Tellmark highlights AI-typical patterns in prose — named, counted and explainable. It is a linter for writing, not a detector: every flag maps to a published catalogue entry, and the same text always scores the same. What you do with the evidence is an editorial decision, and it stays yours.
Tellmark reports observable patterns. It makes no claim about who or what wrote this text.
Every rule is public: read the catalogue, annotated — one article per tell, each assayed by the tool before publication.
specimenen-GB · plain text
0 words · 150 needed for a score
Runs in your browser in under a second. Nothing is uploaded.
Your Tellmark account
Saved reports
Assay
–TDI · tells / 1,000 words
051220+
Tellmark reports observable patterns. It makes no claim about who or what wrote this text.
Saving sends this text to your account, encrypted at rest. Analysis itself never leaves your browser.
Paragraph heat
Annotated galley
Findings
Two definitions
A tell is an observable, countable pattern that occurs at elevated frequency in text produced by large language models, defined by a published catalogue entry.
The Tell Density Index (TDI) is Tellmark's deterministic score: weighted tell instances per 1,000 words, banded green (below 5), amber (5–11.9) and red (12+). Same text, same catalogue version, same score — always.
Questions, answered
What does Tellmark do?
It marks tells — patterns that occur far more often in machine-written text than in unaided human prose — directly in your draft, each with a name, a count and a published rule, then scores their overall density.
Is it an AI detector?
No, and deliberately so. Detectors claim to know authorship and their false accusations have hurt honest writers. Tellmark reports what was found and where; the judgement stays with you.
What is a tell?
An observable, countable pattern defined by an entry in our public catalogue: a word like a famous verb, a construction like the negate-then-elevate template, or a statistic like uniform sentence rhythm.
What is the TDI?
The Tell Density Index: weighted tell instances per 1,000 words. Below 5 is green, 5 to 11.9 amber, 12 and above red. It is deterministic — the same text and catalogue version always score the same.
My text is human-written but scored amber. Is that a mistake?
No. The score says the prose contains AI-typical patterns at a noticeable rate, which readers now recognise regardless of who wrote it. Treat the highlights as an edit list, not an accusation.
Where does my text go?
Nowhere, by default. The assay runs entirely in your browser. Only if you sign in and press Save is the text sent to your account, where it is stored encrypted.
How often does the catalogue change?
Roughly quarterly, as the models drift. Every change ships with a version bump and a changelog note, and each report states the catalogue version it was scored against.
Catalogue updates by email, nothing else. When the pattern catalogue changes version, we send one note saying what changed and why.