Dictate-and-Template AI Story Scaling
Prime ChatGPT with your team's exact template, then dictate context naturally to generate an org's worth of specs.
- Difficulty
- Easy
- Time to result
- ~days to results
- Steps
- 4
- Confidence
- 85%
A concrete workflow for one IC to do work that previously required a whole director-level org: teach ChatGPT the team's exact user-story template, then use voice dictation (Whisper) to talk through the change as if briefing a developer, and let it emit fully-formatted stories. The human stays essential for the one thing AI can't do well — deciding how to logically split a huge change into stories.
Origin
Tal Raviv, applied on a large fundamental-UX-change project at Riverside using ChatGPT plus OpenAI's Whisper dictation and the team's requested Gherkin ('given/when/then') story format.
Core principles
- 01Prime the model with the exact template and role before feeding content; confirm it 'understands' before starting.
- 02Voice dictation lets you brief the model as naturally as you would a new developer — richer context than typing.
- 03AI handles the tedious formatting; humans handle the judgment (how to decompose the work).
- 04Feeding a raw meeting transcript works worse than a deliberate spoken briefing — structure your dictation.
How to run it
- 1
Prime the model with role and template
Open a fresh thread. Tell it 'you're an expert PM / product owner / scrum master,' paste the exact template your team wants (e.g. Gherkin given/when/then), and ask 'do you understand?' before proceeding.
Pro tip Get the template from the engineer/team-lead who'll consume the output so the format is exactly what they want.
- 2
Dictate rich context by voice
Using Whisper dictation (built into ChatGPT's desktop/mobile app, or a standalone app), talk through why you're doing this and the background as if onboarding a new developer, then describe the first change naturally.
Pro tip Whisper is free/open-source from OpenAI and produces high-quality transcription — talking is faster and richer than typing.
- 3
Let AI emit the formatted stories
On hitting enter, the model produces the detailed user story in the exact format with all the edge cases that would have taken you a long time to write by hand.
- 4
Keep the human on decomposition
You still need a person to decide the logical, engineering-sound way to break the huge change into individual stories. Once split, each story generates very well individually.
Watch out Feeding the raw recorded kickoff transcript worked less well than a deliberate spoken briefing — don't assume more raw input is better.
In the wild
For a fundamental UX overhaul touching all of Riverside, Raviv primed ChatGPT with the team lead's Gherkin template, dictated the changes naturally via Whisper, and generated the full set of detailed user stories. The team lead handled splitting the big change into stories.
→ Work that a decade ago would have needed a director-level org — or multiple PMs / a dedicated scrum role — got done by a single IC.
Common mistakes
Dumping the raw kickoff transcript in
Raviv tried feeding the actual recorded/transcribed kickoff into ChatGPT and it worked worse than a structured spoken briefing. Deliberate, ordered dictation beats a raw transcript.
Is it for you?
Best for
A PM or product owner facing high-volume, templated writing (user stories, epics, specs) who wants to avoid becoming the documentation bottleneck.
Not ideal for
Small or nuanced batches of stories where writing them by hand is itself a valuable thinking exercise, or teams without a stable template.
From the transcript
“I gave it to Chad gbt I said you know you're an expert PM product owner scrum Master whatever this is the template do you…”
“use whisper whisper I to dictate and I was like you know let me just tell you a little bit more background I just like…”
“the thing that we still really needed a person for was deciding how to break up this really really big change into those stories”
“experimenting with taking the transcript of the actual kickoff that we did ... I copied it and put it into chat it didn't work as…”
From the episode
Becoming a super IC: Lessons from 12 years as a PM individual contributor
Tal Raviv (Product Lead at Riverside)