Proof-of-Authorship Software: How Universities Should Evaluate Verifiable Writing Records

In brief: Software can preserve evidence about how a document was written, associate a record with an account or attesting person, and in some products make the sealed artifact independently checkable. But proof of an unchanged document is not the same as proof of a person's physical identity or proof that an academic policy was followed. Universities evaluating "proof-of-authorship software" should choose exactly which propositions need evidence, then insist that vendors demonstrate the limits as clearly as their strengths.

Disclosure: AuthMarks publishes this guide and provides writing-process evidence and sealed records. This is a buyer's framework, not a certified ranking or independent comparison of commercial products.

"Prove I wrote it" can mean five different things

The phrase sounds simple until an academic-integrity reviewer asks what a record actually establishes. Separate the following claims:

1. Document integrity: this file is unchanged from the version signed or sealed earlier.
2. Account attribution: the recorded actions or attestation are associated with a particular account or named user.
3. Observed writing activity: the system saw particular typed, pasted, edited or suggested material in a defined capture window.
4. Personal authorship: the physical person who submitted the work also performed the relevant creative work.
5. Academic compliance: that person's use of AI, assistance, references and collaborators complied with the course's rules.

Different tools address different parts of this chain. A verification result for item one cannot by itself settle items four or five. A trustworthy vendor should explain that plainly.

Three software approaches a buyer may encounter

Final-text classification

AI-text detectors analyze a finished submission and estimate whether parts resemble model-generated content. Platforms such as GPTZero and Copyleaks offer such services. The result is not direct observation of the writing process and does not establish personal authorship.

This approach can still be relevant to institutions whose policies allow it as one preliminary signal. It is not equivalent to a signed writing record.

Writing replay and process reports

Grammarly Authorship, GPTZero Writing Replay Anywhere, Turnitin Clarity and Cursive by Blackboard publicly describe forms of writing-process evidence. Their supported editors, source attribution, capture methods, sharing and administrative capabilities differ, and several have functions beyond replay.

These tools can make the observed development of a document much easier to inspect than the finished file alone. An instructor still needs to know what was not observed and who can access the report.

Attested, sealed writing records

AuthMarks documents a process in which a writer explicitly enables recording for a document, the system labels observed writing, unknown/untracked areas remain visible, a named person reviews and attests, and a record is sealed. Its public verification service allows a recipient to check whether a sealed artifact matches the record, without requiring access to private writing evidence. AuthMarks documentation · Security and verification details.

This is a strong fit if the procurement specification asks for both a process record and a separately checkable integrity claim. It does not establish that everyone who interacted with the document was the named account holder or that every part of the work was recorded.

A four-layer "proof" model for procurement

Layer	Example question	Evidence a product may provide	Residual uncertainty
Capture	What happened during the observed writing session?	Changes, paste events, known gaps, timestamps	Anything outside the capture window
Attribution	Who is associated with the activity or review?	Signed-in account, reviewer identity, attestation	Physical person and unauthenticated collaboration
Integrity	Is the retained record identical to what was sealed?	Signature or hash verification over a specified artifact	Truthfulness of every underlying assertion
Governance	How may this evidence be used?	Access logs, permissions, retention, review procedure	Whether an institutional rule was broken

The layers should not be collapsed into a single percentage. For example, an artifact can be perfectly intact even if the account was shared, the student legitimately pasted a quotation or the work occurred partly before capture began.

What should happen when writing was not recorded?

A mature product should never silently convert unobserved text into "human typed." Ask the vendor to demonstrate a document that contains text written before capture was switched on. The report should distinguish the observed segment from a region with missing or uncertain provenance, if the product is designed to provide that distinction.

An unknown area is not automatically an AI-generated area. It is simply an area the product cannot confidently account for. This matters for students who wrote offline, imported legitimate drafts, used approved assistive technology or experienced recording problems.

For buyers, a transparent account of gaps may be more valuable than a simplified dashboard showing 100% confidence.

The difference between a checkable seal and a vendor-hosted report

A vendor-hosted report can be useful: staff sign in, review a timeline and discuss observations with a student. A separately checkable sealed file adds a different question: Can a recipient determine that the artifact still matches the version that was signed, even after it leaves the original workflow?

