The product
Product feed management, down to the field.
Everything happens at field level. Verintra scores the fields you already publish, writes replacements into separate fields, and layers them on at export. Your source feed is never edited.
No login · read only · first score in about two minutes
Slim fit chino, sand, 32
Grade F, score 41. Five findings, four of them one click away.
| Field | In your feed | On export |
|---|---|---|
| title | Chino 32 | Slim Fit Chino Trousers, Sand, W32 |
| description | 41 characters | Cotton twill with a touch of stretch, tapered through the leg. |
| gtin | empty | taken from your catalog, never invented |
| google_product_category | empty | Apparel > Clothing > Pants |
| color, size | empty | Sand · 32 |
Sample data · source feed unchanged
Scoring
Two scores, measured once.
Two engines read every product. One asks whether the data is complete, the other whether a model can understand it. A feed can be valid and still say nothing.
Complete and compliant
Required fields present, identifiers valid, images usable, prices consistent, categories mapped to the official taxonomy. This is the score that decides whether Google and Meta accept the product at all.
Understandable to a model
Whether the title states what the thing is, the description says who it is for, the attributes a shopper would ask about are there, and the product page carries markup an assistant can read.
Same facts, two weightings
Where the two scores disagree is the useful part.
Each dimension is measured once, by one scorer. The two scores are then different weightings of those same measurements. So when a catalog grades B for Google and D for assistants, the gap is not a mystery: it is exactly the dimensions the second column weights more heavily.
| Dimension | Verintra Score | AI-Readiness |
|---|---|---|
| Required fields: price, availability, link | High | Not weighted |
| Identifiers: GTIN, or brand and MPN | Medium | Medium |
| Title | Medium | Low |
| Description and semantic richness | Medium | High |
| Category specificity | Medium | Medium |
| Images and visual richness | Low | Medium |
| Attribute completeness | Medium | Medium |
| Structured data on the product page | Not weighted | Medium |
The eight dimensions that always apply. Two more are conditional: variant and apparel consistency, which only applies to catalogs that have variants, and review signals, which stay unscored until there is review data to read. A dimension with nothing to judge is weighted out rather than scored zero, so no catalog is punished for a field its category does not use.
Fixing
Thirteen fix groups, ranked by what they recover.
Findings collapse into groups, not thousands of rows. Each says how many products it touches and how many points it lifts, so the order of work is decided for you.
Missing identifiers
GTIN, MPN or brand absent, so the product cannot be matched
214 products+9 ptsShort or vague titles
No size, colour or material, so no query matches it
168 products+6 ptsNo product category
Unmapped to the Google taxonomy, so bidding and ranking both suffer
97 products+3 ptsUnusable images
Below the minimum size, or the URL does not resolve
31 products+2 ptsBroken sale dates and prices
A sale window that has closed, or a price the landing page contradicts
12 products+1 ptFive of the thirteen shown. Every group opens the right tool with the affected products already filtered, and every bulk change can be reverted.
AI content
Grounded in your catalog, or it does not ship.
A model that invents a material is worse than an empty field: the empty field fails a check, the invention passes one. So generation is constrained, not trusted.
Written from the fields you have
Shopping-shaped copy in your brand voice, inside the length limits, using only attributes already present on the product. Written to separate fields, so you compare before anything is exported.
Hallucination made structurally impossible
A local pre-filter narrows the 5,500 official Google categories to a shortlist, the model may only pick from that shortlist, and the answer is validated against it. A category that was never offered cannot come back.
Extracted, not guessed
Colour, size, material, gender and age group pulled out of the text you already publish. If the source does not say it, the field stays empty and the check stays failed. That is the correct outcome.
Grounding
The model never sees a wrong answer.
Category matching is where feed tools hallucinate: they hand a model five and a half thousand options and hope. We narrow the list first, so the wrong answer is not on it.
The shortlist is built locally, from the words already on the product, with no model involved. The model only ranks what survives, and the answer is then checked back against the shortlist. A category that was never offered cannot come back.
See it on your own catalog →If nothing on the product supports a category, the field stays empty and the check stays failed. That is the correct outcome, not a gap to fill.
Automation
Write the rule once. It runs on every sync.
Most feed problems are not one product, they are a pattern. Rules turn a pattern into a permanent fix that applies to everything matching it, including products you have not added yet.
Add the size to every trouser title
Runs after each feed sync, before export.
168 products match. Preview the result before you save.
Sample data
Publishing
Thirty three channels out of one catalog.
Twenty eight arrive with a ready mapping, so the feed comes out in the shape the channel expects. Each gets a hosted URL the channel can fetch on its own schedule.
- Ad platforms
- AI shopping surfaces
- Price comparison
- Affiliate
Nineteen of the thirty three. Any channel that reads a feed URL is reachable through the generic export.
Google Merchant Center
Push products straight to your account over the Content API, and read the disapprovals back grouped by reason, mapped to the products they affect and pointed at the fix group that clears each one.
Meta catalog
Connect a catalog and Verintra keeps it in sync in batches, so Advantage+ and dynamic ads read the same improved product data your Shopping campaigns do.
GA4
Thirteen reports over a thirty day window, cached nightly, so feed quality sits next to what the products actually earned instead of in a separate tab.
A/B title tests
Run two title variants against each other on real traffic and keep the one that earns more, rather than the one that reads better in a spreadsheet.
Broken size runs suppressed
When only the sizes nobody wears are left in stock, the product is held back from export instead of spending budget on a click that cannot convert.
Supplemental feeds
Override individual fields on individual products, joined by id and merged into the parent feed on export, with their own registrable URL if a channel needs one.
Getting started
It starts with a URL you already have.
Nothing to install and no plugin to keep updated. If your store publishes a product feed, Verintra reads it from that URL. Sixteen platform guides tell you where to find yours.
Guides forShopify·WooCommerce·ikas·Ticimax·İdeasoft·Akinon·Magento and Adobe Commerce·PrestaShop·Shopware·BigCommerce·OpenCart·Ecwid·T-Soft·Inveon·Google Sheets·Any direct feed URL
These are guides to where each platform hides your feed URL, not integrations to install. Verintra reads any feed URL, whatever produced it.
What happens after you paste it
Step 5 is the only step that changes what a channel sees, and it never touches the feed your store publishes.
See it on your own catalog first.
Reading a feature list tells you less than one grade on your own products. The scan is read only, so there is nothing to undo afterwards.