Start with a six-field factual record

Write a private record before a polished sentence. Polished wording can read as complete while hiding a missing fact. Create one entry for every tool-task pair that materially influenced the submitted work, with six fields:

1. Tool: the product used. Record the version or model only if it is known and the instructions ask for it; otherwise do not guess.
2. Task: a precise verb, such as brainstormed, outlined, drafted, rewrote, corrected, translated, summarised, analysed, or generated code.
3. Scope: the affected chapter, paragraph, figure, dataset, calculation, function, or stage.
4. Retained output: whether generated ideas, wording, translations, summaries, analysis, or code remain in the submission.
5. Verification: a check you genuinely completed, not the one that sounds most reassuring.
6. Responsibility: confirmation that you reviewed and accept responsibility for the work.

Worked example. A student asked a tool for four possible headings, kept two after revising them, wrote the paragraphs independently, and checked the structure against the assignment question. The record converts to: “I used [tool] to suggest four essay headings. Two revised headings remain; I wrote the paragraphs independently. I checked the structure against the assignment question and take responsibility for the submission.” The phrase “the tool helped with planning” would hide the retained headings, while “the tool wrote the essay” would overstate its role.

A decision tree for classifying AI involvement: ideation, drafting, editing or grammar, translation, and data analysis

This decision tree classifies AI involvement into ideation, drafting, editing or grammar, translation, and data analysis, and it gives a rule for which levels normally require disclosure. It helps you name the contribution precisely; it cannot decide whether that contribution was permitted.

1. Ideation (brainstorming): did the tool shape the research question, angle, examples, or structure? A retained outline can be material even when every final sentence is your own. Rule: disclose when a generated idea or structure survives into the submission.
2. Drafting: did the tool produce submitted sentences, paragraphs, captions, or an abstract? Rule: always disclose retained generated prose and name where it sits. Editing that prose later does not turn drafting into ideation.
3. Editing or grammar: separate mechanical correction from substantive rewriting. Fixing isolated spelling, punctuation, or agreement without changing meaning is mechanical; changing tone, argument order, or substantial wording is rewriting. Rule: mechanical correction is often exempt, but confirm it against the policy, and rewriting normally requires disclosure.
4. Translation: did the tool create submitted text in another language? Record the source and target language, the affected sections, the retained translation, and the meaning check you ran. Rule: disclose whenever translated text is submitted, because a fluency read is not a comparison with the source.
5. Data analysis or code: did the tool propose a method, calculation, query, function, or script? Name the component and the inspection, comparison, or tests you completed. Rule: disclose any retained method or code, and never claim comprehensive testing after running it once.

When one tool did materially different jobs, such as suggesting an outline and later rewriting a conclusion, give each its own clause. Combining them as “writing support” hides the drafting contribution.

How much detail to include: the material-detail test

Include every tool-task pair that materially influenced the work, its scope, retained output, and completed checks. Leave out experiments that changed nothing, and do not pad the statement with prompt-by-prompt history or version numbers unless the policy asks for a full activity log.

First, ask whether the submission would be meaningfully different without the interaction. An outline that reordered your argument is material even if no generated sentence remains; a rejected response that influenced nothing may be immaterial, unless all use must be reported.

Second, test for concealment. “I used AI to improve my paper” could mean punctuation fixes, rewritten arguments, or drafted analysis. Replace vague verbs like “helped” or “improved” with the task, the location, the retained output, and the real check.

Third, remove padding. Give enough that no material role is hidden, but not so much that marketing, prompt logs, or model version numbers bury what matters. Aim for the shortest statement that leaves every material role clear, unless the instructions request a complete log.

Scenario. A researcher rejects three generated titles, translates two submitted paragraphs, and accepts seven sentence-level rewrites. If the titles influenced nothing, only the translation and the rewrites are material, and each needs its own clause; the abandoned titles need not appear unless the policy requires every interaction. But if one rejected title became the basis of the final title, that interaction turns material even though the wording changed, because materiality is about influence on the submission, not verbatim retention.

Three adaptable statement templates: essay, thesis, and journal submission

These are three adaptable statement templates written as fill-in wording, not examples to copy verbatim. Keep the underlying facts stable and change emphasis, location, and grammatical person to fit the format. Replace every bracket and delete any clause that is not true.

