AI-Assisted Work Evidence

AI-Assisted Work Evidence

AI-assisted work evidence helps show how human work and AI assistance contributed to a final output.

It applies to writing, design, software, research, analysis, images, video, music, business documents, marketing materials, legal drafts, educational materials, product concepts, reports, policies, and other work where AI tools helped generate, transform, summarise, edit, classify, analyse, or refine material.

A final output is not enough.

If AI assistance matters, the evidence should show what humans provided, what the tool produced, what was edited, what was accepted, what was rejected, what source material was used, and what claim is being made about the final work.

The purpose of this guide is to help users preserve AI-assisted work evidence before questions about authorship, originality, accuracy, permission, disclosure, or responsibility arise.

Quick Read

• AI-assisted work evidence should show the relationship between human contribution and machine assistance.
• Strong records preserve prompts, inputs, outputs, edits, review notes, source material, tool context, versions, and approvals.
• AI-assisted evidence supports explanation, but it does not automatically prove ownership, authorship, legality, originality, accuracy, or permission.

What this means

AI-assisted work evidence is the evidence position around work created with the help of AI systems.

It asks whether someone can explain how the output was made and what role AI played.

For a writer, this may include prompts, source notes, draft comparisons, edits, and final human review.

For a designer, it may include source material, generated variations, prompt history, manual refinements, export history, and approval notes.

For a developer, it may include AI-generated code suggestions, human edits, repository commits, tests, review notes, and dependency checks.

For a business, it may include AI-assisted summaries, research outputs, reports, customer communications, policy drafts, internal approvals, and responsibility records.

The evidence should show the working path, not just the final result.

Why it matters

AI-assisted work can become difficult to explain later.

A person may claim authorship without preserving their human contribution. A business may publish AI-assisted material without keeping source records. A researcher may rely on AI summaries without preserving the underlying material. A developer may use generated code without recording review or testing. An organisation may make public claims from AI-assisted analysis without preserving the prompts, inputs, assumptions, or approvals.

That creates evidence risk.

The question is not simply “Was AI used?” The better question is:

What did AI do, what did humans do, and what evidence supports that distinction?

AI-assisted work evidence helps preserve that distinction before the record becomes contested, reused, published, licensed, audited, challenged, or relied on.

What strong AI-assisted work evidence should include

A stronger AI-assisted work evidence position usually includes:

• The final work — the document, design, code, image, video, report, analysis, dataset, message, or output being relied on.
• The AI-use claim — what is being said about the role AI played.
• Human input records — notes, instructions, source material, outlines, decisions, edits, selections, reviews, and approvals.
• Prompt records — prompts, system instructions, user instructions, parameters, tool settings, or workflow instructions where relevant.
• Output records — generated responses, variations, drafts, summaries, code, media, recommendations, or transformations.
• Source material — materials supplied to or used alongside the AI system.
• Tool context — the AI tool, model, service, version, account, workspace, or environment where relevant and available.
• Edit and selection context — what was changed, rejected, accepted, rewritten, corrected, merged, or approved by humans.
• Version context — drafts, iterations, exports, commits, comparisons, or publication versions.
• Accuracy and review context — checks, tests, fact review, legal review, human sign-off, or quality controls.
• Authority context — who had the right to use the source material, tool, account, or output.
• Custody and retention context — where records are preserved and how they can be recovered.
• Disclosure context — whether AI use needs to be disclosed internally, contractually, publicly, or professionally.
• Claim boundaries — what the evidence supports and what it does not support.

The right level of detail depends on the use case and risk.

Common weak points

AI-assisted work evidence is usually weak when:

• only the final output is preserved
• prompts are lost
• source material is not retained
• AI outputs are copied into final work without version history
• human edits are not recorded
• rejected or alternative outputs are not preserved where they matter
• the tool or model context is unknown
• AI-generated summaries are relied on without preserving underlying sources
• generated code is used without review, testing, or dependency records
• images, video, or audio are published without source or alteration context
• contribution claims are vague
• AI use is hidden where disclosure may matter
• human authorship is overstated
• originality, legality, permission, or accuracy is assumed
• public claims imply EviWrite verification where none exists

These weaknesses make later explanation harder.

How to apply this yourself

For each important AI-assisted work, create an AI-use evidence note.

Ask:

• What is the final work?
• What claim are we making about human contribution, AI assistance, authorship, accuracy, or use?
• What human input existed before AI assistance?
• What prompts, instructions, files, datasets, or source materials were supplied?
• What did the AI system produce?
• What did humans edit, select, reject, approve, test, or publish?
• What tool, model, service, account, or workflow was used, where relevant?
• What versions or iterations show the development path?
• What source material, permissions, or rights context matter?
• What review or accuracy checks were performed?
• Is disclosure required or appropriate?
• What does the evidence not prove?

Then preserve the final work, prompts, inputs, outputs, source material, edits, versions, review notes, authority context, and claim boundary together.

Do not wait until someone asks how the work was made.

What this does not prove

AI-assisted work evidence does not automatically prove:

• human authorship
• copyright ownership
• legal ownership
• originality
• non-infringement
• permission to use source material
• accuracy
• truth
• safety
• compliance
• absence of bias
• absence of AI use beyond the recorded workflow
• that a model did or did not train on particular material
• that EviWrite has verified or backed the record

AI-assisted work evidence helps explain contribution and process. It does not settle every legal, factual, or ethical issue.

Framework-aligned claim boundary

A person or organisation may use this guide as part of EviWrite Framework alignment if they apply the guidance honestly and avoid implying EviWrite involvement.

Acceptable wording may include:

“We use the EviWrite Framework to preserve evidence around AI-assisted work.”

It must not be used to imply:

• EviWrite has verified the AI-assisted work
• EviWrite has confirmed human authorship
• EviWrite has confirmed originality
• EviWrite has confirmed permission or legality
• EviWrite has approved the AI tool or workflow
• the record is EviWrite-backed
• the record is EviWrite-certified
• the record carries the controlled ⓔ mark

Framework-aligned means public guidance was followed.

EviWrite-backed means the record was created through EviWrite or an authorised evidencing channel.

Related checklist

Use the AI-Assisted Work Evidence Checklist to check whether prompts, inputs, outputs, human edits, source material, review records, authority context, disclosure position, and claim boundaries have been preserved clearly.