Exercises — Chapter 24: Grant Writing with AI

Work these with your own proposal and your own AI tool (if you use one) in mind. The goal is to build the habits — verify everything, own your voice, protect confidentiality — that make AI a safe, useful assistant rather than a liability.

How to use these: Part A checks recall; Part B applies the chapter to concrete AI-use decisions; Part C asks you to create at a real applicant's level (an AI-use plan, a verification pass); Part M interleaves earlier chapters. Answers to selected exercises (★) are in the back matter.

Part A — Recall and Understand

A1. ★ State the chapter's threshold concept in your own words. What can AI do for a proposal, and what can it never do?

A2. What is a large language model, and why does hallucination follow from how it works?

A3. Name four tasks where AI genuinely helps, and the property they share that makes them safe.

A4. ★ Name four uses where AI is dangerous, and the property they share that makes them risky.

A5. What is human-in-the-loop, and what are its four concrete commitments?

A6. State the peer-review red line and why it exists.

A7. Define: hallucination, data privacy, disclosure, authorship/accountability.

Part B — Apply

B1. ★ Safe or dangerous? For each AI use, decide safe or dangerous and why: - (a) Asking AI to tighten three paragraphs you wrote. - (b) Asking AI for five statistics about your problem to anchor the needs section. - (c) Asking AI to critique your draft as a skeptical reviewer. - (d) Pasting a colleague's confidential unpublished proposal into a public AI tool to summarize it. - (e) Asking AI to summarize a funder's guidelines document you paste in. - (f) Asking AI to generate plausible preliminary results to strengthen a thin section.

B2. The fabricated citation. AI gives you a perfectly formatted citation supporting a key claim. Describe exactly what you do before it could appear in your proposal, and why.

B3. ★ Rewrite for voice. Take a generic, AI-sounding sentence (write one or use a real one) and rewrite it into a specific, vivid voice grounded in a real project's details. Explain why the reviewer prefers the result.

B4. Reviewer ethics. You're reviewing a confidential proposal and tempted to use AI to help draft your critique. What's the rule, what's the risk, and what do you do instead?

B5. Disclosure decision. You used AI to edit your prose and brainstorm framings. How do you decide whether and what to disclose? Walk through your reasoning.

Part C — Analyze and Create

C1. ★ Write your AI-use plan. Using the Section 24.6 checkpoint, write a short AI-use plan for your proposal: where you'll use AI, where you won't, your verification commitments, and your disclosure decision. This goes in your "My Proposal" document.

C2. Run a verification pass. Take a passage where AI helped (or simulate one) and run the Section 24.4 verification pass: flag every factual claim, verify each against a real source, check meaning survived, rewrite for voice, scan for funder fit, confirm confidentiality. Document what you found.

C3. ★ The intern test. For five specific grant-writing tasks, decide whether you'd delegate each to "a brilliant but sometimes-fabricating intern" and what verification you'd apply. Explain the line you drew.

C4. Catch the AI tells. Write (or find) a short paragraph that has the "AI tells" — generic voice, a too-confident unsourced claim, a plausible-but-suspect citation. Then revise it into something true, specific, and yours.

C5. Confidentiality audit. List everything in your proposal process that is confidential or sensitive (your unpublished ideas, others' data, personal information), and define your rule for what may and may not go into an AI tool — and why.

Part M — Mixed and Interleaved Review

M1. ★ (Ch 23 + 24) How does the "one voice from many authors" discipline apply when AI is one of the contributors to a draft?

M2. (Ch 8 + 24) Why does AI's fabricated-citation problem make Chapter 8's citation-honesty discipline more critical, and how do you operationalize it?

M3. ★ (Ch 2 + 24) Why is AI's generic, voiceless prose a particular liability given who actually reads your proposal?

M4. (Ch 9 + 24) How can AI's devil's-advocate critique support the pitfalls-and-alternatives thinking of Chapter 9 — and what's the limit of trusting its critique?

M5. (Ch 11–12 + 24) Why is a budget exactly the kind of content you must never build from AI estimates? Connect to the verifiability principle.

M6. (Ch 15 + 24) How does checking a funder's current AI policy fit the compliance discipline of Chapter 15, and why is "assume it's allowed" dangerous?

🪞 Metacognitive check-in. Be honest about your own pull toward AI under deadline pressure. The temptation to accept the fluent draft, trust the confident citation, and ship the generic prose is strongest exactly when you're most rushed — and that's when verification matters most. Did these exercises make you more aware of when you're staying in the loop versus outsourcing judgment you're still accountable for? That awareness is the real skill. The tool is fine; the question is always whether you're still the author.