PlagiatScanner

AI transparency · guidance · KIT-08

How to disclose AI-assisted outlining

In brief: If an AI system influenced chapter functions, sequence or argumentative shape, retain your starting outline, the prompt and proposal, the changes adopted, and your academic reasons. A before-and-after outline log shows more than the fact that a tool was present: it identifies the structural judgement you tested and accepted. The rules governing whether and where to declare that use remain those of your particular assessment.

Before-and-after outline log

A decision record for substantive outline changes
Area Starting version System proposal Final choice and reason
Chapter 2 One background chapter Separate concepts and literature review Accepted: different evidence and clearer functions
Chapter 3 Method before material Describe material first Rejected: method defines selection logic
Chapter 4 Results by case Results by analytical category Accepted: matches pre-defined coding framework
Discussion Single section Separate limitations Accepted: assumptions become easier to inspect

Add date, service, displayed model label, prompt reference and any method or source that informed each decision. A screenshot preserves a proposal but not your reasoning.

An outline is intellectual architecture

Headings distribute jobs across a thesis. The introduction delimits the problem; a conceptual chapter defines terms; the method justifies a procedure; results report observations; discussion interprets them. Changing this architecture can alter what counts as evidence and where a reader encounters qualifications. Outline assistance can therefore matter more than surface proofreading even when no generated paragraph survives.

Generic systems do not automatically know your marking criteria, disciplinary conventions or complete literature. A polished standard outline may be unsuitable for a design project, doctrinal analysis, qualitative study or mathematical proof. Do not retain a heading because it sounds academic. Ask what question it answers, which evidence belongs there, what duplication it prevents, and how it advances the research question.

A disciplined route from suggestion to structure

  1. Freeze a baseline: Save a dated outline that includes one-sentence chapter purposes.
  2. Constrain the request: Ask for alternatives to a defined problem rather than an entire thesis plan without context.
  3. Classify the response: Separate genuinely new moves, restatements and conflicts with the project.
  4. Test disciplinary fit: Compare each move with question, method, material and assessment brief.
  5. Record the reason: Give one content-based reason for each important acceptance or rejection.
  6. Map the final version: Connect each log entry to its final chapter number.

Rejected options belong in the trail when they clarify why a consequential path was not taken. Minor renumbering need not swamp the record. Focus on decisions that alter logic, scope, evidence or the separation between reporting and interpretation.

A proportionate disclosure statement

“On 14 August 2026 I used [service and displayed model] to produce two alternatives to an outline I had already drafted. I adopted separate sections for the conceptual framework and literature review, and a distinct limitations subsection. I retained the original order of method and material because the sampling procedure determines the material. The starting outline, proposal and reasons are recorded in Appendix A.”

This statement neither attributes the whole design to the system nor reduces its influence to “general support”. If only a heading was shortened, say so. If the generated architecture became the main structure, the disclosure must reflect that wider contribution and the local rules may require further discussion.

Review the final outline without the prose

Read the table of contents alone and complete the sentence “This chapter is necessary because …” for every heading. A section without a defensible answer may be decorative or repetitive. Check that concepts are defined before use, methodological choices precede the results they produce, and results and discussion retain different functions.

  • Does the structure answer the research question without creating a second project?
  • Are headings at the same level logically comparable?
  • Did any AI-proposed section survive despite lacking evidence?
  • Are counterevidence and limitations given a real place?
  • Do the log, contents page and AI declaration agree?

Use the adaptable declaration for the overall statement. A technical assessment of final prose cannot reproduce this history; the AI detection overview explains the distinction. The AI transparency hub connects outline evidence with prompt and model records.

Protect unpublished and sensitive material

Do not insert identifiable interviews, confidential partner information or unpublished findings merely to obtain a better outline. Abstract the topic and constraints where possible, follow the approved data plan, and use institutionally authorised systems when required. A public appendix should not expose protected prompts. A redacted record can identify the information category and explain what was removed.

The log establishes provenance, not quality. A fully documented outline may still be academically weak. Quality depends on fit between question, evidence, method and reasoning. The record should make those human evaluations visible rather than treating system output as validation.

After the first full draft

Repeat the outline review once prose exists. Note which sections genuinely carry evidence, where results appear too early, and which transitions are verbal rather than logical. If a system is consulted again, preserve it as a new round tied to the current draft. Justify every change through question, material and method. This distinction shows whether the tool shaped the original conception or merely identified a later structural problem.

Sources

Sources reviewed 4 September 2026.

Related guidance