Skip to main content

How to read your detection score

James
Updated

Quick answer: The percentage at the top is the overall AI probability. The colored highlights in your text show which parts read as AI (red), mixed (yellow), or human (green). The grid below shows 8 detectors' individual verdicts.

The three things you see

1. The overall score

Close-up of the detection gauge, outlined in red — a half-circle gradient from green to red with the headline "100% AI GPT" in the center.

A single percentage — the model's combined confidence that the text is AI-generated. Higher = more AI-like.

  • Below 25% — most detectors read it as human.
  • 25-60% — mixed signals.
  • Above 60% — most detectors are flagging it.

2. The color highlights in your text

Input text panel on the left, outlined in red, with every sentence shaded in the red "AI" tint. The detection gauge, per-detector grid, and color legend (AI-generated / Mixed / Human-written with their key colors) appear on the right.

Every sentence (or short segment) gets a color:

  • Green — reads as human.
  • Yellow — mixed; some detectors flag it.
  • Red — most detectors flag it.

Use the colors to find the segments that need attention. A red sentence in an otherwise green paragraph is a candidate for Selective Rephrase.

3. The per-detector grid

Per-detector grid outlined in red — "Cross-checked with:" panel listing all 8 detectors (Turnitin, Copyleaks, OriginalityAI, GPTZero, Crossplag, Sapling.ai, Gowinston.ai, ZeroGPT) in a 4×2 layout, each marked with its individual verdict.

Eight rows, one per detector — Turnitin, Originality.ai, Crossplag, Gowinston.ai, Copyleaks, GPTZero, Sapling.ai, ZeroGPT. Each shows pass or fail for the input.

This grid is the most honest part of the score. Detectors disagree. If 6 say "human" and 2 say "AI," that is a real signal — not a bug.

Note: GPTinf does not pick a "winner." It shows you every result so you can decide which detectors matter for your context (academic vs. corporate vs. SEO).

What a "good" score looks like

It depends on the audience:

  • For an essay where Turnitin is the gatekeeper, you need that specific row to pass.
  • For a blog post, the overall percentage matters more than any single detector.
  • For a research paper, the mix of low overall + at least 6 of 8 passing is a strong sign.

FAQ

Why does the score change when I run the same text again? Some detectors are non-deterministic — they sample. Small variation between runs is normal.

Can I share the score with someone? The score is rendered in-browser. Take a screenshot if you need to share it.

Why is one detector always strict? Some detectors have lower thresholds. That is a property of the model, not your text.

Related

Comments

0 comments

Please sign in to leave a comment.

Powered by Zendesk