Many people approach AI tools as one-shot interactions. They write a prompt, receive output, and move on. Either they use what they got or they start over with a different prompt. This approach leaves significant quality on the table.
The most effective approach is iterative. You generate initial output, evaluate it critically, then have a conversation with the AI tool to refine it. You give feedback on specific issues, request adjustments to tone or emphasis, combine sections from different generations, or pivot the approach entirely. Through this iterative cycle, you move from acceptable output to excellent output.
This lesson teaches you how to have productive conversations with AI tools. You'll learn the iterative refinement cycle, what kinds of follow-up prompts work best, when to keep refining versus when to start over, and how to combine multiple pieces of output into something better than the sum of its parts.
The basic iterative cycle has five steps:
Using a well-engineered prompt (like the CRAFT templates from previous lessons), generate the first version of whatever section you're working on. Don't expect perfection. Your goal is to produce something that captures the core idea and that you can build on.
Read the output carefully. What's working? What's not? Is the tone right? Is the detail level appropriate? Does it match your voice and strategy? Do the facts align with what you know? Identify 2-3 specific things you want to improve or change. Don't try to fix everything at once.
Give the AI tool specific feedback about what you want to change. Not "this doesn't sound right" but "this paragraph uses too much jargon—it should be more accessible to donors without nonprofit experience." Not "I want more detail" but "add one specific example of how our mentoring relationship changes lives." Specificity drives better refinement.
Based on your feedback, the tool regenerates the output with your requested changes. Review the revised version. Is it better? Does the change address your concern? Are there new issues or unintended consequences of the change?
If the revision improves things and you're moving toward your ideal, continue iterating (return to Step 2). If the revision doesn't help or makes things worse, consider starting over with a different approach (return to Step 1 with a revised prompt). If the section is now satisfactory, move on to the next section.
This cycle might repeat 2-3 times for a section, or occasionally more for complex sections. The goal is not to generate perfect output on the first try, but to collaborate with the AI tool through refinement toward output that's genuinely good.
Not all follow-up prompts are equally effective. The most useful follow-up prompts are specific, bounded, and actionable. Here are categories of follow-up prompts and examples:
Example: "This is too academic. Make it more accessible to donors without nonprofit experience. Remove jargon and use simpler language without losing depth."
Why it works: Specific about what's wrong (academic), clear about the target audience (donors without nonprofit experience), and precise about the solution (simpler language, accessible, not dumbed-down).
Better than: "Make this sound less formal" (too vague) or "Rewrite this completely" (throws away potentially useful content).
Example: "This needs to emphasize impact over process. Right now, most of the paragraph describes what we do; shift it to focus on what changes as a result. Cut the activity description from 40% of the paragraph to 20%, and expand the outcomes from 20% to 50%."
Why it works: Specific about what's currently there, clear about what you want instead, and includes a rough ratio to guide the adjustment.
Better than: "Make this more impactful" (unclear what needs to change) or "Rewrite focusing on impact" (might throw out useful activity description).
Example: "This is too detailed about our staffing structure. We need only two sentences on staffing, and they should focus on qualifications relevant to this program, not organizational structure. Cut it from 5 sentences to 2."
Why it works: Clear about what's wrong (too much staffing detail), specific about the target (two sentences, focus on qualifications), and includes the specific edit instruction.
Better than: "This is too long" (doesn't specify what to cut) or "Cut this section by half" (arbitrary percentage without guidance on what matters).
Example: "The third paragraph says we serve 120 participants annually, but our actual number is 95. Also, it says our program runs for 16 weeks, but it's 12 weeks. Please correct these numbers. Keep the rest of the paragraph as is."
Why it works: Identifies the specific errors, provides the correct information, and instructs not to change anything else.
Better than: "These numbers are wrong" (doesn't specify what's wrong) or "Rewrite this paragraph with correct facts" (might change things you liked).
Example: "I like the first three paragraphs of the first version you generated, but I prefer the final two paragraphs from the second version you generated. Can you combine those together to create a single version? Make sure the transition between the third and fourth paragraph flows smoothly."
Why it works: Specific about which parts you want from each version, and includes guidance about making the combination flow well.
Better than: "Combine these two versions" (doesn't specify which parts) or "Create a version that uses the best of both" (too vague about what's best).