When vendors advertise a signature, hash, certificate, notarization or C2PA compatibility, request a live demonstration using an ordinary file. Have the vendor explain:

• What specific bytes or record fields are covered by the check?
• What changes make the verification fail?
• Does the verifier work after the student leaves the platform?
• Can a recipient verify the public integrity claim without seeing private drafting details?
• What additional permissions are required to inspect detailed process evidence?
• What does a positive result not establish?

A signature's technical validity is separate from the human truth of the statements in a record. This is a standard limitation of evidentiary records, not a weakness unique to one product.

Six demonstrations to request before purchase

Demo 1 — Normal draft: a student writes, deletes, revises and adds a sourced quotation. Show what the instructor actually receives.

Demo 2 — Unobserved segment: write part of the document before capture or with recording stopped. Show whether the record identifies uncertainty honestly.

Demo 3 — Named review: show exactly what the person attests, how consent is presented and whether the writer can inspect the record before sharing.

Demo 4 — Export and check: if a vendor offers a sealed artifact, export it and independently validate an unchanged copy. Then make a benign modification to a test copy and repeat the check.

Demo 5 — Instructor access: open the record through the normal LMS or submission route, not through the seller's administrator account alone.

Demo 6 — End-of-term retention: demonstrate what the institution still has when a student withdraws, a course ends, a share link changes or the institution terminates a licence, subject to applicable legal obligations.

Use a worksheet with three verdicts for each requirement: Demonstrated, Not demonstrated, Not applicable. Never automatically replace "not demonstrated" with "unsupported."

Which capabilities are most important for a university?

For individual students, ease of use and the ability to show how their own work was produced may be enough. For a university, the purchase often depends on a longer list: data agreements, role-based access, learning platform integration, retention schedules, disability accommodations, auditable reviews, support, policy control and reasonable classroom burden.

That is why a free individual tool and an institution-managed service should not be compared solely by monthly sticker price. A university is usually purchasing governance and operational reliability around the record.

AuthMarks' current public pricing distinguishes free individual recording/sealing from quote-based education plans with LTI 1.3, instructor tools and configured retention. Verify the scope of your proposed deployment with the vendor rather than assuming every feature is in the free tier. AuthMarks pricing · LMS setup.

Three red flags in proof-of-authorship marketing

"This proves no AI was used." That conclusion can exceed what recorded events establish, particularly where policies permit AI brainstorming, editing or other assistance.

"A signature proves the student typed it." A signature can verify an artifact's integrity and may associate an attestation with an account, but does not alone establish physical identity.

"Our score finds cheating automatically." Cheating is a policy-defined conclusion and requires institutional review. A probabilistic or behavioral signal is not a substitute for student response and procedural fairness.

Any vendor—AuthMarks included—should be judged against these standards.

A buyer-ready statement of requirements

We seek software that preserves reviewable evidence of a student's writing process, clearly distinguishes captured from uncaptured work, and defines the identity and integrity claims supported by its output. Our selection will consider student consent, instructor permissions, accessible participation, LMS integration, export or verification capabilities, retention, and the institution's ability to review concerns fairly. We do not require software to issue an automated finding of misconduct.

That paragraph can be attached to an RFI. Add local privacy requirements and the institution's own academic-integrity policy before issuing it formally.

Explore an AuthMarks pilot

If independent inspection of a sealed writing record and named attestation are central requirements, AuthMarks for institutions explains its proposed evidence workflow. The public verifier is available without an account for supported checks. For a faculty trial, request an institutional pilot; individual writers may start free.

Frequently asked questions

Is there software that can prove a student wrote an essay?

Software can provide evidence of observed writing and an accountable record. It cannot alone prove every off-device action, the typist's physical identity or adherence to all course rules.

Can a PDF signature prove the writing was human?

No. An integrity signature can support a claim that a specific artifact has not changed since signing. Human effort and authorship require additional observed evidence and contextual review.

Is a writing replay a form of proof?

It can be useful process evidence when the collection method, scope and limitations are known. A replay alone is not an absolute guarantee of who physically produced the work.

What if a student cannot consent to recording?

Institutions should establish a proportional, accessible alternative assessment or evidence route consistent with their policies and applicable law. Never treat a lack of consent as automatic proof of misconduct.

Related: Student authorship verification software · Writing provenance software · Writing process tools for educators · AI detection alternatives · Grammarly alternatives