The AI visibility score is a number from 0 to 100. It adds up the points of a fixed set of checks, grouped into two pillars:
The catalog is the same one AI agents read through Shopify's agentic commerce tools, which is why the second half matters as much as the first: a perfect listing the catalog cannot see is still invisible to an assistant.
The points below are what each check is worth right now; they always add up to 100. A product's breakdown shows, for every check, the points earned, the evidence, and the remediation when points are missing.
| Check | Points | How it is measured | How to earn the points |
|---|---|---|---|
| Catalog eligibility | 10 | Whether Shopify's global AI catalog returns the product at all when Octolift searches for it in your store. | Keep the product active and the online store public, then give Shopify time to sync. Assistants cannot recommend what the catalog does not return. |
| Core fields present | 6 | Whether the catalog's copy of the product carries a title, a description, a price, media and variants. Points are pro-rated per field. | Fill in whatever is missing on the product in Shopify and let the catalog pick it up. |
| Variant data structure | 6 | Whether every variant the catalog returns has a price. All priced earns full points, some priced earns half, none earns nothing. | Give every variant a price in Shopify, including the ones you rarely sell. |
| Policy clarity | 8 | How many of the four policy pages Shopify exposes to agents are published: refund, shipping, privacy and terms of service. Points are proportional. | Publish the missing pages under Settings, then Policies. The breakdown links straight there. |
| Availability hygiene | 6 | Whether what the catalog says about availability agrees with your admin. Any mismatch earns nothing. | Usually a sync delay after a stock or status change; re-scan after a while. Persistent mismatches point at a variant that is unavailable in one place and not the other. |
| Media quality | 8 | At least three images, all with alt text, earns full points. Fewer images or missing alt text earns half. No images earns nothing. | Add images and write alt text that says what is in each one. The breakdown links to the product's media. |
| Review signal | 6 | The review count the catalog reports: five or more earns full points, at least one earns half, none earns nothing. | Collect reviews, and make sure your review app passes them to Shopify so the catalog can see them. |
| Check | Points | How it is measured | How to earn the points |
|---|---|---|---|
| Title | 10 | Judged by the AI rater: does the title name the product type and one distinguishing attribute, without all caps or keyword stuffing. | Something like Merino wool crew sweater, navy rather than SWEATER or a string of keywords. Improve with AI suggests up to three. |
| Description | 15 | Judged by the AI rater: is it at least 150 words, does it say who the product is for and how it is used, and does it carry concrete facts an assistant can quote. | Write for the shopper who is comparing: audience, use cases, materials, dimensions, what is in the box. Improve with AI drafts this for you from your own facts. |
| Variant attribute completeness | 5 | Whether variant option values are specific enough to match a shopper's request. All values specific earns full points, at least half of them earns half, fewer earns nothing. | Rename values like Option 1, Style A or Default Title to what they actually are: Navy, 42, Left-handed. |
| Category fields | 14 | The share of the category's attributes (the fields Shopify adds for that category) that have a value. A broad, non-leaf category earns at most half, because most attributes only exist on the specific subcategory. | Pick a leaf category, then fill the attributes: colour, material, size, pattern and whatever the category defines. Improve with AI picks values your text supports. |
| Category match | 6 | Full points unless the category audit finds a clearly better category in Shopify's taxonomy for what the text describes. A product with no category earns nothing here. | Assign the most specific category that fits. When the audit finds a better one, Improve with AI offers it with one click. |
| Rating | Score | What it means |
|---|---|---|
| Strong | 80 and above | Well optimized |
| Good | 60 to 79 | Solid visibility |
| Weak | 50 to 59 | Missing key signals |
| Critical | 49 and below | Needs attention |
The Products page shows the rating next to each score and lets you filter by it. The thresholds are the same for every store.
The two text checks are judged by one language model, the same for every store, whichever suggestion model you pick on the Plan and model page. A score is only comparable if one rater produced it: measured on the same listings, models disagree with each other by a wide margin, so a score that changed with your model choice would rewrite your own history and would not mean the same thing as anyone else's. The rater is also given the exact word count and the rubric, so the judgement is about the content rather than about counting.
A score is only produced when the rater has judged both text checks; if it does not answer, you see a message and can try again rather than a half verdict. When the catalog could not be reached, the score is computed and stored without the agent readiness half and marked incomplete, so you can tell the two apart.
The last score of every product is kept, so the Products page can show the score before and after a change you applied.
Not discoverable. The catalog did not return the product when Octolift searched for it in your store. Every agent readiness check then earns nothing, and the breakdown says so. Common reasons: the storefront is password protected, the product is new or was just changed and has not synced, the product is not active, or it is not available in the checked market. Confirm the storefront is public and the product is active, give Shopify time to sync, then re-scan.
Incomplete scan. The catalog could not be reached at all (a timeout or an error on the catalog's side). The agent readiness checks are then left out of the score rather than counted as zero, the page shows a warning, and a Try again button re-runs the scan. This is temporary.
Open Support in the app menu to chat with us, or email support@craftshift.com. Tell us your store URL and the product you were looking at and we will pick it up from there.
What the suggestions cover, the rules they follow, how applying a change works, and what is written to your store.
How AI assistants see your product, how shopper questions are ranked per country, and how tracked questions work.
What a plan includes, what counts against the monthly allowance, when it renews, and how the suggestion model changes it.