When you write, you leave behind more than just the words on the page. You also leave a trail of micro‑behaviours: how quickly you type, where you pause, which phrases you rewrite three times before moving on.
Taken together, those behaviours form a kind of behavioural fingerprint. It is not about who you are as a person, but how you tend to write when you are actually doing the work.
What we mean by “keyboard biometrics”
In security research, keyboard biometrics refers to identifying a user based on typing dynamics: keystroke timings, key‑to‑key latencies, and error patterns. At TypeTrace, we care less about identity and more about the authenticity of a writing session.
Some of the signals we look at include:
- Cadence — the rhythm of keypresses over time.
- Bursts and pauses — where thought happens versus where text appears.
- Revision density — how often text is deleted and rewritten.
- Error patterns — small, consistent typos and corrections.
These are not about catching people out. They are about providing richer evidence that a document was, in fact, written by a human in a plausible way.
Human writing has a shape
If you plot keystrokes over time for a typical essay, you don’t get a flat line. You get a waveform: bursts of fast typing, slow sections where thinking dominates, and small spikes where a sentence is revised repeatedly.
A few examples:
- Students who type quickly through sections they understand well, then slow down around arguments they’re still forming.
- Analysts who iterate heavily on the introduction and conclusion, but breeze through the methodological middle.
- Writers who pause briefly before inserting a quote, then adjust punctuation and citations afterward.
These patterns are not “good” or “bad” — they are simply human. They are difficult to fake because they emerge naturally from thinking and typing over time.
How this helps in integrity reviews
When an instructor or manager is reviewing a provenance report, they’re not usually trying to determine the exact identity of the writer. They want to know:
- Was this work actually done by a person at a keyboard?
- Did the writing process match what we’d expect for this task?
- Are there sections whose behaviour looks very different from the rest?
Keyboard‑level signals make those questions easier to answer. For example:
- A 5,000‑word essay written in one continuous 12‑minute paste event is suspicious on its face.
- A report with deep, iterative editing over several days looks much more like genuine work.
- A single section with no visible edits and near‑perfect fluency might warrant a closer look.
Distinguishing genuine help from substitution
AI tools complicate this picture, but they do not erase it. Many legitimate workflows include AI assistance, whether to rephrase a paragraph, generate options, or check for grammar issues.
In a provenance trail, that looks different from wholesale substitution:
- Assistance tends to show up as smaller modifications, followed by further human edits.
- Substitution often appears as a large, clean block of text pasted in with minimal follow‑up changes.
Keyboard biometrics don’t tell you which tools were open in other tabs. They do tell you whether the resulting text looks like the outcome of thinking and typing, or like something dropped in from elsewhere.
Privacy and limits
Any time you talk about “biometrics”, privacy should be front of mind. TypeTrace treats typing behaviour as part of a document‑scoped provenance record, not as a permanent identity profile.
Concretely:
- We use behavioural signals to describe how this document was written, not to track you across the web.
- We don’t share raw keystroke data with third parties.
- You control which documents generate proofs, and which proofs you export or delete.
The goal is to protect honest writers, not to build a new surveillance surface.
Where this is heading
Over time, richer keyboard‑level signals will let us answer more nuanced questions:
- “How much of this draft represents new thinking versus reused text?”
- “Does this student’s writing pattern today match their past work?”
- “Can we spot when a report was heavily re‑authored by someone else after the original writer finished?”
Those capabilities need careful guardrails and transparent policies. But done right, they offer something that style‑based AI detection never can: evidence rooted in how work is actually done, not in how a finished document happens to look.
That is why keyboard biometrics sit at the heart of TypeTrace’s approach to integrity. They let us make fewer guesses about authorship and more grounded statements about process — exactly what students, professionals, and institutions need in an AI‑saturated world.