Respect the ERA Living Guidelines
Catch Me If You Can Series
Responsible AI use starts with what you do not hide, leak, fake, automate, or outsource.
EvalCommunity Tutorial
How to Respect the ERA Living Guidelines on Responsible Generative AI Use in Research and Evaluation
A practical guide for using AI responsibly in research, evaluation, proposal writing, evidence synthesis, reporting, and assessment — without hidden prompts, ghostwriting, data leakage, fake citations, unauthorised recording, or integrity risks.
Tutorial Summary
The ERA Living Guidelines explain how researchers and organisations should use generative AI responsibly. For evaluation professionals, the key message is simple: AI can support expert work, but it does not remove expert responsibility.
This tutorial introduces the main rules and risks to avoid: hidden prompts, AI ghostwriting, confidential data leakage, fake citations, unauthorised AI meeting recorders, unsafe AI use, and improper AI use in peer review, proposal assessment, scoring, or ranking.
What You Will Learn
- What the ERA Living Guidelines mean for researchers and evaluators.
- How to distinguish light AI support from substantial AI use.
- How to avoid hidden prompts and document manipulation risks.
- How to protect confidential, personal, unpublished, and proprietary data.
- How to avoid fake citations, incorrect summaries, and unsupported conclusions.
- How to use AI safely in proposal writing, research, and evaluation practice.
- How to avoid unauthorised AI recording in confidential meetings.
- How to avoid substantial AI use in sensitive assessment activities.
- How to document and disclose AI use clearly.
Authoritative Sources Used
This tutorial is based on the European Commission’s ERA Living Guidelines on the responsible use of generative AI in research and related responsible AI guidance. Because the guidelines are living documents, users should check the latest version before applying them to formal research, funding, proposal, or evaluation processes.
Responsible AI Use Workflow
Use this workflow before applying generative AI to research, evaluation, proposal writing, reporting, or assessment.
Step 1
Define Task
Clarify what AI will support.
Step 2
Check Data
Protect confidential and personal data.
Step 3
Classify Use
Light, moderate, or substantial.
Step 4
Verify Output
Check sources, claims, and summaries.
Step 5
Disclose Use
Document substantial AI use.
Step 6
Keep Humans Responsible
Final judgment stays human-led.
1. Why the ERA Living Guidelines Matter
Generative AI is now used across research and evaluation workflows. It can help with writing, editing, summarising, translation, literature organisation, coding support, proposal drafting, and reporting.
But AI also creates risks. It can invent citations, misread documents, generate biased summaries, expose confidential data, and blur the line between human expertise and machine-generated content.
Core Message
AI can support expert work, but it does not remove expert responsibility. Researchers, evaluators, proposal teams, and reviewers remain accountable for the final output.
2. What the ERA Living Guidelines Say in Simple Terms
| Principle | Meaning for EvalCommunity Users |
|---|---|
| Transparency | Disclose substantial AI use when AI meaningfully shapes research, proposal, or evaluation work. |
| Responsibility | Humans remain accountable for final outputs, findings, recommendations, and decisions. |
| Verification | Check AI-generated citations, summaries, facts, and interpretations against original sources. |
| Confidentiality | Do not upload protected, unpublished, confidential, or personal data into external AI tools without safeguards. |
| Fairness | Avoid substantial AI use in sensitive activities such as peer review, proposal assessment, scoring, or ranking. |
3. What Counts as Acceptable AI Support?
AI use is usually lower risk when it supports editing, formatting, structure, or non-sensitive preparation — and when a human reviews the final output.
Usually Acceptable
- Improving clarity and readability.
- Editing grammar.
- Translating non-sensitive text.
- Creating outlines.
- Summarising public documents.
- Organising notes.
- Creating checklists.
- Drafting non-final text for human review.
Use More Caution
- Literature review synthesis.
- Qualitative coding support.
- Data interpretation.
- Methodology support.
- Analytical conclusions.
- Proposal development.
- Findings and recommendations.
4. What Counts as Substantial AI Use?
Substantial AI use happens when AI meaningfully shapes the intellectual, analytical, methodological, or evaluative content of the work.
Practical Test
Ask: did AI only help with wording, or did it shape the thinking? If AI shaped the research design, analysis, methodology, conclusions, findings, or recommendations, treat it as substantial use.
5. Main AI Integrity Risks to Avoid
This overview introduces the main responsible AI risks covered in the Catch Me If You Can Series. Use these short guides to understand what to avoid and how to apply safer AI practices in research, evaluation, proposal writing, review, and assessment.
Respect the ERA Living Guidelines
Understand how to use generative AI responsibly in research and evaluation while keeping transparency, verification, confidentiality, and human accountability at the centre.
Using AI Safely
Learn the basic safeguards for using AI tools safely, including checking data sensitivity, verifying outputs, documenting use, and keeping expert judgment human-led.
