LocaleGuard AI — Multilingual Product Specification Audit

Check whether two localized pages for the same product publish different specifications. Get original field evidence, normalized values, product identity, source URLs, timestamps, content hashes and an actionable HTML/CSV/JSON report.

What it checks

Compare already localized pages, in either direction, using explicit JSON-LD Product fields, additionalProperty entries, specification tables, definition lists and adjacent labelled specification rows. This tool does not generate translations.

Supported specification vocabularies: English, Italian, French, German, Spanish, Portuguese, Dutch, Polish, Turkish, Russian, Japanese, simplified Chinese and Arabic. Any two different supported languages can be paired; no Italian or English page is required. Examples: French ↔ German, Spanish ↔ Japanese and Arabic ↔ Portuguese.

The profile is taken from HTML lang, including regional tags such as fr-CA and de-DE. If HTML has no language declaration, provide sourceLanguage and/or targetLanguage (codes: en,it,fr,de,es,pt,nl,pl,tr,ru,ja,zh,ar). A hint cannot override a conflicting declaration. The tool reads declared language; it does not independently certify the language of all text. Unsupported languages return incomplete instead of a guessed result.

Supported fields: material/composition, internal lining, dimensions, weight, capacity, connector and explicitly included/excluded accessories. Decimal commas, supported unit conversions and an explicit vocabulary of common material/accessory names are normalized. It is not a complete dictionary for these languages. Unknown text, uncertain inclusion claims and conflicting information on the same page remain unverified.

Product identity requires a shared GTIN, an exact SKU on the same host or brand, or an exact MPN on the same host or brand. Conflicting GTINs stop comparison; a shared GTIN is the primary identifier. Pages with multiple Product nodes require an exact expectedSku. Product titles alone are never sufficient.

Quick start

Choose Synthetic example and Start to see a German/French example with two intentional differences: material and charger inclusion. Dimensions expressed in equivalent units agree. This example uses no website requests or external AI services.

For your products, choose Audit my product pairs:

{
  "mode": "audit",
  "pairs": [
    {
      "id": "my-product-123",
      "sourceUrl": "https://your-shop.example/fr/products/item-123",
      "targetUrl": "https://your-shop.example/de/products/item-123",
      "expectedSku": "ITEM-123",
      "ignoreFields": []
    }
  ]
}

Replace the example URLs with your public product pages. For a fully offline audit, supply both sourceHtml and targetHtml as HTML strings in the pair. These snapshots are supplied by you and are not independently verified as live website content.

Maximum: 20 pairs per run; 2 MB per page/snapshot, 8 MB total snapshots. Public HTTP(S) pages only. Private networks, credential-bearing URLs, nonstandard ports and access challenges are rejected. robots.txt is checked, requests and redirects are bounded. No login, proxy, CAPTCHA solving, browser rendering or site changes.

Read the result correctly

Status	Meaning
differences_found	At least one supported specification differs; review the two sources.
no_differences_in_checked_fields	Compared fields agree. This does not verify the entire page.
incomplete	Insufficient, ambiguous or unsupported data.
unmatched	Same-product identity could not be established.
fetch_error	The page could not be retrieved under the access rules.

Coverage lists every supported field, including missing/ambiguous and deliberately excluded fields. Neither language page is assumed authoritative. Consider intentional market differences and variants before correcting content. Changes are always manual.

Pricing

USD0.20 per usable product comparison, with no LocaleGuard subscription, plus Apify's USD0.00005 Actor-start event for this 256 MB Actor. Platform usage is included. A pair with established identity and at least one comparable supported field counts as one comparison, whether it contains a difference or not. The synthetic demo, unmatched pairs, fetch errors and entirely incomplete pairs trigger no comparison event. See the Pricing tab for effective prices and any pending price transition. For example, 20 useful pairs cost USD4 plus one start event.

Outputs and integrations

Open REPORT for the HTML report, COMPARISONS for CSV, OUTPUT for JSON and the default dataset for pair results. Results are suitable for API/MCP, scheduled audits, spreadsheets and automation workflows. Runs and supplied HTML stay in your own Apify storage according to its retention/access settings.

Scope and limitations

This is a focused deterministic audit, despite the product name AI. No LLM or external AI API is used. It is not a translation quality certification, legal review or complete semantic checker. Free-form marketing claims, prices, currencies, shipping, returns, image content, screenshots, colors and sizes of selectable variants are outside this release. JavaScript-only, blocked or unsupported markup may return incomplete. Unknown words and differing dimension counts are not guessed. No automatic correction and no guaranteed impact on sales.

Run locally

Requires Python 3.12+ and BeautifulSoup 4.13.5. From this directory:

python -m src.cli
python -m src.cli --input examples/audit.json --output validation/my-audit
python -m unittest discover -s tests -v

The local CLI does not use Apify credentials or charge comparison events.

Validation

The multilingual release passes 56 local test methods, including all 156 ordered pairings of the 13 profiles on controlled seven-field fixtures, plus negations, unknown non-Latin words, regional tags, explicit language hints and French/German composition differences. A live French/German manufacturer pair was also retrieved successfully: its shared GTIN matched and material, weight and capacity agreed. Dimensions presented as several separate labelled axes were left incomplete instead of guessed. The previous release compared two real Italian/English product pairs locally and on Apify, with five concordant supported specifications. These tests do not certify all language expressions or websites. A deliberately changed local snapshot was correctly flagged. Broader website coverage and a real customer's payment remain untested.