See how AI reads a product in your catalog

The How AI sees this product panel answers “why does this product rank where it does?” by showing exactly what Layers indexed for a product, what it ignored and why, and how variants were treated. It reads from the state the indexing pipeline already persisted, so opening it doesn’t trigger new work or extra model calls.

​

When to use this

• A product isn’t ranking where you expect and you want to see what the index actually contains for it.
• You want to confirm a metafield, tag, or collection made it into the index after a Shopify edit.
• You need to check how a product with variants was split into indexed entries.
• You see something that looks wrong in the indexed content and want to send it to the review team.

​

Open the panel

1. In the dashboard, go to Catalog and open a product. See View product details.
2. In the sidebar, find the How AI sees this product card.
3. Review the freshness pill, signal counts, and one-line strategy summary.
4. Select See full breakdown to open the slide-over.

​

What the sidebar card shows

• Freshness pill — Whether the indexed content reflects the current product data.
  • Up to date — The recorded selection matches the applied embedding.
  • Needs a refresh — The product has changed since it was last indexed.
  • Never indexed — Layers has not indexed this product yet.
• Signal counts — Tags, collections, and metafields shown as included of total. For example, 6 of 14 means six of the fourteen tags on the product made it into the index.
• Strategy summary — One line that says whether the product was indexed as a single entry or split into several entries by a variant option (for example, Indexed as 3 entries, split by Color).

If a product has never been indexed, the sidebar card still renders. The slide-over then explains there is nothing to show yet.

​

What the full breakdown shows

The slide-over has two panes.

​

What the AI reads (left pane)

This pane shows the indexed content verbatim. It is the same structured view the embedded document is built from. Fields include Title, Brand, Category, Attributes, Variant options, Variant metafields, and Description. Long descriptions are clamped with a Show full description control. When the product is split into multiple indexed entries (for example, one per color), a selector at the top of this pane lets you pick which entry to view. Entries are grouped by the distinct value of the splitting dimension, not per variant. Variants that share the split value produce the same document and count as one entry.

​

Signals and variant treatment (right pane)

The right pane has two tabs. Signals lists what was included and what was ignored for Tags, Collections, and Metafields. Included values render as green pills. Ignored values render as dashed pills and are grouped by the reason they were dropped. There are three ignore reasons. Each maps to a distinct stage of the indexing pipeline:

• Not shopper-facing content — Dropped before the selector saw it. Common examples are metafield definitions your store excludes, values that are only Shopify IDs, and empty values.
• Consistently unhelpful across your catalog — The field kept getting judged unhelpful for other products in your catalog, so Layers stopped considering it.
• Judged not descriptive for shoppers — The value reached the descriptiveness model, or an overriding verdict said it wasn’t useful to shoppers.

Counts read included of total. The total includes the values dropped before the selector saw them, so it matches what you’d see for the product in Shopify. Variant treatment shows how variants were split into indexed entries. Each entry is labeled by its splitting-dimension value and links to the corresponding entry in the left pane.

​

Flag a product for review

If something in the indexed content looks wrong and you want the review team to look at it, use the Flag for review action in the panel header.

1. Open the slide-over via See full breakdown.
2. Select Flag for review in the header.
3. Pick a reason:
   • Text that should not be there
   • Something important was left out
   • Variants are handled wrong
   • Something else
4. Add an optional note (up to 2,000 characters).
5. Select Send.

Once sent, the panel header shows the flag was submitted. Select Undo to withdraw the flag. Flags are unique per product and user. Sending a second flag replaces the first.

​

Freshness and the sidebar counts

The freshness state and signal counts are derived from the recorded indexing state, not by re-running any selection. This means:

• Opening the panel is instant and doesn’t cost model or API calls.
• If you edit a product in Shopify, the freshness pill reads Needs a refresh until the product is re-indexed. Trigger a re-index with Manually resync a single product, or with Push to Index from the bulk actions in the catalog list.

​

Tips

• If a value you expect to see is missing, check the Signals tab first. It will either be in the ignored list with a reason, or missing from the total. Which one it is tells you whether the value exists on the product at all.
• Field labels in the left pane are shown exactly as the index labels them, including underscores. That is on purpose. It is the label the embedding sees, not a prettified name.
• For a split-indexed product, use the entry selector in the left pane to compare what a shopper searching for one variant value sees versus another.

​

Next steps

• View product details
• Manually resync a single product
• Resync your full catalog