A few AI prompts to help you actually understand your work data

I've spent a lot of time over the past five months feeding data into AI—reader survey results, click data, open rates—and the difference between a useless response and a genuinely insightful one usually comes down to how you frame the ask. Here are the prompts that have worked for me, made generic enough to use on whatever data you're staring at.

1. Pressure-test your own read on the data.

The biggest unlock for me wasn't asking AI to summarize the data—it was writing down my own takeaways first, then asking AI to check them against the numbers by uploading the raw data files to a project:

"Here are my takeaways from the data for [X project]: [paste your takeaways]. Analyze the data itself and tell me which of my conclusions it supports, which it contradicts, and which it complicates. Be specific about why, and flag anything important I haven't noticed. Before you get started, is there anything you need to know about the project itself?"

That ordering matters. If you read AI's analysis first, it'll anchor your thinking and you'll lose the chance to compare your instincts against the data cleanly.

2. Find the patterns you're not seeing.

Then, in a fresh chat (so it isn't anchored by what you've already told it), go fishing:

"Analyze this data and identify the patterns that show up consistently. Then tell me: What's surprising here? What would you expect to see that's missing? If you could only tell me three things from this data, what would they be and why?"

The "what's missing" part is the underrated bit—absence can often be more interesting than presence (or at least spark new questions).

3. Make it repeatable.

If you look at the same kind of data regularly, ask AI to help you build a reusable structure:

"Based on this analysis, create a template of questions we should analyze this data against every [week/month]. Structure it so results are comparable across time periods and we can see patterns emerging."

I did this for our click data, which I review in depth every month, and it means each analysis builds on the last instead of starting from scratch. And, again, AI is only as good as the data you give it. Garbage in, confident-sounding garbage out.

If you’re using AI to dig into data in an interesting way, or have a topic you want us to dig into, let us know at [email protected].