AcademyLv 1 · 0AcademyLv 1 · 0Anatomy of a great conversation
12 min · up to 60 XP
Why AI conversations go stale, and the simple structure that keeps them useful.
Priya's colleague Ben has watched her get uncanny results from the AI assistant, so he tries it himself. He types "write a job ad", hits enter, and gets back the most average job ad in history: dynamic team, fast-paced environment, passion for excellence.
He swivels his chair round: "It's rubbish. Yours must be a better version." Same model. Same everything. So what actually went wrong?
Check your intuition
Before Priya says a word, you diagnose it. Why did "Write a job ad for our open role" produce clichéd mush?
Here's the mental model that fixes almost everything: treat the AI like briefing a bright new colleague on their first day. They're sharp. They write beautifully. They've read practically everything ever published. And they know absolutely nothing about your company, your customers, or your boss's pet hates, unless you tell them.
You'd never say "write a job ad" to a new hire and walk off. You'd brief them. That briefing fits in your opening message, and two levers do most of the work. Try them.
Play with it
Priya's about to ask the AI for help with a tricky supplier. Flip the two levers of her opening message and watch what the reply becomes.
Try every combination · 1/4 seen
No context, no goal. The model fills every gap with averages: polite, fluent, useless. "Generic tips for difficult conversations…"
Watch it play out
Ben's second attempt, sitting with Priya. Watch the conversation start badly, then watch the repair.
Write a job ad for our open role.
The lazy one-shot. No company, no role, no tone, no length. The model has nothing to work with but averages.
So what goes in a good opening message? Three things: context: who you are, what this is for, what the model can't possibly know. Task: what you actually want it to produce. Constraints: length, tone, format, audience, anything with a hard edge.
No special syntax, no "prompt engineering". Write it the way you'd brief that bright new colleague: two plain sentences of background, a clear ask, the boundaries. Thirty seconds of typing that routinely saves three rounds of disappointment.
Build the prompt
Priya has a tense call tomorrow with a packaging supplier who keeps missing delivery dates. She wants the AI to draft her talking points. Build the strongest opening message: one choice for each part.
1 · Context: what the model can't know
2 · Task: what you want produced
3 · Constraints: the hard edges
A conversation has momentum. Every reply builds on everything said so far: your context, its drafts, your corrections. That's why follow-ups feel effortless: you never repeat the brief, you just steer. Say "shorter", "more formal", "now as a table". The conversation itself is doing half the work.
Predict the response
Later, the AI has drafted Priya's follow-up email to the supplier. It reads well. Before sending, she tries Ben's new favourite move.
What did you assume in this email that I didn't actually tell you?
What do you think the AI will do?
One last skill: knowing when a conversation has run its course. Momentum is a gift, until it isn't. That supplier chat that drifted into holiday cover, then into rewriting the team rota? Everything said earlier is still in the room, quietly colouring every new answer. Long, meandering chats go stale.
The fix costs nothing. New topic, new chat. And if a long conversation starts feeling muddled (answers dragging in things you've moved past), start fresh and paste in only the bits that still matter.
Fill the gaps
Lock in the conversation habits. Complete the sentences:
Key takeaways