Separate the dimensions
Ideas, structure, prose, facts, and review are different kinds of contribution. The profile describes each on its own terms.
The Provenance Profile · An open framework for human–AI authorship
The Provenance Profile is a compact disclosure that shows how human intention, AI assistance, and human judgment combined in a finished work—without reducing authorship to a misleading percentage.
A reader does not merely want to know whether AI was present. They want to understand what, exactly, they are receiving from the named author.
Did the author originate the ideas? Determine the argument’s sequence? Choose the wording? Did AI edit a dictation, propose a framing, or generate the first draft? And, above all, did the author examine and adopt the finished work? A binary disclosure cannot answer those questions.
The logic is not new. Scholarly publishing already separates contribution into typed roles—the CRediT taxonomy—and leading journals now require disclosure of AI use while reserving authorship, and its responsibilities, for humans. The Provenance Profile extends that discipline to everyday professional writing.
The profile is built to answer the reader’s real questions: it separates kinds of contribution, names each in fixed language, and keeps responsibility with the person who publishes.
Ideas, structure, prose, facts, and review are different kinds of contribution. The profile describes each on its own terms.
Controlled vocabulary makes one profile comparable with the next while remaining honest about what cannot be precisely measured.
Origin and accountability are distinct. AI may contribute an idea; the person who publishes it must still evaluate and own its use.
The goal is not to maximize human provenance. It is to maximize the value of the work while making its provenance legible. Foundational principle
“78% human” looks objective, but unless something was counted it is only an impression wearing the costume of a measurement. Different models—or the same model on different days—may assign different figures.
Percentages are appropriate for verifiable evidence such as retained wording, sentence traceability, or sequence overlap. They should not be used as a single proxy for ideas, authorship, voice, and responsibility.
Measurement is also what keeps a profile honest. A profile is a claim about the drafting record—dictation transcripts, prompts, successive drafts—and a challenged label is settled by that record, not by assertion. Labels carry weight precisely because they can, in principle, be checked.
Each dimension asks one distinct question and uses labels that are meant to explain themselves when attached to a memo, article, report, or book.
Choose the narrowest label that truthfully reflects the process. Add a short clarification only when the standard label would omit something material.
A profile describes the dominant mode of the whole work. Note material exceptions in the Context line; a long work may carry per-section profiles instead of a single misleading average.
Who introduced the work’s substantive claims, insights, and conclusions?
“Author-adopted” means the author evaluated the AI-originated idea and deliberately made it part of the work.
Who determined the sequence, framing, and argumentative architecture?
“Author-defined” covers structure substantially supplied by the author—whether laid out in advance or shaped through the author’s iterative direction.
How did the final sentences come into being, and how substantially did AI shape them?
One spectrum covers both production and reshaping: each label names who wrote the sentences and who reworked them. Prose describes production—not ownership of the underlying idea or the author’s approval of the result.
Who supplied the evidence, data, and checkable claims—and were they verified?
A conditional dimension: include it whenever the work makes checkable claims; omit it for opinion, narrative, or purely personal notes. “Not independently verified” is an honest label, not a failing—use it rather than implying verification that never happened.
What level of human scrutiny and deliberate acceptance did the final work receive?
Review is the profile’s accountability axis. “Reviewed for substance” means claims and conclusions were checked but wording was not scrutinized—appropriate for quick internal notes, not outbound work. Factual verification is claimed on the Facts line, never implied here.
The labels do not form a ladder from “less human” to “more human.” Read them horizontally as a concise history of the work.
These reveal where the argument came from and who shaped the way it unfolds.
It tells you how literally to treat the phrasing as the author’s unaided language.
These show whether claims were verified and whether the author deliberately accepted the work presented.
There is no ideal combination. The right profile is the one that accurately represents how the work was made.
The author speaks the argument in sequence. AI compresses repetition and improves transitions.
The author stands behind the complete memo and its final wording. No Facts line: a personal memo makes no checkable claims.
AI proposes a useful distinction and alternative outline; the author tests, reshapes, and adopts both.
The author adopted the analysis after review and accepts responsibility for its claims.
The author specifies the audience, objective, evidence, and constraints; AI produces the first complete draft.
The use of AI increased execution speed; it did not transfer the author’s responsibility.
A profile is a disclosure, not a grade. More human contribution does not automatically mean better work; more AI contribution does not mean worse work.
How human and AI contributions combined in the finished work.
The value, authenticity, truth, or intellectual worth of the result.
Place the profile near the title, byline, or end note. Use one label per dimension, then close with a direct statement of accountability.
Label the dominant mode of the work and note material exceptions in the Context line. Include the Facts line whenever the work makes checkable claims—and claim verification only when it has actually occurred.
For bylines, footers, and signatures, use the short form below. Its codes map one-to-one to the full labels, so a reader can always expand it.
Provenance Profile Ideas: [Author-originated / Jointly developed / AI-originated, author-adopted] Structure: [Author-defined / Jointly developed / AI-proposed, author-adopted] Prose: [Author-written / Dictated, AI-refined / Jointly written / AI-drafted, author-revised] Facts (when the work makes checkable claims): [Author-supplied / Jointly sourced / AI-sourced, author-verified / Not independently verified] Review: [Reviewed for substance / Fully reviewed / Fully reviewed and revised] Context (optional): [One sentence covering material exceptions or contributions the labels do not capture.] Accountability: I reviewed the complete work and stand behind its [arguments, conclusions, and final wording].
Keep the selected labels and delete the alternatives. Adjust the accountability line only to make it more accurate, never stronger than the review performed.
PP/0.2 · Ideas: author · Structure: author · Prose: dictated+ai · Review: full+rev
Codes per dimension — Ideas & Structure: author · joint · ai-adopted. Prose: author · dictated+ai · joint · ai-drafted. Facts: author · joint · ai-verified · unverified. Review: substance · full · full+rev. Omit Facts when the work makes no checkable claims.
The best profile is not the “highest” one. It is the most honest one.