Skip to main content
checkreal.ai

Methodology

How checkreal.ai scores content.

The Truth Score is a single 0–100 number that summarizes a set of signals specific to the content type. This page explains what each part means and where the engine is honestly limited.

Last updated 2026-05-10.

The score

  • 80–100: likely authentic. Most signals point to a real, unmodified source.
  • 50–79: inconclusive. Mixed evidence — recommend human review.
  • 0–49: likely synthetic or manipulated, depending on content type.

The score is not a probability and not a verdict. It is a calibrated summary of the underlying signals.

Signal models per content type

Image. Texture continuity, edge and lighting coherence, compression and metadata layering. See the image detector.

Video. Frame-to-frame temporal consistency, facial-landmark stability, audio-video alignment. See the video detector and deepfake detector.

Text. Sentence-rhythm regularity, generic-transition density, claim specificity, urgency and credential-risk markers. See the text detector.

Audio. Voiceprint stability, prosody markers, background continuity. See the voice detector.

Music. Composition coherence, vocal/instrumental texture, production-artifact patterns. See the music detector.

Screenshot. UI layout consistency, transaction-risk language, image-integrity checks. See the screenshot detector.

What we explicitly don’t do

  • • Confirm a video, image, or voice is real. We report signals; provenance verification is separate.
  • • Identify which generative model produced a given output.
  • • Match a voice clone against a specific real speaker without a reference.
  • • Determine whether a claim made in a piece of content is factually true.
  • • Operate as a forensic tool admissible in court.

Known limitations

The current public engine is a demo for evaluating the format and workflow. Heavy compression — TikTok reposts, social-app screenshots, phone-call audio — strips signals that the engine relies on. Adversarially-tuned generation can pass through. False positives are most common on studio photography and formal academic English. See why AI detectors are not 100% accurate for the full discussion.

How to use a score responsibly

  1. 1. Check the source: who posted, what is their history.
  2. 2. Run the detector for a structured second opinion.
  3. 3. Cross-reference with other reporting or contacts.
  4. 4. Decide. Never let the score alone decide for you.