Product Feed Health Checker

Independent preflight checker — not an official Google or OpenAI approval service

About

About this checker

An independent preflight checker for product data, built on versioned rules with an official source and a verification date attached to every one of them.

64 versioned rules across 3 rulesets · ruleset 1.0.0 · rule data verified 11 November 2026 · no account needed for a local run

Rules

64 versioned rules across three rulesets

Ruleset version

1.0.0 (schema 1.0.0)

Verification date

11 November 2026

Oldest rule verification

11 November 2026

What Product Feed Health Checker is

An independent preflight checker for product data. You import a product export, choose a destination ruleset, and get row-level findings that name the rule, the field, the raw value, the evidence and the official source. Then you repair deterministically, export, and revalidate.

It exists because the useful information in a feed failure is almost never the headline message. “Invalid value” for an image does not tell you whether the file is 300×300, 18 MB, an animated GIF, or an HTML error page served with a .jpg name. A checker that says which one, and links the source requirement, saves the round trip.

What it is not

  • Not an official Google or OpenAI product, and not affiliated with either.
  • Not an approval, eligibility or account-health service. There is no approval signal in the output.
  • Not a feed management or publishing tool. It does not upload, schedule or submit a feed to a platform.
  • Not a Business Hub integration. Business Hub readiness is a manual, informational checklist.
  • Not a crawler-policy enforcer. It can read robots.txt evidence, but a robots allow rule is not an access guarantee.
  • Not a page renderer. JavaScript-only pages become Needs review, never a guess.

How rules are versioned and verified

  • Rules live in versioned data, not in code. Each rule carries a rule ID, a rule version, a requirement type, a severity, an expected outcome, an official source URL, a source type and a last-verified date.
  • Each rule declares its applicability: always, only when a field has a certain value, only before or after a date, only when remote evidence exists, only when a destination is selected. An applicability that cannot be established produces Needs review, never a guessed error.
  • Every rule evaluation records an explicit as-of date. Dates are never inferred from the clock, because an image that passes today can fail after 2027-01-31.
  • Where official sources conflict, the registry records both and states which one the checker follows and why. The Google 500×500 warning-date conflict (2026-04-14 versus July 2026) is the worked example.
  • A source change produces a new ruleset version and a changelog entry, so a report produced today can be compared with a later one and any difference attributed to the rules rather than to the file.
  • Sources are re-checked before release and at least monthly. Competitor pages are never treated as platform specifications.

The rulesets, in numbers

Shared

10 rules

Platform-neutral catalog health: parsing, duplicate headers, blank values, identifier uniqueness, URL and money syntax.

Google Merchant Center

35 rules

Product data specification plus the image specification, including the date-aware 500×500 change.

OpenAI Stable

19 rules

The file-upload product-discovery schema: nine required basics plus price, variant and optional-field handling.

Primary sources followed

Google:

10 official pages

OpenAI:

1 official pages

Shared/standards:

6 published standards

Each rule row on a `/rules/*` page lists its own source URL and last-verified date. The full source register is maintained with the ruleset and re-checked monthly.

How to read a result

Six states, no score: Error, Warning, Needs review, Pass, Not applicable, Not checked. A pass means the implemented check passed with the evidence available — not that the platform will accept the item. A “Needs review” is a deliberate refusal to guess, and it is the honest answer more often than a checker's marketing suggests.

See the full rule reference

Contact and corrections

Rule corrections matter more than feature requests: if an official page has changed and a rule here is stale, that is a defect. The changelog records every ruleset change with the source that triggered it, so a disagreement can be checked against a specific date rather than a memory of what a page used to say.