How userbugs.ai builds its bug reports

Where the data comes from

Our sources are public web pages — forums, review pages, and community discussions that anyone can open in a browser. We discover them through broad, multilingual web search, so reports from regional communities (not just English-language ones) are represented. The discovery and ranking techniques themselves are proprietary; what matters for trust is that every source is public and every published bug links back to it.

How reports are extracted

AI models read the discovered pages and extract structured defect reports: what product, what failed, under what conditions. We never republish verbatim source text — only the structured findings, in our own words. (Which AI providers we use is disclosed in our Privacy Policy’s sub-processor list.)

Citations and claim verification

Every published bug cites its source URLs. Before a report is shown, an AI claim-verification step checks whether the cited page actually supports the claim. Verdicts are supported or partial; reports whose claims are unsupported by their citations are hidden from the site.

Corroboration tiers

Corroboration counts independent source domains and regions — not raw mention counts:

• 1 source — anecdotal: one place on the web reports it
• 2 independent domains — corroborated
• 3+ independent domains — stronger corroboration
• 3+ regions — multi-region: the issue is reported across geographically distinct communities

Official recall data

Recall notices shown on userbugs.ai come from the U.S. Consumer Product Safety Commission (CPSC) via the public SaferProducts.gov recall API. CPSC recall data is a public-domain work of the U.S. government; we display it with courtesy attribution.

• Every recall we show links to the official cpsc.gov notice, and the hazard wording is the notice’s own — we do not paraphrase it.
• Matching a recall to a product in our catalogue is precision-biased: uncertain matches are reviewed by a human before they are displayed.
• Recall information is never paywalled or tier-gated — it is shown to every visitor, signed in or not, on every plan including free.
• EU Safety Gate (RAPEX) coverage is planned, with attribution per Commission Decision 2011/833/EU.

Severity and reproducibility

Severity and reproducibility ratings are AI-assessed estimates based on what the cited sources describe. They are not lab measurements — we do not physically test products.

AI-generated content (EU AI Act transparency)

Bug summaries on userbugs.ai are AI-generated from the cited sources. This disclosure is made under Article 50 of the EU AI Act. The summaries are synthesised in our own words; we never republish verbatim text from source pages.

Limitations

• Source pages can go offline after we cite them — a citation link may stop resolving.
• Reports may lag reality: a defect may have been fixed by a firmware or hardware revision after the sources were written.
• AI extraction and verification reduce errors but do not eliminate them — a report can still be misclassified or out of date.

What we do NOT do

• No authenticated scraping — we never log in to a platform to collect content.
• We respect platform terms — we do not collect from sources that prohibit it or that require logging in.
• No raw page storage — we keep structured findings, not copies of source pages.
• No personal-data profiles of forum users — we track products and defects, not people.
• Erasure requests are honoured — see our Privacy Policy for how to exercise your rights.

Disputes and corrections

If you are a vendor and believe a published report about your product is wrong, contact us at legal@userbugs.ai. We will review the cited sources and correct or remove reports that don’t hold up.

See also our Beta Terms of Service and Privacy Policy.