Product Feed Health Checker

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

Local preflight · CSV, TSV, JSONL

Product feed validator

Validate a product feed export before a platform does: parsing fidelity, identifier uniqueness, URL and money syntax, then the destination rules you actually selected.

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

Formats

CSV, TSV, JSONL (UTF-8, optional BOM)

Shared checks

10 rules

Total rules

64 across three rulesets

Last verified

11 November 2026

What “valid” actually means for a product feed

A feed is not valid or invalid in the abstract. It is valid for a destination, for a product context, and as of a date. A file that is perfectly acceptable for a US-only catalog of new items can fail as soon as a row claims a used condition without a condition value, and an image that passes today can fail after 2027-01-31. Any tool that returns one number hides exactly the information you need.

So the checker separates three layers. Shared catalog health is platform-neutral: parse integrity, blank values, duplicate identifiers, URL and money syntax. Platform rules are then applied only to the destination you selected. Evidence-based page checks are a third, separate, opt-in step.

What the import and shared checks verify

  • The file decodes as UTF-8 (BOM tolerated) and parses as CSV, TSV or JSONL without a structural error — an isolated quote or a truncated line is reported with the row number, not silently skipped.
  • Column headers are preserved exactly; duplicated headers are blocked and must be resolved explicitly rather than guessed.
  • Identifiers stay strings. A GTIN with a leading zero, a 13-digit EAN, or a numeric-looking SKU is never coerced into a float.
  • The original row and the normalized value are kept separately, and unmapped columns are preserved so a later export cannot silently drop them.
  • Blank values are detected after Unicode whitespace trimming — a field containing only non-breaking spaces fails the same way an empty field does.
  • Item IDs must be unique per selectable destination; duplicates are reported on every affected row.
  • URLs must be absolute HTTP(S), without embedded credentials; protocol-relative, data, file and script URLs are rejected.
  • Money and date syntax is validated per destination, and a currency is never inferred from a symbol.
  • The row cap is enforced with an explicit error. Data is never silently truncated.

What it cannot tell you

  • Whether the platform will accept the item. Approval depends on account context, policy review and data you did not send.
  • Whether the product page agrees with the feed. That requires fetching the page, which happens only if you opt in to remote checks.
  • Whether a category-specific rule applies when the category cannot be determined from the file — that becomes Needs review, never a guessed error.
  • Whether an identifier is real. The checker validates the GTIN check digit and length; it cannot know whether the manufacturer assigned that GTIN to that product.

What a report looks like

Each issue carries the rule ID and version, the platform, the category, the field, the raw value from your file, the normalized value the checker used, a plain-English explanation, the fix suggestion, the official source URL, and the as-of date. Rows are filterable by platform, severity, field, product and result state, and every issue links back to its source row so you can find it in the export.

The same run records the ruleset version, so a report produced today can be compared with a later run even after the rules changed — and a cross-version comparison is labelled as such rather than silently blended.

Input limits

  • Up to 10 MiB and 10,000 records per import, with an explicit error when exceeded. These are engineering defaults, not platform channel limits.
  • Compressed archives (ZIP, GZ) are out of scope for the MVP; decompress before import so the parser sees the real row count.
  • XML feeds are out of scope for the MVP. Convert to CSV or JSONL first.
  • Parsing happens locally. No account is required for the first diagnostic run.

Frequently asked questions

Do I need an account to validate a file?

No. Local parsing and the shared/Google/OpenAI rules run in the browser. An account is needed only for remote scanning and for a paid plan.

Is my feed uploaded?

Not by default. The file is parsed in your browser. Only URLs you explicitly submit go to the remote scanner, and never the feed rows themselves.

Why does the tool refuse to fix some issues automatically?

Because some values cannot be derived from the file: a real GTIN, a real brand, a real price or a working image URL. Inventing them would create a feed that passes a checker and fails the platform. Only deterministic repairs are offered, always previewed first.

Can I validate an OpenAI feed with the Google ruleset?

No, and neither can this tool: the rulesets are isolated. Google-specific semantics (for example its availability enum) must never be applied to OpenAI fields. Shared catalog checks run across both.

Related

Official sources for the shared checks are the RFC 4180 CSV format, the WHATWG URL standard (url.spec.whatwg.org), ISO 4217 for currency codes and ISO 8601 for dates. Each rule row on a `/rules/*` page lists its own source and last-verified date.

See what the free tier covers