Avoid Hidden Prompts
Hidden prompts are concealed instructions placed in documents, metadata, formatting layers, comments, or PDF structures to influence how AI reads or evaluates content.
Avoid AI Ghostwriting
AI may polish or structure text, but it should not replace human experts in designing methodology, implementation logic, work packages, or final judgments.
Avoid Data Leakage
Do not upload personal data, unpublished research, confidential reports, proposal drafts, or proprietary methods into external AI tools without safeguards.
Avoid Fake Citations
AI may invent citations or misrepresent real sources. Every source, summary, quotation, and claim should be checked manually before use.
Avoid Unauthorised AI Recording
AI meeting recorders and transcription tools should not be used in confidential meetings without informing participants and obtaining consent.
Avoid AI in Sensitive Assessment
Substantial AI use should be avoided in peer review, proposal assessment, scoring, ranking, or evaluation of unpublished work.
6. Responsible AI Use in Proposal Development
AI can support proposal writing, but humans must lead the project design, methodology, work plan, partnerships, budget, and final decisions.
| Responsible Use | Risky Use |
|---|---|
| Improve clarity and consistency. | Let AI design the core methodology. |
| Summarise public background information. | Let AI invent work packages or partnerships. |
| Create non-final drafts for human revision. | Generate unsupported claims or fake references. |
| Prepare a compliance checklist. | Replace partner consultation or technical expertise. |
Audit-Ready Practice
Keep evidence that the core concept, methodology, implementation logic, and work packages were developed through genuine human expertise and partner collaboration.
7. How to Disclose AI Use Properly
Transparency should be clear and proportionate. Basic editing may not need a long statement, but substantial AI use should be documented and disclosed where required.
General Disclosure Template
Generative AI was used to support selected parts of this work, including [editing / summarising public documents / organising notes / drafting non-final text]. All AI-generated outputs were reviewed, verified, and edited by the authors or evaluation team. Generative AI was not used to make final judgments, replace human analysis, or independently determine findings or recommendations.
Proposal Disclosure Template
Generative AI was used as a support tool during proposal preparation, including [language editing / consistency checking / section structuring / summarising public background information]. The project concept, methodology, work plan, partnerships, budget, and final submission were developed and approved by the human project team.
8. Expert Checklist Before Using AI
- Is this task appropriate for AI support?
- Is the task editorial, organisational, analytical, or judgment-based?
- Does it involve personal, confidential, or unpublished data?
- Do I have permission to use this data?
- Is the AI tool secure enough for this task?
- Could the output affect people, funding, assessment, or professional judgment?
- Have I checked citations and summaries?
- Should I document or disclose AI use?
- Is a human expert still responsible?
9. Practical Exercise: Catch the Risk
Review the examples and decide whether each AI use is low risk, requires caution, or should be avoided.
| AI Use | Risk Level | What to Do |
|---|---|---|
| Grammar editing of a non-sensitive paragraph | Low | Review output before use. |
| Summarising a public EU guideline | Low to moderate | Verify against the original source. |
| Analysing identifiable interview transcripts | High | Do not upload unless approved and protected. |
| Adding hidden prompts to a proposal PDF | Avoid | Treat as an integrity risk. |
| Using AI to score another applicant’s proposal | Avoid | Treat as a sensitive assessment activity. |
10. Frequently Asked Questions
What are the ERA Living Guidelines?
They are European Research Area recommendations for responsible generative AI use in research. They focus on transparency, responsibility, verification, privacy, confidentiality, intellectual property, AI literacy, and avoiding substantial AI use in sensitive activities.
What is substantial AI use?
Substantial AI use occurs when AI meaningfully shapes research aims, methodology, literature review, analysis, findings, conclusions, recommendations, or assessment.
What is a hidden prompt?
A hidden prompt is a concealed instruction placed inside a document, metadata field, formatting layer, or file to influence how an AI system reads, summarises, or evaluates content.
Can AI be used for proposal writing?
Yes, but carefully. AI can help with clarity, structure, and consistency. Humans must lead the core concept, methodology, work packages, partnerships, and final submission.
Can AI be used in peer review or proposal evaluation?
Substantial AI use should be avoided in sensitive assessment activities such as peer review, proposal evaluation, scoring, ranking, or assessment of another researcher’s work.
What is the safest rule for AI use?
Use AI only where the task is appropriate, the data is safe to process, outputs are verified, and a human expert remains accountable for final decisions.
11. Final Takeaway
Nothing to Hide. Everything to Verify. Humans Accountable.
Responsible AI use is not about avoiding AI. It is about using AI with transparency, verification, privacy protection, professional judgment, and human accountability.
AI can help improve clarity, structure, productivity, and learning. But final research design, evaluation judgment, evidence interpretation, findings, recommendations, and ethical decisions must remain human-led.