Example: "This needs one specific example showing how a participant's life changed through our program. Add a brief example (2-3 sentences) into the second paragraph about outcomes. Use our client Maria, who moved from homelessness to stable housing within 18 months and is now working full-time."
Why it works: Clear about what's missing (specific example), where it should go (second paragraph), how long it should be (2-3 sentences), and what example to use.
Better than: "Add more examples" (too vague) or "Include a case example" (doesn't specify which example or where).
Example: "This approach isn't working for me. Instead of organizing the needs statement geographically, let's organize it by population group. The problem statement should still be the same, but the evidence section should show how this problem affects [Group A], then [Group B], then [Group C]. The urgency section should emphasize how each group is affected differently."
Why it works: Acknowledges that the current approach isn't working, proposes a clear alternative, and explains how the new structure will work.
Better than: "I don't like this" (doesn't propose an alternative) or "Try again" (doesn't specify what to try differently).
The best follow-up prompts are specific about what you want to change, explain why the current version isn't working, and guide the revision without dictating every word. They acknowledge the AI tool's work and build on it rather than dismissing it. When you follow this pattern, refinements almost always improve the output.
Let's walk through a realistic example showing how the iterative cycle works with a specific grant section.
Generate an initial needs statement using my CRAFT prompt about our youth workforce program. [Initial CRAFT prompt provided]
"Youth unemployment is a growing concern in our community. Young adults aged 18-24 face barriers to employment including lack of job training, limited work experience, and disconnection from the labor market. Our city's youth unemployment rate is 14%, compared to 6% for adults. This difference represents a significant barrier to economic mobility for young people. Our organization has worked with justice-involved youth for 8 years and understands the specific challenges they face. We propose a 12-week intensive job training program targeting 40 participants annually to address this need..."
This is a good start, but it doesn't create enough urgency. Right now it reads like a problem statement, not a compelling needs statement. The second paragraph should emphasize what happens to young people who don't get job training—what are the long-term consequences? Also, the third paragraph should come earlier; position our organization's experience right after the problem statement, not at the end. Can you revise with more emphasis on consequences and reorder the organization information?
"Youth disconnection from the labor market creates a cascade of consequences that affect entire lives. Young adults aged 18-24 who don't secure stable employment are at increased risk for housing instability, criminal involvement, and long-term economic marginalization. Studies show young people without job training are 4x more likely to experience poverty in their 30s. Our city's youth unemployment rate is 14%, compared to 6% for adults—a disparity that signals systemic barriers rather than individual failure. Our organization has worked with justice-involved youth for 8 years, building trusted relationships and understanding the specific barriers they face. We've achieved 70% job placement rates among our participants, demonstrating that targeted support changes outcomes. Yet only 40 young people per year access our program—far below the estimated 400+ who could benefit from intensive support. This gap between need and current service capacity is the problem we're addressing..."
Much better! This has urgency now. One adjustment: the stat "4x more likely to experience poverty in their 30s" feels disconnected from our program. Can you remove that sentence and instead add something about the school district's dropout rate or another local stat that connects directly to the employment challenge? Also, the final sentence about "gap between need and current service capacity" is good but it reads like a program description. Can you rephrase it to stay in the problem/need space without moving into what we're proposing to do?
"Youth disconnection from the labor market creates a cascade of consequences that affect entire lives. Young adults aged 18-24 who don't secure stable employment are at increased risk for housing instability, criminal involvement, and long-term economic marginalization. According to our school district, 28% of students who enter high school don't graduate—a critical gateway point where disconnection begins. Our city's youth unemployment rate is 14%, compared to 6% for adults—a disparity that signals systemic barriers rather than individual failure. Our organization has worked with justice-involved youth for 8 years, building trusted relationships and understanding the specific barriers they face. We've achieved 70% job placement rates among our participants, demonstrating that targeted support changes outcomes. Yet we know that hundreds of young people in our community lack access to the kind of intensive support that changes employment trajectories. The gap between the number of disconnected youth and available support services represents an urgent need..."
This is solid. Two small tweaks: (1) Change "We've achieved 70% job placement rates" to "We've achieved 70% job placement rates for participants ages 18-22"—we need to be precise about which ages we're serving. (2) In the final sentence, "employment trajectories" is a bit jargony—can you use simpler language? Maybe "meaningful jobs" or "stable careers"? After these two changes, this will be ready to use.
