Responsible-use guide
How to interpret an AI detection score
An AI percentage can look precise while answering a narrow question. Learn what Genutext's scale means and how to choose the next review step.
What the Genutext score represents
Genutext displays an AI-oriented score from 0 to 100. A higher number means the submitted English prose showed stronger patterns associated with AI-generated writing in the current analysis pipeline. A lower number means it showed fewer of those patterns.
The analysis provider returns a human-oriented probability. Genutext converts it by subtracting that number from 100. For paid scans with sentence results, the converted sentence scores are averaged and rounded to form the overall display. If sentence results are unavailable, the converted document score is used.
How Genutext labels the result
| Score | Displayed label | Appropriate reading |
|---|---|---|
| 0–29 | Likely Human | No strong AI-writing signal was returned. Do not treat this as proof that the complete writing process was human-only. |
| 30–59 | Mixed Content | The result is inconclusive or varies across the text. Review the sample, sentences and writing process before drawing any conclusion. |
| 60–100 | Likely AI Generated | Stronger AI-like signals were returned. Examine the detailed context and alternative explanations; do not treat the label as a finding of misconduct. |
These boundaries are product display thresholds. They do not come from an institution's assessment policy and should never be copied into one as automatic guilt thresholds.
Read the document score and sentence signals together
A paid Genutext result can include an overall score and a score for each returned sentence. The overall number is useful for orientation; the local results help you see whether the signal is distributed or concentrated.
If most sentences cluster together
A consistent pattern may explain the overall score, but consistency itself has more than one cause. Genre conventions, a rigid template, translation or extensive editing can make human prose unusually regular.
If a few sentences are much higher
Read those passages in context. They may reflect copied boilerplate, a quotation, a definition, a different drafting session, permitted assistance or generated wording. The score cannot choose between those explanations.
If sentence results are missing
Do not infer hidden passage evidence. The current interface uses the converted document-level provider score when sentence-level results are not available.
Choose the next step from the consequence, not the colour
- 01
Low-stakes quality review
Use the signal to decide where to read more closely. A content editor may simply check factual claims, citations and voice without investigating authorship. - 02
Student feedback
Ask how the argument and sources developed. If AI use was allowed, focus on whether it was disclosed and whether the student owns the reasoning. - 03
Potential policy concern
Preserve the complete report, check the applicable rule and collect process or source evidence before making an allegation. - 04
Possible adverse action
Use the institution's established procedure, give the writer a meaningful opportunity to respond and require evidence beyond the detector result.
If the result is surprising, review possible false positives before escalating it.
How to record the result without overstating it
Accurate wording preserves the distinction between a model output and a human conclusion. A useful review note should make it possible for another person to understand what was tested and why the result mattered.
- Record the product, date and relevant workflow or version information available
- Identify the exact text or document section that was submitted
- Record whether the input was pasted or extracted from a PDF and whether extraction was checked
- Write ‘the detector returned’ or ‘the report classified’, not ‘the detector proved’
- Keep the numeric result and sentence context together rather than quoting the percentage alone
- List other evidence reviewed, including information that challenged the concern
- Record the policy-based human decision separately from the automated result
For the technical boundaries behind the number, see the Genutext methodology and limitations.