Prompt literacy: the teachable core
The gap between disappointment and delight is usually four elements the user didn't supply. Teach the frame relentlessly:
Goal + Context + Source + Format. 'Summarise' vs 'Summarise the attached contract's termination clauses (goal) for a non-lawyer sales lead (context/audience), using only this document (source), as five bullets with clause references (format)'. Same engine, different decade of output. Add the two behaviours that compound: iterate in-thread (refine, don't restart — context accumulates) and reference explicitly ('/'-mention files, name the meeting) rather than hoping retrieval guesses (semantic-index concept: help the pipeline help you).
Anti-patterns to name in training, because naming kills them: the one-word prompt, the twelve-asks-in-one prompt, trusting without opening citations (foundations: citations are the audit handle — teach clicking them as hygiene, which doubles as your injection-awareness layer).
The adoption engine (what actually moves the weekly-active number)
Champions per role family (the pilot cohort's enthusiasts, given a channel and a rotation of demos), role-specific prompt libraries (fifteen prompts that matter for finance ≠ fifteen for field service — generic galleries convert nobody), the weekly drumbeat (one prompt, one use case, ninety seconds — in the channel people already read), manager modelling (the single strongest correlate: teams whose lead uses recap in the Monday meeting adopt), and the feedback loop: every 'it can't do X' ticket becomes either a capability-table update (apps concept), a literacy example, or a governance finding — triage them like the signal they are.
What to watch (proofs)
- Prompt-quality proxy: average conversation depth (iterations per session) from analytics — deepening threads = literacy landing; one-shot-and-leave = the frame hasn't.
- Library usage: which role prompts get copied (where instrumented) — the curriculum's own hit parade, feeding the next training wave.
- Champion coverage: role families with an active champion vs without, overlaid on utilisation — the correlation that funds the programme.
- Ticket triage yield: monthly count of tickets converted into capability-notes/training/governance items — adoption's flywheel, measured (and the local edition of the what-users-ask page writing itself).