Responsible use
When You Shouldn't Rely on an AI Detector
Use nine stop and caution rules to keep sentence labels, unsupported formats and high-stakes decisions within defensible limits.
On this page
Do not rely on an AI detector when the input falls outside the tool's documented scope, the result is the only evidence, or the decision could significantly affect a person's education, employment or reputation without meaningful human review.
AI detection can support screening and conversation. It is a poor substitute for evidence about the actual writing process.
Nine stop or caution situations
1. The sample is too short
Very short text provides too little pattern evidence. A sentence label is particularly easy to over-interpret.
2. The format is unsupported
Code, poetry, scripts, bullet lists, tables and heavily formulaic material may sit outside a text detector's intended use. Turnitin explicitly excludes several of these from reliable qualifying prose.
3. The language is unsupported
Do not run one language through a model validated only for another and treat the output as meaningful.
4. The work mixes authors or assistance
Group work, templates, quoted material, translation and iterative human/AI editing make a single authorship label too coarse.
5. The detector is the only evidence
Turnitin says its report may misclassify writing and should not be the sole basis for adverse action. A score is not a drafting history.
6. The policy is unclear
A detector cannot decide whether brainstorming, proofreading or generation was permitted. Clarify the rule before judging the behaviour.
7. The tool is not approved for the data
Uploading student, client or confidential work to an unapproved service can create privacy, contractual and intellectual-property risks.
8. The decision is high stakes and automated
Grades, misconduct findings, hiring and publication decisions require accountable human judgement, transparent evidence and a route to challenge.
9. You are scanning until a tool agrees
Repeatedly trying detectors until one returns a severe result is confirmation bias. Different tools use different thresholds and score definitions.
Try the text in context
Run a free AI-writing signal check
Paste 300 characters to 350 words. The sample is analysed for a first-pass signal and then discarded.
Why sentence-level verdicts are fragile
A sentence contains limited evidence and is strongly shaped by its topic. “The results demonstrate a statistically significant difference” is conventional academic language. Highlighting it does not show who wrote it.
Sentence-level outputs can still help a reviewer locate passages, but they should be read in document context:
- Is the language normal for the discipline?
- Does the passage connect accurately to cited evidence?
- Did it develop across drafts?
- Can the writer explain the claim and wording?
- Is the model documented for this language and genre?
Avoid writing “this sentence is 95% AI” unless the provider explicitly defines and validates that unit. Most tools cannot support the forensic certainty the phrase implies.
What to use instead
Depending on the question, stronger evidence may include:
- Authorship: outlines, drafts, version history, notes and an explanation of choices.
- Source use: a similarity report, source verification and citation review.
- Factual accuracy: direct inspection of cited sources and reproducible calculations.
- Policy compliance: the assessment rule and a documented disclosure.
- Learning: oral explanation, formative checkpoints and authentic assessment design.
- Publication quality: editorial review, fact checking and source provenance.
No single alternative is perfect. The strength comes from independent evidence converging on the same account.
A proportionate use framework
| Use | Detector role | Required safeguard |
|---|---|---|
| Personal curiosity | Low-stakes signal | Read limitations and avoid rewriting to a score |
| Editorial triage | Prioritise review | Human editor checks sources and context |
| Classroom conversation | One prompt among others | Neutral questions and process evidence |
| Formal misconduct case | Limited supporting evidence | Published procedure, disclosure, response and reasoned decision |
| Automatic penalty | Do not use alone | Meaningful human decision and independent evidence |
Genutext deliberately describes its output as a signal for human review. Its accuracy methodology and detection-as-evidence guide explain the boundary.
Frequently asked questions
Should teachers use AI detectors?
They may use an approved detector as a limited signal if institutional policy permits, but should not treat it as proof or the sole basis for an allegation or sanction.
Can an AI detector prove cheating?
No. It classifies text patterns. Cheating depends on conduct, permission, disclosure and evidence assessed under a policy.
Is it safe to check one sentence?
Sentence-level results are fragile and context-poor. Use them only to locate material for review, not to declare sentence authorship.
What if my institution bans AI detectors?
Follow that rule. Do not upload work to an external service independently; use approved process evidence and seek guidance from the relevant academic-integrity or data-protection lead.