How to make AI a more useful thought partner
By: Sarah Pita
Generative AI tools like Claude, ChatGPT, and Gemini, can be great thought partners. They can also be really terrible thought partners.
If you’ve ever tried talking through an idea with AI, you know what I’m talking about. These tools are designed to be engaging, to “listen” and give every appearance of hanging on your every word. So if you just start talking or typing to AI, you can quickly find yourself in what feels like a long conversation.
What you won’t necessarily get, though, is disagreement.
“Sycophancy” is the term for AI’s tendency to flatter and agree with us, in order to keep the conversation going. This tendency can be dangerous, especially if you’re asking for advice about something. Intentionally or not, you may not be sharing all the context. You may be looking for validation and agreement, and by default, often, that’s what you’ll get.
As Mike put it in a recent LinkedIn post, “Be thoughtful about how you use AI in this process. Tools like ChatGPT, Gemini, or Copilot are great for helping you expand an idea or clarify your thinking. However, they are not great for deciding whether the idea is a good one. They are trained to agree with you, not to protect your budget or your capacity.”
There are ways to make AI a better (if less comfortable) thought partner, though.
Change your settings. In both Claude and ChatGPT, in my settings, under “personal preferences”, I added the instruction “Ask clarifying questions before providing detailed answers.” This effectively frames more skeptical conversations. AI will ask you for context, and that makes AI’s voice a little less immediately validating.
Assign a perspective. Share your idea and ask “who would be likely to be skeptical of this idea, and why?”
Set the tone. You don’t need to write an elaborate prompt. When you ask for feedback, say “be brutally honest” or “don’t worry about being nice.”
Ask for a grade. Producing something, like a grant narrative, that is being graded on a rubric? Share the rubric and ask the AI to evaluate critically based on the rubric. You can even ask for a grade. Ask “How can this be better?”
Ask for the negatives. You shared this amazing idea you just had, and you’ve been happily building castles in the air. At some point, make yourself ask questions like “what am I missing?” and “what are the reasons this might not be a good idea?”
Request a pre-mortem. You’ve shared your idea. Now ask AI to imagine that it is six months, or a year, later, and the idea failed. Ask why it failed. This shifts the focus from evaluating the idea to analyzing why it might fail.
All of these techniques will help to temper that tendency toward excessive validation, so that you get responses that will help you improve instead of just making you feel good.
Even so, and this is important, don’t let AI have the final word before you let your ideas out into the world. Talk to your people. They actually do care, and if they agree, or disagree, it won’t be just because you told them to.
About the Author:
Sarah Pita is a fundraising professional with 25+ years of experience and a dynamic speaker who makes AI approachable and immediately useful for nonprofit teams. She leads practical, engaging trainings and workshops on using AI for fundraising and has presented at groups such as Women In Development NYC and at the AFP GPC Leading Philanthropy conference, among others. Sarah is currently Director of Development at the Center for Independence of the Disabled, New York.
Interested in an AI workshop or training? Contact Sarah here.