What does this prompt really need to know?

What we found in the news

In March 2023 guidance on large language models (opens in a new tab), the UK National Cyber Security Centre advised:

“not to include sensitive information in queries to public LLMs”

— National Cyber Security Centre

The advice addressed public services and their risks at that time. It is not an assessment of every present-day product or an organization’s approved environment. Its practical question remains useful: what information does this particular task actually require?

Why this matters to us

Picture a staff member preparing an invitation to a community outing. The task is to explain the destination, time, accessibility contact, and how to express interest. A participant list is open nearby, including personal circumstances that are relevant to staff planning.

Those details do not need to travel into a general invitation-writing prompt. The writer can describe the event and intended tone without supplying anyone’s private story.

At genus, we are interested in making the useful task clear enough that people do not have to guess what information belongs in it. A helpful draft should begin with a deliberate choice about its ingredients.

Write down the job before adding the context

Describe what the assistant is being asked to produce. An invitation for adults considering a museum visit is a different task from an individual support plan.

For the invitation, useful inputs might include the public event details, a preferred reading level, and the organizer’s contact route. The participant’s name, personal history, or internal notes may add nothing to that job.

A staff member could first prepare a short, approved event brief. That also gives a human reviewer something reliable to compare with the finished draft. If the destination or time is uncertain, leave it marked for confirmation rather than filling the gap with a plausible detail.

This is ordinary editorial preparation. AI makes it especially worthwhile to decide what context is necessary before copying everything within reach.

Know which workspace is approved for the task

An organization needs to make its approved tool and permitted uses understandable to the people doing the work. A familiar-looking chat window does not establish what data it may receive.

Explain where staff can find the relevant guidance and whom they should contact if their task falls outside it. The answer may depend on the information involved, the organization’s agreement, access controls, and how that particular service handles the data.

Removing a name is not a complete privacy check. A combination of circumstances can still identify someone. When a task needs personal information, use the organization’s agreed process and obtain appropriate guidance rather than improvising a workaround.

Make the supported route practical enough that a busy colleague can recognize it and use it.

Check the draft for details that slipped in

After generating the invitation, review both its facts and its audience. Did the draft import something from earlier context? Did it turn a general accommodation contact into a claim about a particular participant? Does it invite people to submit information through the correct route?

In our imagined outing, the public invitation might simply explain how to discuss access needs with the organizer. Private planning can remain in the appropriate staff workflow.

Keep the participant involved in how the invitation sounds. Someone can tell the team whether it is welcoming or clear without having their personal record supplied to a drafting service.

If the tool cannot do the task within the approved boundaries, a human-written invitation remains a perfectly useful result.

What this means for the people you serve

People should not have to trade unnecessary personal details for a clearer event notice. Give them an understandable explanation when information is needed and a suitable place to discuss individual arrangements. Their choice about participating is separate from the writer’s choice of drafting tool.

The genus connection

Today, genus care-team tools provide configured workspaces with role-appropriate access and supported assistance for particular tasks. A connection between tools does not give every user or assistant access to all underlying information.

We would confirm the intended drafting task, available modules, and information permissions with the organization. The broader principle is simple enough to use at a desk: include what the approved task needs, understand where it goes, and review the result before it reaches someone else.

One question to take away

Could we prepare this useful draft with less personal information—and is the workspace we are using approved for what remains?