They Can Tell You Used AI
"Write me a follow-up email to marketing asking them how they want to apply the funnel analytics."
I've seen some version of that prompt a lot. It's a reasonable request. Marketing wants to talk through funnel analytics for a new signup flow, and the engineer wants to know what they're after before the meeting.
The email that comes back usually reads something like this:
Hi team,
I hope this message finds you well. I wanted to follow up regarding the funnel analytics initiative. To ensure we're aligned, could you share more information on how you'd like to apply funnel analytics? Understanding your goals will help us deliver the most effective solution.
Looking forward to your thoughts!
Marketing can tell this was written with AI almost instantly. They will skim it, see vague information, and an open-ended question. If you get a response, it will be just as vague:
Happy to chat! Can you send over a few times?
You wanted input. Instead, you got another meeting.
People tend to put about as much effort into a reply as they think you put into the ask. AI makes it easy to send a polished message without saying what you actually think. Getting a useful answer back means putting your perspective into the prompt and keeping your voice.
People Can See Your Effort
A generic email usually feels like it hits the mark for the sender. You know what you meant, so you don't notice what's missing.
Peter Cardon and Anthony Coman surveyed about 1,100 professionals about a congratulatory email from a manager, written with different amounts of AI help. When respondents were told the manager used minimal to no AI, 83% saw them as sincere. When they were told the manager used a lot of AI, that fell to as low as 40%.
People in the study also rated their own AI use more generously than other people's. So an AI-drafted email from you can look fine but is perceived as low-effort by the reader.
Start With a Context Dump
In Why We're Writing More and Saying Less, I advocated for having a clear direction before you prompt. That works when you already have a point. With cross-functional or complex tasks, you often have a pile of half-formed thoughts. Usually the answer is already in your head. You just need to get it out.
Here's how the marketing prompt changes as you add more of your thinking.
The task alone:
Write me a follow-up email to marketing asking them how they want to apply the funnel analytics.
The task with your position:
Marketing wants to talk through the funnel analytics for the new signup flow. I think it should focus on each field, but I want to make sure we're not over-collecting information. Draft a short email to send before the meeting.
The context dump:
Goal: get marketing and me on the same page about what we're measuring in the new signup flow before Thursday.
They want funnel analytics on the signup flow. I think it should be per field so we can see where people quit. Not sure if they actually want that or just overall conversion, which is already in the dashboard. I don't want us tracking what people type into the fields, just whether they finished each one. Didn't legal flag phone numbers last time? Company size is optional, and I don't think anyone really fills it out; it might be a reason to cut it. Want to know what they'd actually change based on the data. Don't want to instrument fields we might drop next quarter.
The third version uses a context dump or stream of consciousness to provide enough information. Start with a goal, then toss in everything you're thinking, messy and incomplete. The half-formed parts, like the question about legal or the hunch about company size, are details AI can't generally assume. They're also usually left off of prompts.
Funnel analytics measures how many people make it through each step of a flow, like a signup form, and where they drop off. If you haven't worked with it, Metabase has a useful walkthrough of funnel analysis. |
Doing your own thinking before bringing in AI may also keep you more engaged in the work. Researchers at the MIT Media Lab had people write essays using ChatGPT, a search engine, or no tools at all, while measuring their brain activity. Engagement dropped as people relied more on outside tools, and most of the ChatGPT group couldn't accurately quote their own essay minutes after writing it. In a later session, people who wrote the content first and then used ChatGPT to rewrite showed higher engagement than the group that used it from the start.
|
|
Where can AI save you time?
My friends at Big Creek Growth put together a quick survey to spot the repetitive work you can hand off to automation.
|
|
Use AI to Organize and Find Gaps
After you've written the context dump, use follow-up prompts to sort through it:
- Ask AI to turn the context dump into the two or three questions you need marketing to answer. Everything else is background you can leave out of the email.
- Ask what you're missing and where marketing is likely to push back. It might point out that they'll want to compare the new flow against the old one, which changes what you need to track.
- Ask it to pressure test your assumptions. If your assumptions hold up, you can be more confident in your plan. If they don't, you can change the plan before reaching out.
AI can lay out different ways to scope the work. You still have to pick one and explain your reasoning.
A Microsoft Research and Carnegie Mellon survey of 319 knowledge workers found that the more people trusted AI with a task, the less critical thinking they reported doing. People who were more confident in their own ability did more of it.
Writing down your position first gives you something to check AI's suggestions against. If you're not sure whether a task should go to AI at all, the three questions in When Not to Use AI are a good place to start.
Send It in Your Own Words
Putting your thinking into the prompt doesn't help much if the email sounds like every other AI-written message. AI tends to rewrite everything in the same polite, generic tone. A few ways to keep the content authentic:
- Ask for suggested edits instead of a new draft. You get more control over the tone and voice.
- Tell AI what you would say instead when lines don't feel like your own writing.
- Before you send, add in small details and context only you would know to avoid the content feeling generic.
This is what the email to marketing looks like when we put everything together:
Hi Jordan,
Before Thursday, I want to make sure we have a clear direction for the new signup flow. I was thinking we track each field so we can see exactly where people are dropping off. I'd record whether each field was completed, but ensure we don't track any PII in our analytics tool. I know legal flagged a few items last time we talked about tracking signup abandonment.
Are there any fields we don't need to track as part of the funnel? For example, do we still need company size? If nobody's filling it out, that might be easy to drop.
Marketing can answer most of that in a reply. They know what you think, what you don't want to track, and what you need from them.
Looking back at the context dump, most of that email is already thought through: the per-field tracking, leaving out what people type, the hunch about company size. AI tightened the wording, but the critical thinking came from you.
Sometimes you'll write the context dump, read it back, and realize you don't need AI to finish the email at all. Next time you open an AI tool to send a message, try writing the context dump first and see how much is left for it to do.