Use AI for project preparation without exposing sensitive information
AI can help organize a draft, but fluent text is not proof of accuracy, compliance or eligibility. Keep responsibility with your team and do not treat generated advice as an official interpretation of a call.
1. Define a low-risk task
Good starting tasks include improving a public summary, suggesting questions for a team meeting or identifying unclear passages. Decide what a useful result looks like before opening the tool.
Avoid asking a model to invent evidence, quotations, partner commitments or budget rules. Use the official call documents as the authority.
2. Minimize the information you share
Remove names, contact details, participant records, confidential budgets and unpublished partner documents. Use placeholders and public or synthetic examples. Check the provider's retention settings, contractual terms and your organization's rules before use.
Removing a name does not necessarily anonymize information: combinations of details may still identify someone. Seek appropriate advice where personal-data processing is involved.
- No credentials, personal records or confidential files in prompts.
- Permission to use any shared material confirmed.
- Provider and organizational data-handling rules reviewed.
3. Ask for structure, not invented certainty
Give a limited task and require the model to identify gaps instead of filling them with assumed facts.
Using only this public, non-confidential summary, propose an outline with objectives, activities and questions we still need to answer. Do not invent statistics, organizations, costs, sources or eligibility rules. Mark unsupported points as 'needs verification'.
4. Review before using the output
A team member must check every factual claim and cited link against its primary source. Verify calculations separately and check whether the proposed work fits your real capacity. Review language for bias, accessibility and exaggerated promises.
Record which tasks used AI and follow any disclosure requirements in the applicable call or organizational policy. Never submit generated content without accountable human review. This guide is not legal advice.
Sources and further reading
Practical guides