Should You Upload Your Draft to an AI Detector? Three Universities Have Written It Down

At some universities this is now written down, and it is written as a rule about uploading rather than a rule about cheating. Three examples, all current as of 18 September 2026:

• Leeds tells students why they "should not use AI detector or plagiarism detector websites", in a policy approved by Senate on 8 July 2026.
• Reading tells the staff who design and mark assessments that "Student work must not be uploaded into GenAI tools or AI detection tools under the current interim approach." The page is addressed to staff, not to you, which is exactly the distinction this piece is about.
• UT Austin prohibits third-party AI detection software for evaluating student work without a university contract, and says uploading a student's work into one "may be a violation of that student's copyright and intellectual property rights."

These are not all the same kind of rule and they do not all bind the same people, which is the part worth getting right before you quote one at anybody. The differences are set out below.

What none of them says is that checking your own work is dishonest. The concerns on the page are accuracy, data handling and ownership.

Leeds' Academic Integrity and Assistance Policy carries its own version stamp: "This policy was approved by Senate on 8 July 2026 for implementation from the academic year of 2026/27. This is version 1.0 of this policy which is owned by the Quality Assurance and Enhancement Team."

Inside it, in a list of what the university will do for students, is this:

The Turnitin tutorial explains how Turnitin is used in the assessment process, and why students should not use AI detector or plagiarism detector websites.

That sentence is addressed to the student side, not only to markers, which makes it unusual among the statements we have read. It also arrived with a bigger change: Leeds' generative AI hub states that this policy "replaced the three-tiered traffic light system from the start of the academic year 2026/27". If you learned your department's red-amber-green categories last year, the framework they sat in is gone.

The same policy is direct about where responsibility sits: "Students will be responsible for the process of drafting and refining the work submitted for assessment, including generating content, revising, editing and proofreading." The next sentence adds, "Students may be asked to demonstrate authorship to confirm the authenticity of their work."

Most statements of this kind stop at "don't". Leeds' guidance to staff gives its reasoning, and it is the clearest public answer we have found to a question students ask constantly — what happens to a draft you paste into a free checker:

If you suspect that material presented to you, for any purpose, has been wholly or partially generated using AI, you should not use an online AI detection tool. This is because (i) currently these tools are insufficiently accurate for the University to use them as evidence, (ii) submission of the work to such tools may represent a data security breach, (iii) the document being checked becomes available as training data for other AI tools.

Read the three reasons separately, because they fail differently.

The first is about evidence, and it is the one already covered by the universities that limit AI detector evidence. The second and third are about what happens to the file after you press the button, and they apply whether or not anyone is being accused of anything.

Two honest qualifications. This particular passage is written to staff, about other people's work, so the data-protection reasoning is strongest in that situation: the person uploading is not the person who wrote it. And the page carries no version number and no date stamp, so we are citing it as read on 18 September 2026 and nothing more.

The third reason is a claim about detector vendors' data practices, made by a university rather than by us, and it is stated in general terms rather than about any named tool. If you want to know what one specific service does with your text, its own terms are the place to look, and what a third-party checking service can and cannot tell you about storage goes through how to read them.

Reading's current assessment page opens by naming its audience — it is "For University of Reading staff who design, approve, deliver or support assessment and feedback" — and then says plainly: "Student work must not be uploaded into GenAI tools or AI detection tools under the current interim approach." Its list of practical boundaries repeats it: "Student work should not be uploaded into AI detection tools." Both sentences sit under a heading about marking and feedback, alongside "Staff should use institutionally supported tools".

So this is an instruction to markers, not a rule you can be found to have broken. It is still worth knowing, for a reason that has nothing to do with rule-breaking: your university has taken a position on putting student work into these tools, and that position is about the file, not about who is holding it. Reading's student-facing GenAI page, for what it is worth, does not mention detection at all.

Reading is also the clearest example of why the version of a policy matters as much as its wording. Three documents from the same university say three different things about AI detection:

The 2023 wording is the one that circulates, and the PDF still loads for anyone who finds the old link. It is no longer linked from the university's current AI index page. If you are going to cite your institution's position — in an appeal, an email, or an argument with a friend — the version stamp is part of the citation. Reading's own AI page asks for exactly this: "This page is regularly updated. Please revisit this page when seeking advice and avoid making local copies that may become out of date."

