intent by design. See the AI visibility score definition for how the primitive rolls up into a composite number.

Why share of voice can mislead

Share of voice (AI) sums citation counts for a brand across a prompt set, then expresses the brand's share relative to competitors. It's useful for comms. For ecommerce, it can move in ways that don't predict revenue. A brand's share of voice can rise while its best-selling SKUs lose citations, because a long-tail SKU grabbing mentions offsets the franchise drop. The rollup hides the product-level pattern.

This is the same reason brand-level tracking misses revenue. Any metric that aggregates above the SKU loses information ecommerce teams need.

What to track week over week

At weekly cadence, track: citation count per SKU per engine, source mix (how many mentions come from your PDP vs a review site), characterization accuracy flag (did the engine get your product's specs right), and visibility score movement. Weekly data is noisy, so expect ±10 percent week-to-week variance on many metrics even with no real change. Look at the noise floor for each metric once before reading trends.

What to track quarter over quarter

At quarterly cadence, roll up to: percentage of top-revenue SKUs with visibility score above a set threshold, source-mix trend (is your PDP share of citations rising), share of model within your competitive set, and a composite catalog visibility score. Quarterly metrics smooth the weekly noise and align with how leadership reads the story. Don't try to make quarterly metrics do weekly work — they're different instruments.

What NOT to track

Skip metrics that feel rigorous but add no decision value. Examples: total prompt-set impression counts (ambiguous denominator). Single-engine share of voice as a solo metric (loses multi-engine picture). AI-sentiment scores that aren't calibrated to product specifics. Click-through rate from AI answers until the attribution layer is more mature — most of what's reported is inferred and noisy. And anything where you can't define what "success" looks like before running the measurement.

The dashboard that actually helps a merchandiser

The dashboard that moves work is deceptively simple. One row per SKU. Columns: revenue weight (percent of catalog revenue), visibility score per engine, recent citation count, trend arrow. Filters for category, collection, variant. A single sort: revenue-weighted visibility gap, descending. The top 10 rows are your PDP triage list for the next two weeks. Everything else is read-only context.

Most AI-visibility tools eCommerce Insights has seen in 2026 bury this view under brand-level charts. The brand-level charts are fine for executive readouts; they're the wrong thing for the person actually fixing PDPs. Build or buy for the dashboard that drives action first.

Tying metrics to revenue

The honest answer on attribution: it's not mature yet. AI engines don't consistently pass referrer information when a user clicks a cited link, and some answers are read without any click at all. What works as of Q1 2026 is directional correlation — track visibility score and revenue by SKU, look for relationships in the SKUs where visibility has moved meaningfully, and report the pattern with appropriate caveats. Causal click-level attribution may land in 2027; until then, treat visibility as a leading indicator, not a direct revenue metric.

Key takeaways

• Measure per-SKU, per-engine, per-query-intent. Roll up only after.
• Share of voice is a comms metric; it can mislead when used as an ecommerce revenue metric.
• Weekly: citation count, source mix, characterization flag, visibility score movement.
• Quarterly: percent of top-revenue SKUs above threshold, source-mix trend, share of model, catalog composite.
• Skip metrics where "success" isn't defined before measurement — usually a sign the metric is ornamental.
• Attribution isn't mature; treat visibility as a leading indicator for now.

Related reading