AI Training
Good communicators get more from AI. We teach the method.
Practical training for communications and marketing teams in the seven skills that decide whether AI gives you work you can put your name to.
Taught on the tools your team already has, whether ChatGPT, Claude, Copilot or Gemini.
A typical first instruction
“Draft a communications strategy for our acquisition announcement.”
What it leaves out
- Which audiences come first: staff, investors or regulators?
- What can we say before the deal closes?
- Who signs off: legal, the board or both?
- Which questions are we least ready to answer?
- What have we already promised these people?
Why training
Using AI looks easy. You type, it answers. The hard part is invisible.
Knowing what to tell it, what to give it, how to check it and how far to trust it. Teams that learn these skills get far better work from the same tools. Teams that skip them get confident, plausible mistakes, with your name on them.
The seven skills, in the order the work happens.
Before you start, in the environment AI works in, and once the work comes back.
Part one
Before the work
What you ask for, and what you forgot to say.
Briefing and specification
Saying exactly what you want, why it matters and what good looks like. The clarity of the brief sets the ceiling on the output.
Requirements discovery
Finding what your instructions leave out, from audience and approvals to legal limits, before the AI starts guessing.
Part two
The working environment
What AI knows, the words it uses and the steps it follows.
Context design
Deciding what AI knows and which rules it follows: style guides, approved facts, past work and house rules, organised so it can use them. Often called context engineering. It matters as much as the model you choose.
Domain language
One agreed vocabulary for your organisation, used by the team, the AI and the audience. Loose words produce loose work, and loose public answers about you.
Workflow design
Turning recurring work such as coverage reports, monitoring and briefing notes into repeatable steps AI can follow reliably.
Part three
Checking and control
What to trust, and how far.
Verification and quality control
Knowing what to trust, how to test it and how to stop the same mistake happening twice. Invented quotes, wrong figures and made-up sources are caught by process, not luck.
Agent supervision
Reviewing AI work, coordinating several AI agents at once and giving them more freedom only as they earn it.
The trust ladder
Trust is earned, one step at a time.
Most teams are on the first step. We find where yours is and train for the next one, not the one after.
Most teams start here
Step one
Assisted
AI drafts and summarises.
A person checks every line.
Step two
Repeatable
AI follows agreed workflows with set inputs.
A person reviews each output against checks.
Step three
Connected
AI works from your own files, style rules and approved facts.
A person signs off what goes out.
Step four
Supervised agents
Several agents run longer tasks in parallel.
A person sets goals, reviews plans and approves results.
Shaped around your team, not a syllabus.
Every programme starts with a short discovery call. Then we agree three things.
Who
The people in the room
Leaders, a whole team or one function: corporate communications, media relations, brand, content or marketing operations.
Where you start
Your step on the ladder
From occasional chat use to running workflows. We assess this first, so nobody sits through what they already know.
How
The format
A single session, a workshop or a short programme, in person or remote, with follow-up afterwards.
Formats
Use one, or combine them into a programme.
- Leadership briefing
- What has changed, the main risks and the decisions leaders need to make.
- Team workshop
- Hands-on training on your own kind of work: research, planning, content and review.
- Role-specific training
- Focused sessions for one function, built on its everyday tasks.
- Follow-up clinics
- Short sessions after training to review real use and correct weak habits.
What your team keeps.
A training day alone does not make AI use safe. That comes from practice and checks, which is why we offer follow-up clinics.
- Exercises built on your own work, not generic examples
- A short set of reference notes we keep up to date
- A verification checklist for your kind of work
- A first context pack: the style rules and approved facts your AI work should start from
Taught from practice
We run Adelbury this way.
Our research, articles and website are produced with AI agents working from written instructions, style rules, approved facts and approval steps.
Our UK political podcast tracker was built by one AI agent and checked by another.
What we teach is what we use every week.
See Adelbury Labs
Taught by Stephen Davies, not a slide deck.
Founder of Adelbury. More than 20 years in communications, at Edelman, Weber Shandwick and PwC. To talk about training for your team, email hello@adelbury.com