Facts first. Conclusions next.

FactifAI is designed to make a fact check inspectable - you can see what was checked, which records informed it, what the evidence supports, and what remains uncertain.

Our standard

A verdict is the end of the trail - not the source of truth.

Language models can organize and explain evidence, but they can also be incomplete or wrong. FactifAI therefore uses web research to retrieve source material before it asks the model to synthesize a report.

Every factual conclusion should be traceable to a displayed source. When the record is thin, contested, or time-sensitive, the report says so.

The research path

From a sentence to a sourced report.

  1. 01

    Frame the claim

    Identify checkable assertions, key entities, time period, and implied comparisons. Opinion and prediction are labelled as such rather than forced into a true-or-false verdict.

  2. 02

    Retrieve the record

    Search for relevant primary records, official datasets, research, and established reporting. User-supplied context helps disambiguate the statement but is never accepted as proof by itself.

  3. 03

    Compare evidence

    Look for support, contradiction, scope changes, outdated framing, and missing context. Sources are assessed by proximity to the underlying fact and methodological transparency.

  4. 04

    Explain the conclusion

    Return a structured verdict, short justification, atomic findings, key facts, source links, and limitations. Paid reports add deeper analysis and a PDF.

Verdict scale

Labels with room for nuance.

The label summarizes the evidence; the justification and citations carry the real meaning.

True

The central factual claim is supported by strong available evidence.

Mostly true

The core is supported, with a limited omission or imprecision.

Mixed

Material parts are supported and contradicted, or framing changes the meaning.

Mostly false

A small element is accurate, but the central impression is unsupported.

False

The central factual claim conflicts with strong available evidence.

Unverifiable

Reliable public evidence is insufficient to reach a responsible conclusion.

Not fact-checkable

The statement is opinion, prediction, rhetoric, or too ambiguous to test.

Source quality

Closer to the fact is usually better.

No source type is automatically correct. We favor evidence with direct access to the underlying event or data and clearly described methods.

  1. Strongest signalPrimary records and official data

    Laws, filings, transcripts, raw datasets, court records, and direct institutional releases.

  2. CorroborationPeer-reviewed and expert research

    Work with visible methods, appropriate scope, and relevant expertise.

  3. ContextEstablished reporting

    Transparent journalism that cites records, named experts, and on-the-record sources.

  4. Lead onlyUnverified posts and summaries

    Useful for discovery, but not treated as decisive evidence without corroboration.

What FactifAI cannot promise.

Public records can be incomplete. New evidence can change a conclusion. Search results can omit relevant material, and AI can misread a source despite safeguards. FactifAI is a research aid - not a substitute for professional legal, medical, financial, or investigative judgment.

Full analysis performs a deeper research pass and may revise the initial label. When that happens, the report preserves the basic verdict and explains why the assessment changed.


Check a claim