Copyrightable does not guess and does not count words. It assesses the kind of human creative control you exercised, measured against the standard the US Copyright Office actually applies. This page is the public, versioned specification of that assessment.
US copyright protects work that is the product of human authorship. For AI-assisted work, the Copyright Office asks whether a human exercised sufficient control over the expressive elements of the work - through selection, arrangement, or modification. There is deliberately no quantitative threshold; the Office decides case by case.
Critically, the Office has been explicit that prompts alone are not enough, no matter how detailed. A prompt conveys an idea; copyright attaches to expression and to the human control over it. So our assessment rewards control over expression, and refuses to reward effort or volume.
Grounded in: USCO Registration Guidance (2023); Copyright and Artificial Intelligence, Part 2: Copyrightability (2025); Thaler v. Perlmutter (2025); Zarya of the Dawn registration decision.
You chose this output over others, or chose what to keep and discard. Selecting expressive material is a recognized authorial act.
You ordered, structured, or composed the material - sequencing, layout, what goes where. Creative arrangement can itself constitute authorship.
You altered AI-generated material beyond the trivial - rewrites, refinements, edits that change the expression.
You drove repeated rounds of revision toward a result, evidencing control over the final expression.
You steered the AI away from its output toward a different expressive result - a strong signal of creative control.
These are deliberately not rewarded, because the law does not treat them as authorship:
We report an evidentiary profile, never a percentage - a percentage would imply a precision the legal standard explicitly rejects. Each filing gets a qualitative tier plus the underlying evidence: counts of your selection, arrangement, and modification decisions, iterations, redirections, and how much AI output you modified rather than accepted unchanged.
Substantial, repeated human control over expressive elements - modification, selection or arrangement, and iteration. The kind of creative control the USCO test looks for.
Some human control evidenced - modification, selection, or iteration present but limited. May support authorship for the human-controlled portions.
Little evidence of human control over expressive elements. Largely unmodified AI output or bare directives.
This is Methodology v0.1. Every certificate records the methodology version it was assessed under, and stays valid under that version. As the law and our methods evolve, new versions can re-assess prior filings from the same immutable evidence, so the change is auditable rather than silent. The exact weights and thresholds are versioned internally and not published; the framework, factors, and tiers above are.
What v0.1 can and cannot see: modification, iteration, and redirection are observed reliably from the capture chain. Selection and arrangement are recorded only where the tool surfaces them (choosing among outputs, reordering); we do not infer them when they are not evidenced. The profile reports what was observed, never what was assumed. When no control signals are captured at all, we say so plainly rather than assigning a tier as if control were measured and found absent - the two are different, and we keep them distinct.
Where the signals come from: the capture surfaces record what happened on your machine as you worked. The score is only as honest as the record beneath it, which is the point of capturing at the moment of creation rather than reconstructing after the fact. The assessment normalizes by the human actions in a session, so tool-call noise and the sheer size of an AI response never move the result.
Provenance vs. authorship: these are two separate facts and we never blend them. The tier above says how strong the human control was. Provenance says how trustworthy the capture of it is. Live capture (the hook and hosted connector) records on your machine as the work happens. An imported transcript is supplied after the fact, so it is inherently easier to stage and carries weaker provenance. We label which one a filing used rather than hiding the difference. Provenance never changes the authorship tier, it sits beside it. See what this proves, and what it doesn't.
For the limits of what this evidence does and does not establish, read what this proves, and what it doesn't. Questions about how your work was assessed? Try it free and see your own profile, or see pricing.