Google Merchant Feed Validator & Price Checker

Check a Google Shopping / Merchant Center product feed before upload. Find malformed prices, image URLs, duplicate IDs and other selected feed issues, with the row, product ID and correction guidance in one report. Optionally compare a small sample of product pages with their published Product/Offer JSON-LD.

Up to 10,000 feed rows per report. Base price: $1.00 per completed report + $0.00005 per start; platform usage included. HTML, CSV and JSON are included, with no separate fee for each product or finding. The live Pricing tab shows your current price and tier discounts.

Preview the correction report

The prefilled synthetic feed has 2 products. The second product produces these 4 findings:

Field and sample value	Report guidance
image_link: /missing-image.jpg	Use an absolute HTTP(S) image URL
price: 12.345 USD	Review price precision above two decimal places
sale_price: 15.00 USD, regular price 12.345 USD	Sale price should be lower than regular price
availability: preorder, no availability date	Provide the expected availability date

The report identifies product P2 on line 3 and separates 2 errors from 2 warnings. This is a static output preview; reading it does not start a run.

Run the two-product sample with the displayed Actor price, or replace it with your own feed. The sample keeps website checks off.

File validation is the default. Website comparison is optional and samples 1–10 rows; unsupported or inaccessible page markup returns unknown. This independent utility checks a stated subset of Google's requirements and does not guarantee Merchant Center approval.

What the report checks

• Required core fields; title or structured_title and description or structured_description alternatives.
• Duplicate IDs, ID/title/description length and inconsistent row widths.
• Price/currency syntax, extra price decimals, sale-price comparisons and zero-price review.
• Link/image URL syntax, supported availability and condition values, and dates for preorder/backorder.
• Optional single-GTIN length/check digit, plus explicitly labeled identifier-review cases.

Checks distinguish errors, warnings and review findings. Conditional cases receive review findings when the feed alone cannot establish eligibility. Structured title/description values satisfy the presence check, but their subfields, content and AI provenance require separate validation. Missing GTIN, brand, MPN or condition is not automatically labeled invalid.

Run it

1. Choose Paste CSV / TSV text or Upload CSV / TSV file under Feed source. Include lowercase English headers on the first line. Upload mode ignores the demo text, so you do not need to delete it.
2. Select CSV (comma) or TSV (tab). The delimiter is explicit; it is not guessed.
3. For a private uploaded file, select the temporary key-value store created by the upload under Uploaded file store (read access). Only that store receives READ permission. Pasted text and valid signed upload links do not need a store selection.
4. Leave page checks off for feed validation, or enable Check product-page prices and availability, confirm your permission to check the supplied pages, and choose a sample size of 1–10 rows.
5. Start the Actor and open Review report. Download CSV findings or use JSON report in an automation.

Example input:

{
  "sourceMode": "text",
  "feedText": "id,title,description,link,image_link,availability,price\nAPRON-1,Cotton apron,Plain cotton apron.,https://example.com/apron,https://example.com/apron.jpg,in_stock,19.95 USD\n",
  "delimiter": "csv",
  "checkPages": false
}

For uploaded input, use sourceMode: "upload", feedFileUrl and, for a private upload, sourceStores: ["YOUR_UPLOAD_STORE_ID"]. The file picker provides the Apify record URL. Only HTTPS api.apify.com/v2/key-value-stores/.../records/... URLs are accepted. Signed URLs must contain only their storage signature; API-token query parameters are rejected. Existing API clients that send feedText and omit sourceMode continue to work. When the source mode is omitted, supply exactly one nonblank text or upload source.

The prefilled example is synthetic, produces four feed findings, and makes no product-page requests. For a first test, run it unchanged. Replace the example.com product links with your own authorized product pages before enabling online checks.

Example: invalid GTIN digits and identifier conflicts

Check invalid GTIN digits in a product feed CSV contains three synthetic products:

Product ID	Selected check result
DEMO-DIGIT	GTIN_CHECK_DIGIT: final check digit does not match
DEMO-CONFLICT	IDENTIFIER_CONFLICT_REVIEW: GTIN is supplied while identifier_exists says no
DEMO-VALID	No findings in the selected rules

The report has 1 error and 1 review finding. The identifiers are checksum fixtures, not claims that those identifiers belong to the synthetic products. A valid check digit does not verify assignment or Google approval. Page checks stay off. Reading this preview does not start a run; running the example uses the displayed Actor price. At base pricing, one report plus startup totals $1.00005.

Optional website comparison

The checker reads server-rendered JSON-LD from the first 1–10 valid-width product rows in feed order; the default sample is 5. It compares a selected Product/Offer's price, priceCurrency and availability with the feed. Product SKU or GTIN and exact product/offer URLs help select the product and variant. Queries remain part of internal variant matching but are omitted from report evidence.

Each field is match, mismatch or unknown. A mismatch means the feed and observed JSON-LD differ; it is a review warning, not a claim that the checkout is wrong or Google will disapprove the product. A page can have a definite availability difference while its price remains unknown.

