The AI Operating Brief: the 20-minute fix for generic Copilot

You’re reading The AI Directive — practical AI intelligence for
UK business leaders. Subscribe here if
someone forwarded this.

Most people treat AI like a brilliant new starter who receives no
induction.

No background. No examples. No preferences. No idea who the audience
is. No definition of what good looks like. No rules for when to ask a
question rather than charging ahead with a confident first draft.

Then we complain that the output is generic.

Fair, up to a point. But also slightly harsh on the machine.

If you asked a human colleague to write a board update with no
context, they would either ask ten questions or produce something that
sounds like it escaped from a consultancy template folder. AI tends to
do the second one, because it is trying to be helpful and has no social
embarrassment about producing beige prose at speed.

Last week I wrote about voice profiles: giving AI evidence of how you
think and write, instead of asking it to guess. That is still the first
move.

An AI Operating Brief is what you build around that
voice profile.

Not instead of it. Around it.

The voice profile says: this is how I think, write and make judgement
calls.

The operating brief says: and this is the job, the audience, the
structure, the source material, the boundaries and the point where you
stop and ask.

An operating brief is not a clever prompt. It is not a magic phrase.
It is a short working document that tells an AI assistant how to work
with you repeatedly.

It answers questions like:

That last one matters. AI adoption gets sloppy when people only think
about output quality. The better question is operational: what should
this thing be allowed to do, and under what conditions?

For most leaders, the win is not writing longer prompts. It is
reducing how often you have to explain the same working preferences from
scratch.

A good operating brief turns AI from a vending machine with grammar
into something closer to a junior colleague who has at least read the
induction pack.

Still needs review, obviously. I am not suggesting you leave it
unattended with the board papers and a LinkedIn account. We have enough
problems.

But it changes the starting point.

Instead of this:

You can say:

That is a different conversation.

The AI now has a frame for the work. It knows the audience, the voice
profile, the structure, the judgement style and the escalation rules. It
is not guessing what “good” means from the entire internet, which is not
exactly short of bad examples.

The operating brief also makes AI safer inside organisations.

It lets you encode rules like: do not invent facts; separate evidence
from opinion; use approved source material only; ask before making
external claims; flag sensitive content; never send, publish or change
records without explicit approval.

Those are boring instructions.

Good.

Boring is underrated when systems are near customer data, board
decisions, suppliers, HR processes or anything with a login button.

The companies that get value from AI will not just be the ones with
the best tools. They will be the ones that teach the tools how work
should be done.

If you want to build an AI Operating Brief, do not overcomplicate
it.

Start with five sections.

1. Role and audience

Who are you, what sort of work do you do, and who is the output
usually for?

For example: “I lead transformation in a UK SME. Most outputs are for
directors, operational managers or suppliers. Assume the reader is busy,
commercially literate and allergic to waffle.”

That one sentence already improves the answer.

2. Voice profile and judgement style

This is not a second attempt at last week’s exercise.

Use the voice profile you created from real writing samples. Do not
replace it with a fresh list of adjectives. That is how you end up with
two slightly different versions of “your style”, which is just another
route back to generic mush.

The operating brief should point to the voice profile and then add
the missing working rules around it.

For example:

Then add the judgement rules that are not purely about tone:

That is the link between the two pieces.

The voice profile tells the assistant how the work should sound and
feel.

The operating brief tells it how the work should behave.

3. Preferred output structure

Tell it how you like work packaged.

Do you want the recommendation first? Do you want risks separated? Do
you prefer bullets or short paragraphs? Should it include options?
Should it mark blockers clearly?

This is where a lot of “AI sounds wrong” complaints actually come
from. The voice may be closer after last week’s exercise, but the
structure still needs to match how the organisation thinks.

4. Decision rules and escalation points

This is the safety section.

Write down what the assistant can do freely, what it can draft but
not send, and what needs human approval.

Examples:

That might sound obvious.

It is obvious right up until someone wires an agent into the wrong
workflow and discovers that “helpful” is not the same thing as
“authorised”.

5. Clarification rules

Tell the assistant when to stop and ask.

For example:

That is one of the most useful instructions you can give.

It prevents both extremes: the assistant that asks permission to
breathe, and the assistant that confidently remodels the kitchen because
you asked it to tidy up.

The video version of this idea is coming shortly on the channel.

For now, the related video is here:

Don’t Lose Your Voice
to AI

That covers the voice profile piece. This issue should sit directly
on top of it: same voice evidence, now wrapped in role, audience,
structure, source and safety rules.

1. Do not paste confidential examples into public AI
tools just to build a better profile. Use approved tools, sanitised
examples, or internal platforms.

2. A team operating brief should not erase
individual voice. It should decide which voice profile or house style
applies, then add shared rules for structure, evidence, review and
approval.

3. The best prompt library is not a folder of 200
magic prompts. It is a small set of reusable working briefs for the
decisions, documents and workflows that actually matter.

“Is this just prompt engineering?”

— Reasonable sceptic

No. Or at least, not in the usual internet sense.

A prompt is the instruction for this task.

An operating brief is reusable context for how work should generally
be done.

The difference matters. Prompt engineering often turns into hunting
for the perfect incantation. An operating brief is closer to onboarding:
role, audience, standards, boundaries and escalation rules.

One is a trick.

The other is an operating model.

If you only do one thing this week, write a one-page operating brief
for the AI work you repeat most often.

Not the whole organisation. Not the grand AI strategy. One recurring
task.

Project updates. Meeting summaries. Supplier replies. Risk notes.
Board papers. Customer responses.

Pick one. Write the operating brief. Test it twice. Improve it.

That is how this becomes practical.

Next week: choosing the right Microsoft AI surface — Copilot, Agent
Builder or Copilot Studio — without turning it into a theological
argument.

See you Tuesday.

— Toby

@TobySmithFirth on
YouTube

These are tools I use personally. Affiliate links marked — I earn
a small commission if you sign up, at no extra cost to you.

Some links in this issue are affiliate links. I only recommend
tools I actually use.