Still looking for the perfect prompt—or having a conversation?
By Ryan Clement · Published
You ask AI to write an email. You read the answer. Then you decide whether AI is good or bad. Sound familiar? A more useful question might be: did I give it enough to work with—and did I do anything with the first response? Getting better at AI doesn't have to mean chasing every new feature. Start by looking at how you work with it.
1. Are you looking for the perfect prompt—or having a conversation?
A first response can give you something to react to. Explain the audience, the situation, and the outcome you want. Read what comes back. Then ask a follow-up that addresses an actual gap.
Try: What assumption are you making? What information is missing? Give me another approach. Under what conditions would this strategy fail?
The goal isn't a longer conversation for its own sake. It's a more useful result. If you keep going in circles, stop and reconsider the task, your source material, or whether AI is the right tool.
2. Are you giving AI tasks—or explaining problems?
Rewriting an email is a perfectly reasonable use of AI. But sometimes the wording isn't the real problem. Maybe the audience is unclear, the request is too broad, or nobody has agreed on the next step.
MIT Sloan's September 2026 article on worker performance describes research involving more than 20 companies. Its practical lesson: define the problem and evaluate results before expanding AI use. Time saved alone can miss quality problems and extra review work. Applying that lesson to nonprofit work, I would start with the decision your team needs to make.
Here's a fictional nonprofit example: a small team has more event follow-up to do than staff time available. Instead of asking AI to write a generic thank-you, explain that you need a manageable follow-up process. Use invented attendee scenarios and ask it to compare possible approaches, identify missing information, and suggest what staff should review.
That gives you options to evaluate. It doesn't give the tool authority to decide who deserves attention or infer a person's giving capacity from event attendance.
3. Are you explaining what good looks like?
If you have a standard in mind, say what it is. Include the task, why it matters, the audience, relevant context, and the constraints. An example of an acceptable result can help make those expectations concrete.
For that fictional event team, you might say: Compare two follow-up approaches for a team with three staff hours available. Use only the invented scenarios below. List the tradeoffs and unanswered questions. Don't infer wealth or invent relationships. End with a checklist a staff member can review.
You now have something to check the answer against. A polished paragraph is easier to assess when you know what it was supposed to accomplish.
4. Do you challenge the answer?
A confident answer can still be wrong. Ask the tool to distinguish supported facts from suggestions and assumptions. Then check the important claims against the underlying sources yourself.
Asking AI to critique its own answer can surface questions worth investigating. It isn't independent verification. A second response can repeat the same error, and a citation needs to be opened and checked to see whether it actually supports the claim.
For prospect research, that means checking identity, dates, sources, and the difference between evidence and inference before anything becomes part of a usable profile. The Follow-Up Prompts for Prospect Research linked below can help structure that review.
5. Are you using AI to save time—or also to examine your thinking?
Saving time is useful. So is noticing a question you hadn't considered. You can ask AI to compare options, suggest objections, or identify what additional information would change a decision.
The same MIT Sloan article cautions that delegating a task to AI can leave workers without the understanding they would have gained by doing it. That's a useful check on a workflow: can you explain and defend the result yourself?
In the fictional event example, ask: What could make each approach impractical for a small team? What would we need to know before choosing? Then decide whether the objections make sense in the actual context.
Treat those responses as proposals to examine. More ideas aren't automatically better ideas, and agreement from a chatbot isn't evidence that your plan is sound.
6. Have you started building repeatable workflows?
When an approach proves useful, save more than the prompt. Save the purpose, permitted inputs, expected output, review steps, and the point where a person makes the decision.
A simple research workflow might be: define the question, gather approved source material, prepare a draft, verify consequential claims, and record what remains unknown. The Research Profile Prompt Guide linked below is a resource you can use alongside your own review standards.
Templates and saved workspaces can reduce repeated setup. They still need maintenance when your sources, tools, or organizational requirements change. A workflow you trust should make review easier to do, not easier to skip.
7. Do you know when not to use AI?
Sometimes a spreadsheet, a conversation with a colleague, or an existing process is the better choice. Sometimes the information is too sensitive for the tool you're considering. Sometimes you don't have a reliable way to check the result.
Before using organizational information, check your organization's rules and whether the tool and proposed use are approved. Removing names alone may not make a record anonymous; combinations of details can still identify someone. Use fictional examples when practicing.
My Should This Be AI? tool is a starting point for thinking through that choice. Deciding to leave AI out of a task can be a sign of good judgment.
Look at your progress, not your ranking
You probably don't need to know whether you're in the top 5%. Look at how you were using AI six months ago. Are you providing better context? Asking more useful questions? Checking what comes back? Have you built a few workflows you can review and improve?
And, most importantly, are you still responsible for the final judgment?
If the answer is increasingly yes, you're progressing. You don't need to chase every new model or feature to become better at AI. Pick one recurring task this week, define what a good result looks like, and add a deliberate review step.
The tool matters. But the thinking behind the tool matters more.