How to Review Authenticity Signals Responsibly

An authenticity or AI-writing signal is a tool’s estimate that a passage resembles patterns it associates with AI-generated text, or that it matches other sources. It is a probability or a highlight, not a factual record of how the text was written. The tool has not watched you work; it is pattern-matching against a model.

That matters because a signal answers “does this look statistically similar to X?”, not “did a person write this?”. Treat a flag as a prompt to look more closely — an invitation to review — rather than a conclusion. This is general guidance; your university’s own definitions and policy always take precedence.

No detector is fully accurate, and none should be treated as proof of misconduct. Detectors produce false positives — flagging genuine, human-written work — particularly for writing that is clear, formulaic, or by students writing in an additional language. They also carry inherent uncertainty: results are estimates, not measurements.

Different tools can return different results on the same text because they use different models, training data and thresholds. A high score from one tool and a low score from another does not mean one is “right” — it shows how tentative these signals are. Responsible institutions treat a signal as one input to a human judgement, alongside context and academic-integrity processes.

If a section is flagged, review it calmly and in context. Read the passage within the whole document, and check your own drafts, notes, outline and version history — these show your writing process and are your best record of authorship. Compare the style of the flagged lines with the rest of your work: is it consistent with how you write elsewhere?

A short example. A student’s introduction is flagged. Instead of panicking, they reread it in context, and confirm they wrote it — they have earlier drafts and notes. They notice the flagged lines are generic, formulaic phrasing, so they rewrite them in their own clearer voice with specific detail. If anything is genuinely unclear, the responsible next step is to ask the tutor for clarification rather than guess. A flag is a reason to look again, not to hide anything.

Universities set their own rules on whether and how AI tools may be used, and those rules vary by institution, course and even assignment. Read your AI-use policy and academic-integrity guidance before you use any tool, and follow any declaration or disclosure requirement — some courses ask you to state what AI use was permitted and how you used it. When in doubt, ask; disclosing permitted use is always safer than assuming.

You can read how we think about responsible AI in study on our AI usage policy and academic integrity pages, and the QAA publishes sector guidance on academic integrity for UK higher education — but your own institution’s policy is the one that governs your work.

The goal is never to disguise writing so a detector stays quiet; that solves nothing and risks your integrity. If your own writing was flagged, the constructive response is to make it clearer and more your own: add specific detail, examples and your own reasoning, and replace generic, formulaic sentences with precise ones in your voice. Writing that reflects your genuine understanding is both stronger academically and less likely to read as generic.

That is a normal editing skill — the same work as turning description into analysis. Reviewing your own draft for clarity and argument is where our Marker’s Eye can help; you make the changes, and every word stays yours.

These responses make things worse:

• Panicking and assuming a flag means you are in trouble.
• Treating a detector score as proof, for yourself or anyone else.
• Trying to “beat” detection instead of improving the writing.
• Ignoring the flag rather than reviewing the section.
• Not reading the AI-use policy or skipping a required disclosure.
• Not keeping drafts and notes that evidence your own process.