Avoid AI Ghostwriting
Catch Me If You Can Series
AI can support the writing — but humans must lead the design.
EvalCommunity Tutorial
Avoid AI Ghostwriting
A practical guide for using AI to improve clarity and structure without letting it replace human experts in methodology design, implementation logic, work packages, findings, or final judgment.
Quick Answer
AI ghostwriting happens when generative AI creates the intellectual core of a proposal, research output, or evaluation report while the final document presents the work as if it was fully designed by human experts.
AI may support editing, clarity, structure, consistency checks, and non-final drafting. It should not replace humans in designing methodology, developing work packages, interpreting evidence, producing findings, or making recommendations.
What You Will Learn
- What AI ghostwriting means in research, evaluation, and proposal writing.
- How to distinguish AI editing from AI ghostwriting.
- What AI can support safely without replacing human expertise.
- What AI should not do in methodology, work packages, findings, or recommendations.
- How to keep audit-ready proof of human design and partner contribution.
- How to use safer prompts that keep AI in an assistant role.
- How to document or disclose AI use when needed.
Main Sources
This tutorial is based on the European Commission’s ERA Living Guidelines on the responsible use of generative AI in research and related EvalCommunity guidance. The ERA guidelines distinguish basic editorial support from substantial AI use, especially when AI shapes methodology, data interpretation, research aims, hypotheses, conclusions, or other core parts of the work.
AI Ghostwriting Prevention Workflow
Use this workflow before using AI in proposal writing, research design, evaluation reports, or methodology sections.
Step 1
Start Human
Develop the core logic first.
Step 2
Limit AI
Use AI for clarity and structure.
Step 3
Protect Data
Avoid confidential uploads.
Step 4
Review Output
Check meaning and evidence.
Step 5
Keep Proof
Show human ownership.
Step 6
Disclose if Needed
Document substantial use.
1. Why AI Ghostwriting Matters
Generative AI can help researchers, evaluators, consultants, and proposal writers improve clarity, structure, and readability. It can make a paragraph clearer, reorganise notes, suggest headings, and support non-final drafting.
The risk begins when AI moves from supporting the writing to replacing the thinking. Research, evaluation, and proposal writing depend on expertise, evidence, context, partner input, and professional judgment.
Core Rule
AI can help improve the text, but it should not become the hidden author of the methodology, implementation logic, work packages, findings, or final judgment.
2. What Is AI Ghostwriting?
AI ghostwriting happens when AI creates substantial parts of the intellectual, analytical, methodological, or evaluative content of a proposal, research output, or evaluation report without transparent human ownership, review, evidence, or disclosure.
The problem is not simply that AI helped. The problem is that AI may have created the core thinking while the document presents the work as expert-designed.
Practical Test
Ask: did AI improve how the work is presented, or did AI create the work itself?
3. What AI Can Support Safely
AI can support writing when it helps improve presentation, not replace expertise.
Language Support
- Improve grammar
- Improve clarity
- Shorten dense paragraphs
- Improve readability
- Translate non-sensitive text
“`
Structure Support
- Reorganise human-written notes
- Suggest headings
- Create a first outline from existing material
- Check consistency between sections
- Format text or tables
Preparation Support
- Create a compliance checklist
- Draft non-final text for revision
- Summarise public background material
- Prepare agenda drafts
- Support final proofreading
4. What AI Should Not Do
AI should not be used as the hidden designer of the work. These tasks require human expertise, evidence, and accountability.
- Do not ask AI to design the full methodology from scratch.
- Do not ask AI to invent work packages or project activities.
- Do not ask AI to create partner roles without consultation.
- Do not ask AI to write findings without evidence.
- Do not ask AI to produce recommendations without analysis.
- Do not ask AI to replace stakeholder consultation.
- Do not ask AI to replace technical expertise.
- Do not ask AI to create a Theory of Change without human validation.
- Do not ask AI to produce final judgments on programme performance.
5. Editing or Ghostwriting?
| AI Use | Classification | Reason |
|---|---|---|
| Improve grammar in a human-written methodology | Usually acceptable | AI improves wording only. |
| Reorganise human notes into headings | Usually acceptable | AI structures existing human content. |
| Create a checklist from public guidance | Usually acceptable | AI supports organisation. |
| Draft a methodology from no human input | Ghostwriting risk | AI creates the core design. |
| Invent work packages for a proposal | Ghostwriting risk | AI replaces project planning. |
| Generate recommendations without human review | Ghostwriting risk | AI replaces final judgment. |
6. Ghostwritten Trap vs Audit-Ready Human Proof
AI-polished writing can make weak ownership look stronger than it is. The safer approach is to keep evidence that humans led the design.
| Ghostwritten Trap | Audit-Ready Human Proof |
|---|---|
| AI generates the core methodology and research aims. | Co-design notes, workshop logs, early drafts, and expert comments show human-led methodology development. |
| AI writes generic literature review logic and partner contribution claims. | Verified sources, partner evidence, consultation records, and source notes show real contribution. |
| AI-polished text hides weak ownership and generic workflows. | Tool disclosure, partner input, pilot evidence, version history, and approval records create an audit trail. |
| AI invents activities that sound plausible. | Work plan decisions are linked to needs analysis, partner roles, budget, and implementation evidence. |
| AI writes recommendations without evidence. | Recommendations are linked to findings, data sources, stakeholder input, and evaluator judgment. |
Practical Rule
AI can make your work clearer. It should not make your work look more human-led than it really was.
