How Teachers Read AI Scores on Year 7 Baseline Writing

September brings a problem detection tools rarely talk about: a high AI likelihood score on a child you have never taught. You run a Year 7 baseline writing task through your detection tool, a 78% comes back, and you have nothing to compare it against. You do not know this student's voice, their range, or what they are capable of on a good day. AI detection for teachers is hard enough when you know the writer; on a baseline task it asks you to make a fair judgment about a stranger.

This post is for the moment the score lands on a new arrival — not for building a case, but for reading the number sensibly when you have almost no context. The question worth answering is "this came back high on a child I do not know yet, so what do I actually do?", not "what makes writing look AI-generated?"

Why Baseline Writing Skews Detection Scores

Year 7 baseline writing is one of the noisiest inputs a detection tool ever sees, and the reasons have nothing to do with cheating. A likelihood score measures how closely the linguistic patterns in a piece — sentence rhythm, vocabulary spread, predictability of word choice — resemble patterns large language models tend to produce. Plenty of perfectly honest baseline writing trips those signals.

Transition writing is heavily scaffolded. Many primary schools teach a "success criteria" sentence formula that produces uniform, polished, slightly mechanical prose — exactly the texture a model flags. Some children had real help at home over the summer, redrafting with a parent until the writing reads above their independent level. EAL students who have learned English through structured, formal models often write in a register that reads as unusually even. None of this is dishonesty, and all of it can push a score up.

Treat a baseline score as a reading about the text, not a reading about the child. You have no behavioural history, no prior books, no sense of their normal voice. The score cannot tell you which of these explanations applies — only you, over the coming weeks, can.

The Fair Process When You Have No Prior Work

The honest answer when a baseline score comes back high is: usually, do very little, and certainly do not confront. On a child you have taught for a year, a high score sends you to their previous books. On a Year 7 in week two, that comparison does not exist yet — which means a single number carries far too much weight to act on alone.

Start by reading the piece yourself, slowly, the way you would moderate any borderline script. Does the vocabulary feel reachable for an eleven-year-old, or are there phrases no Year 7 in your experience would generate unprompted? Was the task done in class under your supervision, or set as a "settling-in" homework where any amount of help was possible? Context, not the percentage, decides whether this is worth a second thought.

If something genuinely does not sit right, the next step is observation, not accusation. Set a short, low-stakes piece of supervised in-class writing within the fortnight and read the two side by side. That gives you the prior-work comparison you were missing — built ethically, in conditions you controlled, on a child who has no idea they are being "checked". Most of the time the in-class piece resolves it one way or the other without a difficult conversation ever happening.

Building a Baseline Record for the Year

The deeper fix for thin context is to stop treating baseline writing as a one-off. The first half-term is when you build the very evidence base a high score later in the year will rely on. A couple of supervised, independent writing samples per child in the autumn term give you a defensible record of each student's genuine voice — the thing that makes any future detection score interpretable.

Keep your notes proportionate and factual. "Baseline 78% likelihood; heavily scaffolded primary style; in-class piece consistent — no concern" is the kind of line that protects both you and the student. It records that you read the score responsibly rather than ignoring it or over-reacting to it. This is exactly the documentation habit covered in our guide to handling a high AI likelihood score fairly, applied to the one situation where you have the least to go on.

How GradeOrbit's Detection Tool Fits a Baseline Task

GradeOrbit's AI detection tool returns a clear AI likelihood score from 0% to 100% for each piece of writing, designed to inform your professional judgment rather than replace it. You choose the depth of analysis: a lighter single-credit check for a quick read across a class set of baseline tasks, or a more thorough three-credit model when one piece genuinely warrants a closer look. Running a whole Year 7 cohort's baseline writing through the lighter model gives you a sense of the spread without committing to a deep analysis of thirty scripts you have no history for.

Crucially, GradeOrbit never stores the student work you upload. The writing is processed to produce the score and is not saved to any database — important when the children are eleven and brand new to your school. Detection is assistive: the number starts a fair process, and you, who will come to know these students far better than any tool, finish it.

Try GradeOrbit for Your Year 7 Baselines

Baseline writing is the hardest case for any detection tool, because the context that makes a score meaningful does not exist yet. GradeOrbit gives you a clear likelihood score and a 0–100% scale to inform — never decide — your judgment, while building the kind of evidence base that makes the rest of the year fairer. Visit GradeOrbit to see how the detection tool works alongside your professional judgment from the very first week of term.