• Multiple unresolved products/offers, aggregate price ranges, lost variant identity, missing JSON-LD and inaccessible pages return unknown.
• Sale prices, price specifications, member/quantity/region restrictions and tax-related price context receive review instead of a definite price mismatch. This first version does not choose an active sale or compute tax, shipping or currency conversion.
• JavaScript-only markup, RDFa, Microdata, rendered prices, cookies, login, browser challenges, remote JSON-LD contexts, checkout, images and website changes over time are outside scope. Pages with unsupported markup may be useful to a browser but unknown to this checker.
• Only public HTTP(S) URLs on standard ports are fetched. Private, loopback, link-local, reserved and multicast addresses, credential-bearing URLs and sensitive query keys are rejected. DNS answers are validated and connections are pinned to a public address. Redirects are capped at 3 and must stay on the same hostname; HTTPS cannot redirect down to HTTP.
• Each page is limited to 512 KB and 6 seconds, with 2 concurrent requests and 32 seconds total for the online phase. Compressed responses are not decoded. The checker requests uncompressed HTML; a server that ignores this request returns unknown. Raw HTML is discarded after extracting the selected fields. There are no proxies, retries, images, AI calls or browser fees.

The report states how many rows were not checked. A sample is not a full-site audit. Separate network failures remain in the report, so one blocked page does not discard useful results from other pages. Choose your sample order deliberately if you want to prioritize high-risk products.

Limits and scope

Each run accepts up to 10,000 nonblank product rows, 2,000,000 characters and 40 columns. Uploaded files must also be uncompressed UTF-8 and no larger than 2,000,000 bytes. Blank lines are ignored. Headers and cell edges are trimmed. Duplicate headers are rejected. CSV quoting supports embedded commas and newlines; a finding's line number is the record's ending physical line. XML feeds and arbitrary remote feed URLs are outside scope.

Reports contain complete finding counts and at most 500 detailed findings. Truncation is explicit. Finding values are capped at 1,000 characters and IDs at 200 characters. HTTP(S) URLs in findings omit query strings; credential-bearing URLs are replaced with a safe placeholder. Fix reported issues and rerun to expose remaining findings. The JSON column inventory identifies fields outside the checked subset.

With page checks off, product URL checks are syntax-only. Neither mode verifies domains, image dimensions, checkout or complete price accuracy. Three uppercase currency letters pass the syntax check even if that currency is unsupported. Prices longer than 30 numeric characters receive review instead of numeric comparison. A valid GTIN checksum does not establish registration or correct product assignment. Multiple GTINs, category/country rules, shipping, tax, account policies and approval status remain outside scope. Date syntax is checked without deciding whether the date is correct or still in the future.

Google's requirements can change. A report with no findings means no problems were detected by this tool's listed rules.

Outputs and automation

• report.html: self-contained, escaped HTML report with no scripts or remote assets.
• REPORT: JSON summary, column inventory, scope notes, findings and pageChecks evidence when enabled. Each page check contains its line/ID, query-free URL, status, match method and selected field values/reasons when available.
• findings.csv: summary row, findings and optional PAGE_EVIDENCE rows. Formula-like cells receive a leading apostrophe for spreadsheet safety; JSON preserves the parsed text within the stated preview limits.
• Default dataset: exactly one summary item per completed report. Product rows are not individual billable results.

Use the run's output links or key-value-store records after the run succeeds. A recurring workflow can submit its fresh export, wait for completion and route the summary or report to its existing review process. The Actor does not submit or alter a Merchant Center feed.

A practical recurring sequence is: create the fresh feed export → submit its text or upload through the Actor input → wait for a successful run → retrieve the default dataset's single summary → inspect rowsWithErrors, rowsWithWarnings, rowsNeedingReview and pageCountsByStatus → route report.html or REPORT to your own review step. Do not mark unchecked/unknown pages as valid or automatically change prices based on this report. Apify schedules repeat saved input; the upstream workflow must refresh that input or upload record when the feed changes.

Pricing

USD $1.00 per completed feed report, plus $0.00005 per Actor start. A completed report totals $1.00005, with platform usage included. There is no separate per-product or per-finding fee. A report that finds problems is still a completed report. Set the maximum run charge to at least $1.00005; $1.01 covers one report and startup. Unsupported input fails without creating a report-result event; the startup fee may still apply. Data is stored in the run's Apify storage.

Rule references

Feed rules reviewed September 5, 2026; upload and structured-data support reviewed September 7, 2026. These primary sources explain the corresponding attributes and platform features; the implementation intentionally checks a limited subset:

• Google product data specification: required fields, alternative structured text and text lengths.
• Price and sale price: amount/currency format, precision, sale comparisons and conditional zero-price cases.
• Availability, availability date, and date formatting: supported values and conditional dates, including vehicle-only build_to_order.
• GTIN and structured-data identifier guidance: supported identifier lengths and check digits.
• Condition: supported condition values and conditional requirement.
• Google Product/Offer merchant-listing structured data and Schema.org Offer: selected page fields and price context. This utility does not implement the complete specifications.
• Apify input schema and resource permissions and dataset schemas: native uploads, selected storage access and typed summary output.

Local verification

Run python -m unittest -q from the source folder. The 56 validator, upload, network-transport, comparison, rendering and lifecycle tests use only the standard library, synthetic data and injected transports; the hosted lifecycle uses Apify SDK 4. They cover private-address/DNS/redirect blocking, private-store authorization, unknown/partial results, time and concurrency caps, spreadsheet/HTML safety and the one-summary billing contract. No customer data or live product requests are needed.

python prepare_package.py regenerates schemas and synthetic reports. example-page-report.html illustrates two observed differences using the local example-product-page.html fixture through an injected fetcher; it is not evidence of a real live-page run. The default example-input.json keeps page checks off.

More previews and related tools

Watch the 32-second synthetic portfolio demo for a walkthrough of the four tools. The video uses sample data.

• Shopify CSV Import Checker & Supplier Catalog Diff
• OpenAPI JSON Diff & Breaking Change Checker
• Greenhouse & Ashby Job Change Monitor