How many of your products carry a description long enough for an AI shopper to reason about.
A product passes at ≥200 plain-text characters and ≥30 words; the signal reads green when ≥80% of readable products pass.
Sourcesread_products
METHOD / FIDELITY / CONSENT
A grade is only useful when its limits are visible.
Catalume reports what synthetic AI buyers observed, how each stage was tested, and when the observation expires. This page is the reference behind every public grade, estimate, and badge.
What is measured
The Agent Readiness Grade combines four pillars: Discoverability, Comprehension, Choosability, and Transactability. A mission can be Won, Placed, Invisible, Misunderstood, or Blocked, with confidence and evidence attached to the outcome.
How the legs work
Catalume tests a shared intent across ChatGPT, Gemini, Claude, Perplexity, and Copilot contexts. Retrieval, reasoning/choice, and checkout walk each carry a separate fidelity label. A checkout walk never completes payment.
What public pages show
- Public scans use a storefront URL and public evidence only.
- Public results are watermarked and coarse.
- Competitor-specific loss detail, full diagnosis, and unwatermarked exports require install or verified control.
- Named grades and leaderboard placements require merchant opt-in and can be revoked.
What a grade does not mean
A grade is not a certification, audit pass, product-quality score, revenue guarantee, or exact ranking inside a consumer AI app. It is a time-bounded observation of an agent-readiness test set.
Revenue-at-stake estimates
Where a merchant connects attribution data, Catalume may show an estimate range derived from consented AI-referral sessions, conversion lift, average order value, and observed Share of Recommendation movement. The formula is AI-referral sessions × max(0, conversion lift) × average order value × uncertainty bounds. Missing inputs reduce precision; simulated missions never enter the estimate; every range carries its source, confidence, and the disclaimer that referral-linked estimates do not establish causation. One exclusion is worth naming: a loss where the winning position appears to be a paid placement is left out of the exposure basis entirely, because no change to a store can outrank a purchased position — counting it would describe money no fix can reach. That detection is best-effort and is always shown with its confidence; the count of excluded losses is shown beside the range rather than folded into it. A range always renders through one shared formatter — for example ≈ $3,800–$5,700 estimated — so no surface can present exposure as a single confident number.
Prediction calibration
Every revenue-at-stake range is scored after the fact: once a fix's outcome is reconciled against the estimate that preceded it, the prediction either landed inside its range or it did not. The calibration hit rate below is that running record — the share of estimates whose realized outcome fell within the predicted range, shown as a percentage interval, never as a dollar figure or a promise of future revenue.
CALIBRATION RECORD · STILL CALIBRATING
Still calibrating — 0 of 20 predictions reconciled
Published only after at least 20 post-fix predictions are reconciled; below that floor the record stays "still calibrating" rather than reporting a thin, over-precise number. The hit rate measures estimate accuracy only — it is not a guarantee, and it does not attribute revenue to any single fix.
OPTIONAL VISITOR CALCULATOR · NO DATA IS SENT
See the range mechanics with your own inputs.
Use monthly revenue and the observed miss rate only as a planning illustration. Catalume does not infer a dollar figure from a public scan.
No dollar figure is shown until you provide inputs.
LAYER 1 · THE FIVE SIGNALS SHOPIFY SCORES
How we read the same five listing signals.
Approximation · our reading of the same five signals — not your Shopify score. Catalume recomputes each one from the merchant's own product and content data. It never reads Shopify's analytics, which would require protected customer data access — so these are our numbers, computed our way, published here in full.
How many of your products carry enough media to be compared visually.
A product passes at ≥3 attached media items; the signal reads green when ≥80% of readable products pass. Media that is not an image counts toward the total, because per-item media types are not readable with the access Catalume requests.
Sourcesread_products
How many of your products carry at least one published review an agent can cite.
A product passes with ≥1 published review; the signal reads green when ≥50% of readable products pass.
Sourcesread_products
Whether every purchasable variant states the option values that distinguish it.
A product passes when it has at least one variant and no variant is missing an option value (a single-variant product passes by definition); the signal reads green when ≥90% of readable products pass.
Sourcesread_products
Whether your privacy, refund, shipping, and terms policies are published and substantive.
All four policies must be present with ≥200 characters of body text; anything less reads red.
Sourcesread_content