Undergraduate essay

AI-use disclosure: I used [tool] during [stage] to [task] for [scope]. [State what output remains.] I checked it by [actual method]. I reviewed and take responsibility for the final argument, evidence, citations, and wording.

Thesis or dissertation

AI-assistance declaration: I used [tool or tools] in [chapters or stages] for [specific tasks]. [Identify retained language, ideas, summaries, analysis, or code.] I checked the output by [method]. I take responsibility for the research design, source use, analysis, and conclusions.

Journal manuscript

During preparation of this manuscript, the author or authors used [tool] for [task and section]. [Describe retained output.] The output was checked using [verification method] and revised as needed. The author or authors take responsibility for the final manuscript.

Use first person only when permitted, and “the author” or “the authors” only where that convention fits the manuscript. Never add a formal claim such as “all output was independently validated” unless you can name the checks behind it. An AI-use disclosure does not replace ordinary source attribution; the citation generator can help format a separate reference, though its input fields and output still require review.

Describe several tools without merging their roles

When more than one tool was involved, give every materially different tool-task pair its own row so the distinct roles stay visible. Split one product across two rows when combining its jobs would hide a difference.

Do not compress these into “I used several AI tools for research and writing.” That wording erases the line between translation, rewriting, summarisation, and code generation. If a verification cell is blank, do not invent reassurance: run an appropriate check, or state the limitation honestly when you ask the responsible authority what is required.

Match your statement to your course or publisher policy

Before submitting, run this checklist against your specific policy language:

1. Locate the governing document: assignment brief, academic integrity policy, thesis regulations, journal author guidelines, or submission form.
2. Compare every record row with the permitted, restricted, or prohibited uses it names.
3. Use the policy’s own terminology only where it accurately describes what happened.
4. Check whether prompts, outputs, drafts, tests, or correspondence must be kept or submitted.
5. Place the statement where required: an acknowledgement section, methods note, declaration field, title-page section, or form field.
6. Read the finished statement against your record and delete any check you did not perform.

If assistance was prohibited, disclosure does not cure it. Where it was permitted, report it factually rather than as an apology.

What if the policy is silent or ambiguous?

Keep your factual record and ask the responsible instructor, supervisor, editor, or research-office contact before submitting. Do not treat silence as automatic permission or automatic prohibition.

Send a decision-ready message: “I used [tool] to [task] in [scope]. Its output [does or does not] remain, and I checked it by [method]. The instructions do not appear to address this. Is it permitted, what must I disclose, and where should the statement go?” If one document permits general editing while another bans generated text, name both provisions and do not assume the broader phrase overrides the specific restriction. If an answer is partial, ask separately about permission, required detail, and placement; if it is given verbally, keep a dated note of what was decided and when. This is recordkeeping guidance, not a claim that every institution or publisher accepts informal permission.

Disclosure is not an admission of wrongdoing

A truthful record of permitted assistance is not an admission of cheating. If the assistance was prohibited, though, disclosure does not make it acceptable.

Disclosure records how the work was produced. AI detection does something different: it estimates patterns in finished text. FiftyGPT’s responsible-use position is that detector results are probability signals, not proof, and should be weighed with writing history, citations, drafts, and human review. If independently authored work was flagged, use the separate guide on how to respond to an AI flag rather than treating disclosure as a defence. More planning and checking resources sit in the academic tools directory.

Frequently asked questions

Do I need to disclose using a grammar checker?

Check the governing instructions and separate mechanical corrections from substantive rewriting. Do not describe extensive changes to wording, structure, or argument as a simple spelling, punctuation, or grammar check.

Where should I put an AI-use disclosure statement?

Use the location named in the instructions, such as an acknowledgement section, methods note, declaration field, title-page section, or submission form. Ask the responsible contact if none is specified.

What if the policy does not mention AI?

Document the tool, task, scope, retained output, and completed checks, then ask the responsible authority before submitting. Do not infer either permission or prohibition from silence.

Should I disclose AI translation or summarisation?

When translation or summarisation materially created submitted text, record the languages or source, the affected sections, the retained wording, and the check against the original, then follow the governing instructions.

Can one statement cover several AI tools?

Yes, provided every material role stays clear. Use separate clauses when combining ideation, drafting, rewriting, translation, summarisation, analysis, or code would conceal an important difference.