Community free access — free until October 13, 2027. All current features are free during the launch period. Fair-use limits apply. No credit card required.
Skip to main content
Verfiai
← Back to workspace

Accuracy and Limitations

We prefer published, checkable results over claimed ones. This page states what has been measured, what has not, and where the tool falls short.

Public dataset evaluation

No public-dataset evaluation of Verfiai has been completed yet. When one is run, we will publish the dataset name, the number of claims tested, the measured accuracy and the test date on this page. Until then, no accuracy figure should be assumed, and you should review the cited sources yourself before relying on any result.

Known limitations

  • Verfiai cannot read paywalled, offline, or non-indexed sources.
  • Search coverage may miss very recent events due to indexing delays.
  • Books, preprints, and references without a DOI may not be found during citation scans.
  • AI engines can occasionally misinterpret nuanced evidence or context.
  • Scanned or image-only PDFs cannot be read; paste the text instead.
  • Verdict labels are the AI's assessment of the sources it found, not proof. Open the sources yourself before relying on a result.
  • Outputs are decision-support tools, not definitive truth judgments.

How verification works

Reports break your question into individual claims, search the live web for evidence, and label each claim supported, contradicted, mixed or insufficiently evidenced. Citation checks query Crossref for retractions and record status. The method is described in full on the methodology page.