Let it interview you.

You ask a big question about your own life and get an answer that would fit your neighbor, your dentist and most of your graduating class. It reads like a horoscope: warm, plausible and about nobody.

Lesson three said to give the assistant your reality. But with the large, foggy questions (where can I get better, what am I avoiding, what should I do next) you often don’t know which facts matter. If you did, you might not need to ask.

So turn the interview around. Ask is the A in DREAM, and it means the assistant asks you. Before it answers, it questions you, and your answers become the reality it works from. For this kind of question, its questions are often worth more than its first answer. Good ones also make you stop and think, which has value before any advice arrives.

We didn’t invent this. Google’s prompting guide for Gemini in Workspace suggests asking the assistant what questions it has for you.

Three small rules make it work. Say “don’t answer yet,” or it may answer first and ask afterward. Ask for the questions one at a time, so each can follow from what you just said. Twelve questions in one message is a form, and you will abandon it like every other form. And set a cap, somewhere between three and eight, so the interview ends in an answer.

With memory and without

Some assistants now remember you between chats. If yours has months of your conversations to draw on, a short version can work: “From what you know about me, where could I get better as a manager? Show me the evidence, and separate what you know from what you’re guessing.” Paste that same line into a fresh chat, or into an app that has never met you, and the best outcome is an assistant that says it doesn’t know you. The other outcome is the horoscope again, delivered with confidence.

The interview version works anywhere, because it gathers its own evidence. That is why the prompts in our library that ask what an assistant sees in you come in two versions, Fresh chat and Knows you, with Fresh chat as the default.

Side by side

The weak version asks a stranger for a verdict. The strong version says so out loud (“you don’t know me”), hands over the microphone for six questions, and then asks for something specific: two changes, each tied to an answer you gave, plus one thing the assistant can’t know from a chat. That last part matters. It reminds both of you that a short interview is a sketch, not a portrait.

Now the assistant knows what you want, where you stand and what good looks like. One problem is left: it tends to tell you what you’d like to hear, and it sounds very sure while doing it.