A guide for schools and universities
AI writing tools and academic integrity
Your students are already using these tools. The question is no longer whether to allow them, but what a workable policy looks like, what detection can honestly tell you, and what to say to a student who asks where the line is.
We make one of these tools, so read this knowing that. Most of what follows would be just as true if you never used ours — and we have tried to write it that way.
What AI detection can and cannot tell you
Detection has been adopted widely and quickly — by early 2026, more than sixty per cent of colleges and universities were using some form of it. It is worth being precise about what a score from one of these systems actually represents.
Detectors do not recognise AI text. They measure statistical properties of writing — how predictable the next word is, how much sentence length varies, how evenly vocabulary is distributed — and report how closely a passage resembles the patterns machine-generated text tends to have. That is a probability, and it is a probability about a style rather than about an event.
Which means the false positives are not randomly distributed. They land on students writing in a second language, on students who write plainly and formally, and on anyone who writes the way academic writing is explicitly taught to be written — short sentences, consistent register, restrained vocabulary. The students most likely to be wrongly flagged are often the ones with the least standing to contest it.
A detector score is an input, not a finding. Used alongside draft history, alongside a conversation, alongside whether the submission sounds like the last three from the same student, it usefully prompts a look. Used on its own as grounds for a charge, it will eventually be wrong about somebody who did nothing.
What a policy that survives contact with reality looks like
The policies that fail are the ones that draw the line at the technology. “No AI tools” is unenforceable and already untrue — spellcheck, autocorrect, grammar checking and predictive text are all AI, and have been for years. The policies that work draw the line at authorship instead.
Ask who produced the ideas, not which software was open
If the argument, the structure and the judgement are the student’s, tools that improve the expression are ordinary academic practice. If they are not, no amount of disclosure makes the submission theirs.
Say it per assessment, not per institution
A reflective essay and a lab report have different reasons for existing. A blanket rule forces departments to either ignore it or enforce something that makes no sense for their discipline.
Require disclosure and make it safe to give
Students conceal use when disclosure is punished. A line on the cover sheet — what was used and for what — produces far more honesty than a prohibition, and it is a teachable moment rather than a hearing.
Assess process where it matters
Draft history, an annotated bibliography, a five-minute conversation about the argument. These are harder to fake than any text is to detect, and they were good pedagogy before any of this arrived.
What to tell students
The question students actually have is narrower than the one policies usually answer. They are not asking whether AI is permitted in principle. They are asking whether the specific thing they are about to do is cheating, at eleven at night, with the submission due at nine.
Most of it can be answered with one distinction: fixing how you said something is ordinary; obtaining what to say is not. Grammar, clarity, a sentence rewritten because it came out badly — these have never been controversial. An essay produced from a title is a different act, and students know it is, which is why they ask.
The one that genuinely confuses people is paraphrasing. The test is not how many words were changed — it is whether the structure of the thought is still somebody else's. Swapping synonyms while keeping the sequence of points is a close paraphrase and needs a citation regardless of how different it looks. Our student page puts that in the words a student would use, and states plainly what we will not help with.
Where ABCcount fits, if at all
There is no essay generator, deliberately. Every mode takes text a student has already written and changes how it reads — give it nothing and it produces nothing. That is the line the product is built around rather than a feature we have not got to yet.
Text is processed and then removed. Neither of the two providers we route to trains on it, and one retains nothing at all; the trust and security page names them and states where our guarantees stop.
Institution-wide access is seat-based and billed once rather than by several thousand individual cards, and it can be a single department rather than a campus. If you want to talk it through — or if you disagree with anything above and want to tell us why — we are at sales@abccount.com.