UnTXT - Product Factsheet

Last updated: October 2026 · For informational use · All details reflect the publicly available product

Basic Information

What UnTXT Does

UnTXT is an online AI pattern checker from Sataklela OÜ that shows people which parts of their writing sound machine-made and why, so they can make it their own again.

You paste in an essay, a report or an application. The passages built from the habits of phrasing and structure that machine writing overuses come back marked FLAGGED, in grey, each with the reasoning written out. The passages that sound like one particular person come back marked AUTHENTIC, in gold. A mark describes how a passage reads and never who wrote it, because people follow those patterns too, most of all when they write formally or in a second language.

There is no percentage anywhere in the product, and nothing is rewritten. The reading is done by two frontier models, Claude Opus 5.5 and GPT-6 Sol, which go through the text separately and argue each passage out, with Claude Sonnet 5 settling whatever they cannot agree on. The reader can see what each of them said.

Target Users

Confirmed Public Features

• Two marks per passage, AUTHENTIC or FLAGGED, each with its reasoning; eleven pattern checks (SPIKE, SYNTAX, FORMAL, HUMAN, TEMPLATE, FRONTED, FORENSIC, RHETORIC, PUFF, FORMULA, LEXICAL) still run in the browser extension's signed-out trial
• One result for every role: the full analysis is the only reading the web app runs
• Run overnight: the same analysis queued between 5 pm and midnight local time for about 40% fewer credits, back within eight hours and never sooner than three, with an email when it lands
• Analysis history with date, word count, flag count, and credit cost per entry; all stored text encrypted at rest with AES-256-GCM
• PDF report export, A4 formatted, for the Teacher and Professional roles
• CVT credit gifting: transfer unused credits to any registered user (one-hop only)
• Three roles (Student, Educator, Professional), which set the plan and wording; the role does not change how a document is read
• Dark mode with per-user persistence
• 7-step onboarding quickstart for new users
• Multi-currency pricing: 14 currencies: GBP, USD, EUR, INR, AUD, CAD, AED, SGD, BHD, CNY, KRW, CHF, NGN, KES
• Google OAuth sign-in
• Marked passages carry a note rather than an editor: clicking one says which verdict it holds and what was noticed about it, and nothing in the app offers to rewrite the sentence for you
• Invite & Earn referral program: personal referral link per user, 10 CVT to the friend on signup, 10 CVT to the referrer once the friend returns on another day, 90 CVT to the referrer on the friend's first subscription
• Cohort: the people referred to or from, or exchanged credit with; free direct send of a check you have already run into a cohort member's history
• The full analysis: two frontier models, Claude Opus 5.5 and GPT-6 Sol, argue each passage out over several rounds, with Claude Sonnet 5 settling ties; the reader sees each model's reasoning; credit is charged on actual usage
• Interface language: English only. Fifty machine-translated languages shipped between 4.5.0 and 4.68.0, and were removed in 4.92.0 because enough of them came back wrong that showing them was worse than showing nothing; text in any language can still be submitted, with the checker calibrated on English
• The full analysis is the only pass the web app runs: pressing Analyse sends the text straight to the debate, with no heuristic pass ahead of it, and an account that cannot run it is told so before anything is analysed or charged
• File upload on the blank canvas: PDF, Word and plain text, identified by file contents rather than by name or declared type
• Account verification at sign-up: institutional address for students and teachers, company registration and tax reference for organisations; accounts that cannot verify pay per analysis run instead
• Organisation accounts for schools, tutoring agencies, student societies, teacher training programmes and enterprises: per-seat billing, pooled credit allocated by an account manager, co-branded portal; tutoring firms verify by phone, card and email rather than a registered business name, enterprise needs VAT/legal-entity details and a 5-seat minimum up front
• Tiered documentation site at untxt.uk/docs: open pages for anyone, deeper reference behind a free account
• Monthly subscription for individuals: role-based tax-inclusive plans in 14 currencies with a monthly CVT allowance; overage metered and billed to the saved card at month end with a 1× fair-usage cap; fully embedded on-site checkout (Stripe Payment Element, no redirect)
• Account & Billing page (untxt.uk/account): separate from the writing tool; plan and usage, projected month-end charge, card update, invoice history, cancel/resume
• Documents: saved work with derived titles, opened into the editor by clicking the row; recent documents and project folders in a drawer down the left
• Editable, versioned recall: a reopened document is fully editable and autosaves as versions rather than overwriting, at no credit cost, with a colour-coded word diff between any two versions
• What's New: a panel listing what shipped in each release, with unread entries marked per account
• REST API and developer portal (untxt.uk/developer): self-serve API keys, an endpoint that runs the analysis, and balance and usage endpoints, billed from a prepaid credit reserve; reference at untxt.uk/docs/api
• Institutional address verification against roughly 10,600 university domains worldwide as well as the .edu, .ac.xx and .edu.xx patterns, so a genuine institution on an ordinary national domain is recognised
• Partner programme for sales representatives, affiliates and content creators: tracked links, a self-serve dashboard at untxt.uk/partners, commission tiered by the plan a referral subscribes to, and payouts made through Wise in the partner's own currency
• Browser extension for Chrome and Firefox, built from one shared source tree: checks selected text anywhere on the web from a side panel; signed in it runs the full analysis, and without an account it gives five trial checks on the pattern modules alone

Pricing Model

What UnTXT Does Not Do

• It does not store submitted text in plain text: everything you have run is encrypted at rest with AES-256-GCM before it reaches the database
• It does not train on user-submitted text
• It will not check your work for plagiarism, since it never compares anything against a library of existing documents
• It does not let an organisation read its members' writing. An organisation sees credit spent, checks run and last-active date, and nothing further; the text itself stays encrypted against the member's own account
• It does not take a whole archive in one upload. The developer API takes up to 50 documents per request, run straight away or queued overnight, so a bigger archive goes in as several batches
• It does not plug into a learning management system, and we ship no connector for one
• It is not forensic evidence you could take to a hearing, as the results are informational and educational only
• It does not publish an accuracy figure, and does not compete on one. Every such number in this market is produced against a benchmark chosen by whoever quotes it, ours included, so we give the reasoning behind each marked passage and leave the weighing to the reader
• It does not show a percentage score anywhere - not in the app, not on a shared link, not in the wording attached to a marked passage
• It will not rewrite your work for you. There is no button that takes a draft and hands back a cleaner one, and there is not going to be

Company Information