Students, teachers and people writing at work all reach for this differently, so here is what each of them actually does with it.
If you are worried that an essay will be wrongly flagged, paste the draft in and read what comes back. Sentences that look machine written are marked where they sit, and each one is coloured by the kind of pattern that it matched, so what you end up with is a specific line to work on rather than a number that you cannot act on. Fix the lines and run it again. You go into deadline day knowing, and not guessing.
Plenty of people use AI to brainstorm, to tidy up wording or to work out a structure, and that is now ordinary rather than suspect. What the highlights give you is the boundary, so that you can see where one ends. Here is the part that a machine shaped, and here is the part that was yours from the start, which means you can rewrite the first in your own words while everything else stays as you had it.
Longer work is easier to check when you take it a section at a time. An EPQ or a dissertation rarely fits into a single pass, so run each chapter on its own and every result stays in Documents under a title of its own, which is what makes the pile navigable three weeks later. Come deadline week you are not hunting through drafts, because the one that you already checked is the one still sitting there.
You get the whole analysis before you change anything. You can see where the marks fall, how they cluster, and what the reasoning says about each one. Read a few of your own drafts that way and the habits start to become obvious, which is the point: the tool is less useful as a filter than as something that teaches you what your writing gives away.
You do not need to be technical, because the output reads plainly enough. Paste in a submission and the passages worth a second look come back marked, each carrying the reasoning that put it there, which is usually all that you need in order to decide whether a conversation is warranted. Actually that is the whole intent of it. It is a faster second opinion and not a verdict, and it does not ask you to keep pace with whatever tool your students happened to find this month.
Any analysis that you have finished becomes a formatted A4 PDF in seconds. Records, integrity reviews, a meeting, or a conversation that you would rather de escalate: the report is annotated clearly enough that you can put it in front of somebody who was not in the room. It downloads when you want it, and nothing about it expires at the moment that the analysis finishes.
Whether it is a handful of essays or a whole cohort, an Educator plan is sized against real marking loads at 1,300 CVT a month. Unused allowance banks to the end of your academic year instead of vanishing at the end of the month, so a quiet term or a long holiday costs you nothing. If a month runs long, the excess is metered at the same rate credit costs to buy, and your marking is never simply cut off partway through.
Clients, colleagues and employers are getting better at noticing AI written content, and some of them will say so. Paste a report, a proposal or any business document in, and the passages that read as machine generated surface early enough that you can do something about them, rather than after the document has reached whoever you were hoping to impress.
CVs, covering letters and written applications can be read more carefully than they usually are. Sections that look heavily AI assisted get marked, and that frees you to weigh the things that genuinely predict a good hire, which are authenticity, the way somebody communicates, and whether the substance underneath actually holds up when you read it twice.
Agencies, editors, recruiters and content teams run it over submitted copy before anything goes live. You can pick out the passages that read artificially, and point at them in a conversation, which lets you hold one standard across work arriving from a lot of different hands. It scales, and it does not need a meeting.
Prompts, chatbots and detection jargon are not things that you need to follow. The tool is built for people whose job is not AI, and what it offers is a practical way of staying current with machine written text. You are not obliged to become expert in a subject that you never chose and would rather not think about.
Share your referral link with classmates or with a study group. Once somebody signs up through it and runs a first analysis, bonus CVT lands on both sides, and nothing has to be bought for that to happen, which is what makes it worth passing on rather than hoarding. It stretches your own free allocation and it points other people at something useful.
Pass your link on to other educators when it suits you. When a colleague signs up and runs a first analysis, credit lands on both sides. Anyone reviewing several classes tends to feel the benefit of the extra CVT quickly, and the link sits in the account menu whenever you want it.
Teams that review documents together can circulate the link internally. Every new sign up that completes an analysis earns credit on both sides, so a team builds its checking capacity gradually and nobody has to raise a purchase order. That means no approval and no budget line to argue for.
UnTXT sits comfortably inside the generative AI policy that Loughborough University applies to assessments. Because it detects rather than generates, students can use it to audit a draft before submission and show good scholarship in doing so. Best practice is to acknowledge that use in your assessment statement.
UCL describes its approach to AI as supportive rather than prohibitive, and UnTXT fits that comfortably. Students are asked to acknowledge GenAI use and to make sure the final submission is substantially their own work. Auditing a draft is how you check the second of those before anyone else does. Acknowledge that use when you write your disclosures.
Imperial College treats assessed work created with AI, where that was not explicitly authorised, as potentially contract cheating. UnTXT works alongside that requirement rather than against it: patterns are identified in the draft, you rewrite them, and the substantial personal contribution the college asks for stays intact. Acknowledge that use in the AI tools statement when you submit.
Russell Group principles apply across member institutions, and that includes Loughborough, UCL, Imperial, Oxford and Cambridge, with an emphasis on transparency alongside substantial personal contribution in assessed work. Auditing your own drafts is what serves both of those principles at once. Even so, the detail varies from one institution to the next, so read the generative AI policy that your own university publishes rather than assuming that the group position covers you.
Whether UnTXT fits depends on the policy that your institution has actually written down. Broadly it supports transparent scholarly practice, because students can audit their own work and educators can hold assessment integrity without having to guess at what a passage is. Read the guidelines that bind you before you rely on any of it.
Academic Weapon runs on every analysis in the web app now, rather than waiting to be asked for it, and your text goes to two frontier models — Claude Opus 5 and GPT-5.6 Sol — which argue over it passage by passage, with only what both of them land on getting marked and a third model, Claude Sonnet 5, reading whatever they cannot settle between them. So the answer carries more weight than any single reading would, including ours.
A run costs roughly 89 CVT against 1 CVT for a preliminary check, and the estimate appears before you confirm rather than after. Results stay probabilistic, as they always have. Agreement between two frontier models with a third breaking ties is a strong signal, though it is not a legal finding, and Sataklela OÜ accepts no liability for decisions taken on the strength of it.