The feed layer for AI commerce
Your feed decides whether AI can recommend you.
Assistants answer from structured product data, not your website. Verintra grades your feed A to F, fixes the gaps, and publishes the formats ChatGPT, Perplexity and Google AI Overviews read.
Full Pro for 14 days · no card · the scan needs no account at all
Score breakdown
Grade distribution across all active products
Fix these first · biggest score lift
Score breakdown · sample data
How it works
Grade. Fix. Publish.
Three steps, in this order. The order matters: nothing is rewritten before you can see what it costs you.
Every product, graded
Ranked fixes, not a list
Thirteen groups, each saying how many products it affects and how many points it recovers. The hour goes where it pays.
The whole catalog, at a glance
The distribution across every product, and where the next points are sitting.
Your source feed is never edited
Corrections are layered on at export. To undo one, you simply do not export it.
One catalog, thirty three destinations
Twenty eight arrive with a ready mapping. Ad platforms, AI assistants and comparison sites read the same corrected data.
- Meta
- ChatGPT
- Perplexity
- Criteo
- +28 more
Sample data · every number shown is a product capability, not a customer result.
Inside the product
One catalog in. Every channel, corrected, out.
Verintra reads the feed you already publish, grades every product against thirty plus checks, writes what is missing without inventing it, and pushes the corrected catalog to thirty three destinations. Your source feed is never edited.
Thirteen fix groups, each showing how many products it affects and how many points it recovers. You spend the hour where the hour pays.
Shopping copy written from the attributes already on the product, into separate fields, so you read it before anything ships.
A local pre-filter shortlists the official Google taxonomy first. A category the model never received is one it cannot return.
IF-THEN rules over thirty seven writable fields. Write the rule and it applies to every matching product on every sync, forever.
Thirteen fix groups, each showing how many products it affects and how many points it recovers. You spend the hour where the hour pays.
Shopping copy written from the attributes already on the product, into separate fields, so you read it before anything ships.
A local pre-filter shortlists the official Google taxonomy first. A category the model never received is one it cannot return.
IF-THEN rules over thirty seven writable fields. Write the rule and it applies to every matching product on every sync, forever.
Products go up over the Content API and the disapprovals come back grouped by reason, mapped to the items they affect, pointed at the fix that clears them.
A real catalog connection kept in sync in batches, so dynamic ads and Shopping campaigns are never working from two different versions of the truth.
Thirteen GA4 reports over a thirty day window, cached nightly, so feed quality sits next to revenue instead of in a separate tab nobody opens.
Two variants against each other on live traffic. The one that earns more stays, rather than the one that read better in a meeting.
Products go up over the Content API and the disapprovals come back grouped by reason, mapped to the items they affect, pointed at the fix that clears them.
A real catalog connection kept in sync in batches, so dynamic ads and Shopping campaigns are never working from two different versions of the truth.
Thirteen GA4 reports over a thirty day window, cached nightly, so feed quality sits next to revenue instead of in a separate tab nobody opens.
Two variants against each other on live traffic. The one that earns more stays, rather than the one that read better in a meeting.
When only the sizes nobody wears are left, the product is held back from export instead of spending budget on a click that cannot convert.
Supplemental layers override individual fields on individual products. Your store’s feed stays exactly as your platform wrote it.
Twenty eight arrive with a ready mapping, each with a hosted URL the channel fetches on its own schedule. Ad platforms, AI assistants, comparison sites.
Every bulk edit is revertible, and every rule can be switched off. Nothing you do here is a one-way door.
When only the sizes nobody wears are left, the product is held back from export instead of spending budget on a click that cannot convert.
Supplemental layers override individual fields on individual products. Your store’s feed stays exactly as your platform wrote it.
Twenty eight arrive with a ready mapping, each with a hosted URL the channel fetches on its own schedule. Ad platforms, AI assistants, comparison sites.
Every bulk edit is revertible, and every rule can be switched off. Nothing you do here is a one-way door.
Every capability here is live in the product today.
What it finds
Thirty plus checks, in the product’s own words.
Two scorers read every product and emit thirty eight distinct findings between them. These ten are the wording the scan actually returns, not a summary written for this page.
Swipe for all ten findings.
What you get back
Every view answers one question.
Three views from the app, on sample data. Pick one.
Which products are costing me the most?
Lowest scoring first, each with the single worst issue and whether a fix is ready to apply.
| Grade | Product | Score | Worst issue | Status |
|---|---|---|---|---|
| F | Wool blend overcoat, navy, M | 38 | No GTIN and no MPN | Action needed |
| F | Slim fit chino, sand, 32 | 41 | Title carries no size or colour | AI fix ready |
| D | Leather chelsea boot, black, 42 | 52 | Description under 120 characters | AI fix ready |
| D | Linen shirt, white, L | 55 | No product category | AI fix ready |
| C | Cashmere scarf, grey | 64 | Image below the minimum size | Action needed |
Swipe the table sideways for the issue and status columns.
Can an assistant actually recommend this catalog?
A second, separate reading. The Verintra Score asks whether the data is complete and compliant. AI-Readiness asks whether a model can tell what the product is, who it is for and when to recommend it.
Why did Google say no?
Disapprovals come back grouped by reason and mapped to the products they affect, each pointing at the fix group that clears it.
Account connected · last sync 04:12
Missing value: gtin
No identifier Google can match to a known product
84 productsIdentifiersImage too small
Below the minimum pixel size for Shopping
31 productsImagesMissing value: brand
Required for this product category
19 productsAttributesMismatched value: price
Feed price differs from the landing page
7 productsPricingGoogle’s wording, not ours.
Sample data
Catalog overview
Lowest scoring
And when you are not at your desk
The same grade, the same ranked list, the same product detail. Nothing here is desktop only, because it is the same app rendered narrow rather than a cut down version of it.
Output
One catalog, every format.
What ad platforms expect and what assistants read are different files. All of them come out of one source, generated together, on a hosted link each channel can fetch directly.
Any channel that takes a spreadsheet, Meta included.
Your own systems, and channels with a custom mapping.
Per product page markup, not a feed you register.
The Agentic Commerce Protocol, for ChatGPT.
Merchant Center, Performance Max and every channel on the Google spec.
Consultancy
Verintra was a consultancy before it was a product.
A boutique, data-driven digital marketing consultancy based in Istanbul. Fixing the feed is one part of a strategy, not the whole of it, and that side of the business is still here.
Digital advertising management
Account structure, bidding strategy, creative cycle and the product data underneath it all.
Growth strategy and CRM automation
Segment architecture, lifecycle flows and retention you can actually measure against a control group.
Performance marketing and analytics
Measurement architecture first, then dashboards. One screen per decision, not forty metrics nobody acts on.
What grade does your feed get?
More than thirty checks on every product, a letter grade for the catalog, and a ranked list of what to fix first. About two minutes.