UT Austin's provost office frames the same subject as procurement and rights rather than integrity:

The University prohibits the use of all third-party software, including AI Detection Software (ADS, software used to flag or otherwise detect output as AI-generated), to evaluate student work or assignments unless a University contract or purchase order is in place.

The page adds that the university "does not endorse the use of AI Detection Software and does not have any central contracts or purchase orders with active AI Detection Software features", and that detection software counts as high-risk software that cannot be bought on a departmental or personal card.

The sentence students should know is this one: "UT Austin students own the intellectual property and copyright of any work they create in a course." The next sentence says that submitting that work into a third-party tool without a contract in place "may be a violation of that student's copyright and intellectual property rights."

Note who that protects and who it restricts. The copyright paragraph is about your work being submitted by someone else; it does not name who that someone is, and the guidance around it is addressed to the people who teach and mark. What the page does not do is carve out an exception for you: its own prohibition in section 3.0 is written without a subject — the use of third-party detection software to evaluate student work is prohibited without a contract, whoever is doing it. Read it as an institutional position rather than as permission. This page carries no version number or date, so it is quoted as read on 18 September 2026.

1. Look up your own institution's current page, not a screenshot of someone else's. These rules differ by university and change by academic year. Leeds' framework changed this September; Reading's position has three versions in circulation.
2. Read who the rule binds, and check the page's own audience line. Two of the three examples here are addressed to staff — Reading's says so in its first sentence — and one, Leeds', is written about students. A rule telling markers not to run student work through a detector is not a rule telling you not to check your own. Both end up in the same summaries, and the summaries are where the distinction gets lost.
3. Remember what the reading would be worth anyway. Detector vendors publish length floors below which they say their own output is unreliable, and a single paragraph sits under most of them — what five of them publish sets out the numbers.
4. Use the channel your institution actually gave you, if there is one. What you can run, and how many times, depends on what your school switched on: how to check your work before you submit.
5. If your institution has written nothing, you are deciding for yourself. The rules above are three universities' answers, not a sector standard. Of the nineteen institutional statements we have collected on AI detectors, not one is addressed to a student checking their own draft — they are all about what staff may run and what a panel may rely on.

Is it academic misconduct to run my own essay through a detector? None of the documents quoted here says that. Leeds' policy tells students not to use those sites and explains why in its Turnitin tutorial; Reading frames it as an interim approach to uploading; UT Austin frames it as procurement and copyright. If your own institution's rules say something stronger, its rules are the ones that apply to you.

My university has not written anything about this. What then? Then there is no rule to break, and the practical questions remain: what the site does with your file, and whether a reading on a partial draft means anything. Both are worth answering before you paste.

Does this apply to grammar and rewriting tools too? That is a separate question with a separate answer, and the answers differ sharply between institutions. Leeds' policy is worth reading in full on this point, because it turns on two definitions and two rules. It defines "technologies" as "digital tools, including generative artificial intelligence (GenAI) and translation aids, that are paid, unpaid or provided by the University", and "paid services" as "an external person and/or company paid for by the student". The rules that follow are: "Students are permitted to use technologies to prepare work for assessment if restrictions are not specified within the assessment brief", and "Students are not permitted to use paid services to prepare work for assessment." Which category a given tool falls into at your own university, under your own assessment brief, is not something anyone else can decide for you.

Where do the schools that restrict detector evidence fit in? They overlap but are not the same group. A rule about what evidence a misconduct panel may rely on is not a rule about who may upload a file. The 19 institutions that limit AI detector use or evidence sorts the statements by what kind of rule each one is.

• University of Leeds, Academic Integrity and Assistance Policy (version 1.0, approved by Senate 8 July 2026, for 2026/27), and the generative AI hub noting that it replaced the traffic light system.
• University of Leeds, guidance to staff on the use of artificial intelligence. No version number or date on the page; read 18 September 2026.
• University of Reading, AI in teaching, learning and assessment and assessment and feedback guidance.
• University of Reading, GenAI tools and assessment, V1.4 (June 2025), and the August 2026 policy statement (version 1.0, approval date 12/08/26).
• UT Austin Office of the Provost, AI detection software guidance. No version number or date on the page; read 18 September 2026.