From Generating to Deciding
Most teams stop at "we tried ChatGPT". We train yours to build assistants that hold a role, follow a structure, and produce work you can hand on without rewriting it.
Book a training consultationWhy the Answers Are Average
Garbage In, Garbage Out
A vague instruction gets a vague answer. If you cannot state precisely what you need, the model will give you something that sounds right and says nothing.
Talking to it like a search box
Keywords worked on Google. A model needs structure — a role, a task, the context, the shape of the output.
No role assigned
Without a persona the model answers as nobody in particular. Give it a role and it can take a position, and change it when you ask.
The Craft of Asking
The PTCF prompt structure
Persona, Task, Context, Format. Four slots that turn "write me something about X" into an instruction with only one reasonable reading.
Chain-of-thought reasoning
Make the model show its working. For decisions with several moving parts, the reasoning is the part you need to check — not the conclusion.
Knowledge bases and RAG
Turn a thousand pages of internal documents into something you can ask questions of, with answers that cite the page they came from.
Many Roles, One Assistant
Structured intake
Used in: customer service, order handling
"Turn how a customer actually talks into a clean, structured record — the way a good order-taker does it."
Simulated review meeting
Used in: project management, risk assessment
"Have the model argue a project risk as the PM, the site manager and the QA lead in turn, and cross-examine each."
Drafting to a house style
Used in: PR, public-sector documents
"Give it the regulation and the facts; get back a release in the format the recipient expects."
Ask Well, Think Better
The gap is not access to AI — everyone has that. It is knowing how to instruct it. That part is teachable.
Start a conversation