Pattern likelihood
The service estimates how strongly text patterns resemble the patterns represented in its detection models.
Responsible interpretation
How to interpret an AI Sentinels result.
Detection helps focus review. It does not know who wrote a text, why a pattern appears, or what institutional decision should follow.
The central rule
A review signal, not a verdict.
A score summarises model output for the submitted text. It should be read with the text, paragraph patterns, assignment context, drafting evidence, applicable policy, and a fair opportunity for human explanation.
Variation
Sentence and paragraph variation can contribute context about uniform or mixed writing patterns.
Predictability
Word-choice predictability can add a signal, but it can also reflect genre, editing, language background, or highly structured writing.
Paragraph context
Paragraph-level indicators help reviewers locate variation inside a document instead of treating the whole text as uniform.
Interpretation workflow
Move from indicator to fair review.
Check the input
Confirm the correct text, version, language, and assignment context were analysed.
Read the pattern
Review the overall result together with paragraph-level variation and the text itself.
Gather context
Use drafting history, sources, process evidence, conversation, and policy-relevant facts.
Make a human decision
Apply the institution’s own process, responsibilities, and appeal or review safeguards.
Do not use alone
A detector result should not be the sole basis for disciplinary, grading, employment, or other consequential action.
Expect limitations
Short, edited, formulaic, translated, or domain-specific writing can require extra caution and context.
Evaluate the workflow, not only the score.
A governed pilot can test how results fit your academic process and review safeguards.
Interpreting a result
- How should an AI detection score be interpreted?
- As a review signal, not a verdict. A score summarises model output for the submitted text and should be read together with the text itself, paragraph patterns, assignment context, drafting evidence, applicable policy, and a fair opportunity for human explanation. Detection helps focus review; it does not know who wrote a text, why a pattern appears, or what institutional decision should follow.
- Can an AI detector prove that a student used AI?
- No. The service estimates how strongly text patterns resemble the patterns represented in its detection models, which cannot establish authorship or intent. A detector result should not be the sole basis for disciplinary, grading, employment, or other consequential action. The institution applies its own process, responsibilities, and appeal or review safeguards.
- What can make an AI detection result misleading?
- Short, edited, formulaic, translated, or domain-specific writing can require extra caution and context. Word-choice predictability adds a signal but can also reflect genre, editing, a writer's language background, or highly structured writing. Paragraph-level indicators exist so reviewers can locate variation inside a document instead of treating the whole text as uniform.
- What is the recommended review workflow after a detection result?
- Four steps. Confirm the correct text, version, language, and assignment context were analysed. Read the overall result together with paragraph-level variation and the text itself. Gather drafting history, sources, process evidence, conversation, and policy-relevant facts. Then make a human decision under the institution's own process, responsibilities, and appeal or review safeguards.