7. How to Use AI Without Ghostwriting
- Start with human thinking: write your own notes, logic, methodology, or outline before using AI.
- Define what AI may do: limit it to clarity, structure, consistency, or non-final drafting.
- Protect sensitive content: do not upload confidential partner documents, proposal drafts, unpublished research, or identifiable data unless approved and protected.
- Review the output: check whether AI changed meaning, added claims, removed nuance, or overstated evidence.
- Restore human judgment: add context, evidence, methodological detail, limitations, and partner-specific information.
- Keep proof: retain records showing that humans led the concept, methodology, work packages, and final decisions.
- Disclose if needed: document substantial AI use according to funder, journal, client, institutional, or organisational requirements.
8. Safer Prompt Formula
Use prompts that keep AI in an assistant role.
Formula
Role + Existing Content + Clear Limits + Human Review
Act as an editorial assistant. Improve the clarity and structure of the text below. Use only the information provided. Do not add new methods, claims, sources, work packages, findings, or recommendations. Do not change the meaning. I will review and approve the final version.
9. Safer Prompt Examples
Editing Methodology Text
Improve the clarity of the following methodology section. Do not add new methods, data sources, sampling strategies, or claims. Keep the meaning unchanged.
“`
Structuring Proposal Notes
Organise the following human-written notes into a proposal section outline. Use only the notes provided. Do not invent activities, partners, outputs, or work packages.
Checking Consistency
Check whether terminology is consistent across these sections. Do not rewrite the methodology or add new content.
Creating a Checklist
Create a checklist based only on the criteria below. Do not assess the proposal or recommend a score.
“`
10. Risky Prompt Examples to Avoid
- Write the full methodology for this project.
- Create work packages for this proposal.
- Invent a strong Theory of Change.
- Generate findings from limited notes.
- Write recommendations even if evidence is missing.
- Make this proposal sound innovative even if the design is weak.
- Create partner roles that make the consortium look stronger.
- Write a convincing literature review with citations.
- Draft the final evaluation judgment.
11. AI Ghostwriting in Proposal and Evaluation Practice
| AI Can Support | Humans Must Lead |
|---|---|
| Language editing and readability | Project concept and problem analysis |
| Section organisation and consistency checks | Methodology and implementation logic |
| Public background summaries | Work packages and partner roles |
| Drafting non-final report text | Findings, conclusions, and recommendations |
| Formatting and compliance checklists | Final submission and final evaluation judgment |
12. Audit-Ready Checklist
- Can we explain who developed the core concept?
- Can we show how the methodology was designed?
- Can we prove that partner input shaped the work?
- Can we link work packages to real planning decisions?
- Can we show verified sources for the literature review?
- Can we link findings to evidence?
- Can we explain who approved the final recommendations?
- Did AI only support wording and structure?
- Did AI add any unsupported claims?
- Should AI use be disclosed?
- Is the final output clearly human-owned?
13. Simple AI Use Notes
For Basic Editing
AI was used for language editing and clarity improvement only. The content, methodology, analysis, work packages, findings, and final wording were reviewed and approved by the human team.
For Proposal Development
AI was used as a support tool for editing, structure, and consistency checks. The project concept, methodology, work packages, partner roles, budget logic, and final submission were developed and approved by the human project team.
For Evaluation Reporting
AI was used to support selected drafting and editing tasks. It was not used to independently analyse data, produce findings, make evaluative judgments, or determine recommendations. All outputs were reviewed and validated by the evaluation team.
14. Practical Exercise: Is This AI Ghostwriting?
| Example | Classification |
|---|---|
| AI improves grammar in a human-written methodology. | Low risk |
| AI organises human notes into headings. | Usually acceptable |
| AI creates a checklist from funder criteria. | Usually acceptable |
| AI writes a methodology from a blank prompt. | Ghostwriting risk |
| AI invents partner roles. | Ghostwriting risk |
| AI writes recommendations without evidence. | Ghostwriting risk |
15. Frequently Asked Questions
What is AI ghostwriting?
AI ghostwriting happens when AI creates substantial parts of the intellectual, analytical, methodological, or evaluative content of a proposal, research output, or evaluation report without transparent human ownership, review, or disclosure.
Is AI editing the same as ghostwriting?
No. AI editing is usually lower risk when it improves clarity, grammar, or structure without changing meaning or adding new claims.
Can AI help write a methodology section?
AI can help improve the wording of a human-designed methodology. It should not design the methodology from scratch or invent methods, data sources, or sampling strategies.
Can AI write evaluation findings?
AI should not independently produce final findings or recommendations. Evaluation findings must be grounded in evidence and validated by human evaluators.
What proof should teams keep?
Teams should keep co-design notes, meeting records, early drafts, partner comments, source verification notes, methodology decisions, version history, and final approval records.
16. Final Takeaway
AI Can Polish the Work. Humans Must Own the Work.
AI can polish, organise, and support writing. But it should not become the hidden author of the methodology, proposal logic, work packages, findings, or final judgments.
Use AI to improve how expert work is presented. Do not use AI to replace the expert work itself.