"Youth disconnection from the labor market creates a cascade of consequences that affect entire lives. Young adults aged 18-24 who don't secure stable employment are at increased risk for housing instability, criminal involvement, and long-term economic marginalization. According to our school district, 28% of students who enter high school don't graduate—a critical gateway point where disconnection begins. Our city's youth unemployment rate is 14%, compared to 6% for adults—a disparity that signals systemic barriers rather than individual failure. Our organization has worked with justice-involved youth for 8 years, building trusted relationships and understanding the specific barriers they face. We've achieved 70% job placement rates for participants ages 18-22, demonstrating that targeted support changes outcomes. Yet we know that hundreds of young people in our community lack access to the kind of intensive support that leads to meaningful jobs and stable careers. The gap between the number of disconnected youth and available support services represents an urgent need..."
Perfect. This version is ready to use in the proposal. Saving this prompt and final output to my library.
Notice what happened in this example: The user didn't reject the AI output or accept it unchanged. Instead, they engaged in a three-iteration conversation that moved the output from "good starting point" to "ready to submit." Each iteration was specific about what to improve and why. The final version is significantly better than what the first generation produced.
Take a grant section you generated in earlier lessons. Evaluate it critically using these questions: Is the tone right for my funder? Does it have enough detail in the right places? Does it emphasize what matters most? Are all facts accurate? Write a follow-up prompt using one of the patterns above to improve the section. Generate the revised version. If the revision improves things, try one more iteration of refinement. Document how the output improved with each iteration. Notice that iteration produces better results than one-shot generation.
Keep iterating when: The output is on the right track but needs adjustment. The core approach is sound, you just need to refine tone, emphasis, detail, or specific facts. Each iteration is noticeably improving the output. You're making progress toward your vision.
Start over when: The fundamental approach isn't working for you. The output is structured wrong and you'd need to change more than half of it. You realize the prompt itself missed something important. You've iterated 4+ times and still aren't getting close to what you want. The output reveals that your thinking about the section isn't clear.
Sometimes starting over with a revised prompt is faster than continuing to refine output that's fundamentally off track. The decision rule: "Am I refining or rewriting?" If you're rewriting, start over. If you're refining, keep going.
Many grant proposals require multiple sections working together. A strong approach is to iterate each section independently until it's excellent, then review how they work together. You might discover that one section is perfectly pitched while another feels tonal y disconnected. You can then iterate to create consistency without redoing everything.
Another powerful move: generate multiple versions of challenging sections and keep the best elements from each. You might use the logic model from Version A, the outcomes from Version B, and the activities description from Version C—combining them into a stronger final version than any single generation produced.
Iterative refinement is a skill that improves with practice. Your first few iterations might feel awkward. You might not be sure what to ask for or how to describe what's wrong. This is normal. After 10-15 iterations with AI tools, you'll develop intuition about:
This intuition makes you dramatically more efficient at using AI tools for grant writing. What takes someone new 30 minutes of refinement takes an experienced user 10 minutes because they know exactly what to ask for and what's worth iterating on.
Iteration is not endless polish. At some point, you need to make a decision and move forward. The law of diminishing returns applies: after a certain number of iterations, additional refinements produce minimal improvement and consume time you could use elsewhere. Know when "good enough" means it's time to move on. Grant writing is about deadlines and submissions, not perfection.
As you become skilled at iterative refinement, consider documenting what you learn. Keep notes about:
These notes become a personal guide to iteration efficiency. Over time, you'll develop patterns that work consistently for your organization's voice, your funder relationships, and your program types.
The most effective relationship with AI tools is collaborative conversation, not command-and-control. You're working together toward a shared goal: excellent grant content. The AI tool brings writing capability, speed, and the ability to regenerate sections quickly. You bring strategy, judgment, knowledge of your organization and funder, and the ability to recognize when something's not quite right.
This collaborative mindset leads to better outcomes than either approaching the tool as a vendor ("I'll tell you what I want and you'll deliver it") or dismissing AI output as inherently inferior ("This AI tool can't do real grant writing"). The sweet spot is in the middle: use the tool's capabilities, refine through iteration, and make final judgment calls based on your professional expertise.
In Lesson 3.7, the final lesson, you'll learn how to organize and maintain a prompt library that becomes your personal competitive advantage.
Continue to Lesson 3.7