Sataklela
Case StudyDocument Intelligence

UnTXT

Be Original Again

Live

Origins

UnTXT grew directly out of the work behind CVaaS. As AI-generated applications became ubiquitous and employers began rejecting mass submissions on sight, the root problem became clearer: people had lost the ability, or the confidence, to express themselves in their own voice. The same pattern was showing up in academic coursework, in professional communications, and in the writing of people who had never touched an AI tool but were being flagged anyway.

Existing detection tools made the problem worse. A single percentage score or a binary verdict doesn't tell a writer what to fix or why it was flagged. It just creates anxiety. UnTXT was built to do the opposite: surface specific patterns, explain them, and give writers something they could actually act on.


The Solution

UnTXT analyses submitted text and highlights passages that match patterns associated with AI-generated writing. Rather than producing a score or a verdict, it returns colour-coded, passage-level annotations showing which specific sentences triggered which detection modules and why.

The design premise is that UnTXT is a tool for writers, not a tribunal. Its purpose is to surface patterns so that authors can make informed decisions about revision, not to make absolute determinations about authorship. Early user testing confirmed this: users wanted more detail so they could learn from the output, not just pass a check.


In Use

Neurodivergent student, UK university

The tutorial is super straightforward and clear, the website isn’t overwhelming… I know this will help a lot of neurodivergent students who struggle with unclear instructions and what is ‘expected’.

Postgraduate student, translated coursework

A coursemate had written his work in Chinese, used ChatGPT to translate it, and was being warned by professors for AI patterns in the output. He switched to UnTXT to identify and revise the flagged passages. No issues reported after that.

Non-technical adult user, verification use case

Standard mode, a traffic-light grading system, was introduced after users outside academia asked for a simpler way to verify texts and emails they received, without needing to parse detailed linguistic annotations.


Detection Approach

UnTXT uses 11 named detection modules. Each module identifies a different class of signal (structural, lexical, stylistic, or rhetorical) that is associated with AI-generated writing patterns. A passage may be flagged by one or more modules; each flag is labelled and colour-coded.

These are signal categories, not absolute indicators. A flag does not confirm AI generation; it highlights a pattern that warrants attention.

SPIKE

Identifies abrupt changes in language register or complexity

SYNTAX

Flags structural sentence patterns common in AI-generated text

FORMAL

Detects overly formal or clinical phrasing atypical for the author

HUMAN

Assesses absence of natural human writing markers

TEMPLATE

Identifies generic structural patterns and filler phrases

FRONTED

Detects fronted adverbials and clause structures characteristic of AI output

FORENSIC

Applies deeper lexical and stylistic forensic analysis

RHETORIC

Identifies rhetorical structures frequently used by language models

PUFF

Flags meaningless amplifying language and superlatives

FORMULA

Detects formulaic construction patterns across paragraphs

LEXICAL

Analyses vocabulary choice patterns associated with AI generation


Key Features

  • Passage-level colour-coded annotations; not a percentage score
  • 11 named detection modules covering structural, lexical, and rhetorical signals
  • Academic mode optimised for essay submissions
  • Freemium model: first 5,000 words analysed free, no card required
  • Role-based entry for students, teachers, and professionals

Target Users

Students

Review drafts before submission to identify AI-pattern passages and revise for authentic voice

Teachers

Assess submitted work and open conversation with students about writing authenticity

Professionals

Review written deliverables (reports, proposals, communications) for AI writing signals before publication


Pricing

LiveFree to start

The first 5,000 words are analysed free with no card required. Premium tiers are available for higher usage. Sign-in unlocks bonus word allocation.


Screenshots

The UnTXT interface requires sign-in to access. Screenshots are available on request; contact us directly.

Submission interface
Screenshot available on request
Annotated output view
Screenshot available on request