When you run AgentReady's audit on your catalog inside the app, you get a number out of 100 and a letter grade. It is satisfying to see, but a score is only useful if you know what moved it and what to do next. So here is exactly what the in-app readiness score measures, how it is weighted today, and how to push it up.
Two scores, two jobs
First, a distinction that trips people up. AgentReady has two different scores:
- The in-app catalog score, the subject of this post. It reads your products through Shopify after you install and scores the structured data AgentReady generates for them, on four parts.
- The free public checker, at /tools/shopify-ai-readiness-checker. It needs no install, reads your public storefront the way a crawler does, and scores five categories (brand identity, products, discovery files, answer-engine basics, trust and policies) on an A to F scale. Its weighting and grade bands are documented in the readiness score reference.
They use different inputs and different bands, so a checker grade and an in-app grade for the same store will not match, and a number quoted from one does not apply to the other.
Why a single catalog score exists at all
AI shopping assistants and agents increasingly read stores through structured data rather than through page layout. "Readable" is not one thing, it is several, and they matter in different amounts. The score rolls those into one number so you have a single thing to watch, and it breaks into four parts so you know where the points went.
What each part measures
The weights below are as of November 2026. We adjust the scoring as we learn, so treat the app's own breakdown as the current source.
Coverage (30 points). The share of your products that have valid generated structured data. This one is blunt on purpose: if only half your catalog has clean structured data, only half your store is described in a form machines can parse. When coverage is low, it is usually because some products have not been processed yet, and it is often the biggest early win.
Structure quality (35 points). Whether each product's data has its required attributes, and whether the values are normalized. Today the required attributes are category, color family and material; each missing one costs points. Values that look like raw data, such as ALL-CAPS tags or strings full of special characters, count as unnormalized and also cost points. For the underlying Product schema fields, our product schema guide walks through what "complete" looks like.
Variant resolvability (20 points). Whether an agent can tell your variants apart. Two things cost points here. First, option values that say nothing, such as "Default", "None" or a blank value on a product that really has variants (a simple product with Shopify's single "Default Title" variant is not penalized). Second, a product that varies on two or more options, say Size and Color, without a declared order for which option comes first.
Trust and safety (15 points). Two concrete checks. Each attribute in the generated data should carry a confidence level, which records how sure the system is of that value; each attribute without one costs a little. And no claim may be marked both safe and unsafe at once; a conflict like that costs more. This part is about whether the data is honest about its own certainty.
Add them up and you get your score out of 100.
What the letter grade means
The grade is a quick read on the number:
- A is 90 and above.
- B is 75 to 89.
- C is 55 to 74. In the app, a product at C or better counts as "AI-ready".
- D is under 55.
These bands are AgentReady's definitions for sorting work, not thresholds that ChatGPT, Perplexity or Shopify publish. No assistant tells anyone what score it needs, so read the grade as "how complete and unambiguous is this data", not as a prediction of what an agent will do. A sensible first goal is to get your catalog to C or better, then work toward A.
How to actually read your score
Do not stare at the total. Look at which of the four bars is lowest relative to its maximum, because that is where the cheapest points are.
- Low coverage usually means products that have not been processed yet.
- Low structure quality means missing or messy attributes. The prioritized fix list names the specific products and fields, and for images specifically, alt text is a common gap.
- Low variant resolvability means placeholder option values or multi-option products without an option order. Standardize how options are written across the catalog.
- Low trust and safety usually means attributes without a confidence level on a subset of products.
Fix the lowest bar first, re-run the audit, and watch it move. It is a faster loop than trying to improve everything at once.
The honest state of this
A readiness score is a proxy. It approximates how complete and unambiguous your catalog data is, and "how machines read stores" is still changing. A perfect 100 is not a guarantee of sales from AI, and a B is not a crisis. The score is most useful as a direction and a diff: it tells you where the gaps are, and it tells you whether the work you just did changed the data.
It also only covers the reading layer, the structured data agents consume. The newer question of whether an agent can operate your store is a separate measurement, which we get into in Lighthouse's agentic browsing score.
That is the point of putting a number on it. Catalog data quality used to be invisible work with no feedback. A score, broken into the parts that actually move it, turns it into something you can see and act on. That is why AgentReady shows it first, and why the first thing worth doing after install is finding your lowest bar. The audit, findings and fix previews are on the Free Audit; publishing the fixes starts on Core. If you have not installed anything yet, start with the free AI-readiness checker, which takes under a minute.

Comments
Every comment here comes from a verified email. Write yours, confirm from your inbox, and it's live.
Loading comments…