Write a concept note with AI

Use an AI assistant to draft a funding concept note faster, with a hand check after every step and clear limits on what data you paste in.

WhatsAppEmail

What You Will End Up With

A concept note is a brief outline of a proposed project, written to be read before a full proposal. By the end of this guide you will have a short one, drafted with an AI assistant and checked by you line by line, ready for your own final check and for review by someone with program knowledge.

It is worth the effort for two reasons. Funders that ask for concept notes use them to check fit with their priorities and to screen out proposals unlikely to be funded. And a concept paper can bring feedback and guidance from a program officer before you sink weeks into a full proposal.

Before You Start

Gather these before you open any AI tool:

  • The funder's own instructions. Length and format depend on the funder, so its call decides. Read it first, and note whether it wants a budget.
  • Your design notes. The need, the objective and the planned activities, written down by you. AI will not rescue a vague project idea, so settle the design before you draft.
  • A cleaned set of source material. Strip out anything confidential, personal or sensitive.
  • Permission to use the tool. Approved enterprise tools and public consumer tools carry different risks. Find out which one your organization allows.

Steps

  1. Read the call and fix the limits. Length is the funder's decision. One guide reports that most funders that set a page range ask for three to five pages, another advises two to three unless the funder wants more, and neither beats the instructions in front of you. Note whether a budget is requested. Include one only if the funder asks; otherwise leave it out. Expect to write a different version for each funder, because a note only has a chance if it fits that funder's framework.

    Check by hand: write the page limit, format and budget rule at the top of your working file so you can see them while you edit.

  2. Settle the design without AI. Decide the objective yourself. It should name the beneficiaries, the geographic scope and the substantive focus, and it should not describe how you will get there. Then test the logic top-down and bottom-up: if the activities are delivered, will the outcomes follow, and will the outcomes meet the objective?

    Check by hand: does your one-paragraph design name the beneficiaries, the geographic scope and the substantive focus, without describing how you will get there?

  3. Draft the need and background. Give the assistant your notes, your target population and the statistics you have sourced.

    Instruction to the assistant: "Using only these notes, draft the background. Explain why the problem needs to be addressed and what has already been done about it. Use the statistics I have supplied and no others. Flag every gap where more data is needed."

    Check by hand:

    • Does the problem statement name the target population?
    • Is every statistic accurate and traceable to a reference?
    • Does it say what has already been done about the problem?

  4. Draft goals, methodology and deliverables. Ask for a set structure, and adapt it, since the best format depends on your project. A common format uses the headings Need, Goals, Methodology and Deliverables.

    Instruction to the assistant: "Draft the goals, methodology and deliverables in that order. List activities in sequence and show how each contributes directly to the objective. Include intangible deliverables such as partnerships, trained groups or process improvement alongside tangible ones. For each result, suggest an indicator with a defined unit and a target covering quantity, quality and timing."

    Check by hand: run the top-down and bottom-up test again on the AI's version.

    • Do the indicators measure skills gained or actions taken afterward, rather than restating an activity such as the number of people trained?
    • Does every activity contribute directly to the objective?
    • Do the outcomes, taken together, meet the objective?

  5. Tighten and adapt. Now ask for polish.

    Instruction to the assistant: "Cut this to fit the page limit. Use clear, concise language. If the reviewer is a generalist, replace technical jargon. Rewrite the opening sentences so they make the reader want to continue."

    Check by hand:

    • Is the title snappy, informative and distinctive?
    • Does the text sound confident and specific rather than vague or unsure?
    • Have you removed detail that overloads the reader?

    The opening sentences decide whether anyone keeps reading, so rewrite them yourself if the assistant's version is flat. If you use the tool for a plain-language summary, fact-check it and confirm it is culturally appropriate for the people it describes.

  6. Verify every claim. Put the draft next to your own records and check each figure, date, place and partner name. You are responsible for the accuracy, completeness and relevance of everything the assistant produced. Remember too that a concept note can bind you: once you are invited to a full proposal, the project aim should usually stay as stated, so do not let the assistant promise more than you intend to deliver.

    Check by hand:

    • Has someone with program knowledge read every claim?
    • If a reviewer asked for evidence, could the team back up each one?

  7. Confirm the disclosure rules and finalize. Some funders require applicants to disclose AI use, and their guidance changes over time. Look up the current rules for this specific submission, not the ones you remember from the last.

    Check by hand:

    • Have you re-read the funder's current AI rules today?
    • If disclosure is required, does your statement say when and how AI was used?

    Even where disclosure is not required, being open about how AI was used in a work product is good practice.

Using AI Safely

Treat AI as a data-risk issue. What you type in matters more than which tool you use.

  • Never paste confidential, personal or sensitive data into a public AI tool. If a document contains it, do not upload the document.
  • De-identify personal data and keep identifiable information off public tools.
  • Treat approved and public tools differently. An enterprise tool your organization has approved is not the same risk as a free consumer tool.
  • Have staff review the output before it is shared outside the organization or used for a decision.

Some research funders allow AI tools to draft, translate and summarize parts of a proposal, and some require you to say so. Because those rules move, re-check them before each submission.

Common Mistakes

Pasting sensitive data into a public tool. It is easy to drop in a field report to save time. See Using AI Safely above.

Sending one generic note to every funder. A note fitted to one framework rarely fits another. Rework it for each funder.

Using AI before the design is settled. The assistant will write fluent text around a weak idea. Finish steps one and two by hand first.

Writing indicators that restate activities. Ask what those people can now do.

Submitting AI text unverified. You are responsible for the accuracy, completeness and relevance of everything the assistant produced.