Workplace scenarios

Scenario 1: The tempting shortcut

Hannah in client services has a long, messy complaint history from a customer and wants an AI chatbot to summarise it before a review meeting. The thread contains the customer's name, address, and account details, and the only tool to hand is her personal free account.

**What good looks like:** Hannah stops before pasting. Customer personal data does not belong in a personal-tier tool whose data handling terms nobody in the organisation has reviewed. She checks which tool is approved for work use, and if none fits the task, she asks IT rather than improvising. If she does use the approved tool, she includes only what the task genuinely needs, leaving out identifying details the summary can work without.

Scenario 2: The citation that never existed

Ben uses an AI assistant to draft a briefing note for a client, and the draft cites a specific industry report with an author, a year, and a convincing title. It reads perfectly, and the deadline is close enough that skipping the check is tempting.

**What good looks like:** Ben treats every citation as unverified until he has found the source himself. When he searches, the report does not exist; the tool hallucinated it. He removes the claim, replaces it with a genuine source, and reminds himself that a confident tone tells him nothing about accuracy. Nothing AI-generated reaches the client until the facts, figures, and quotes have been independently checked.

Scenario 3: Sifting the applications

A hiring manager, Priya, has forty applications for one role and wonders whether an AI tool could shortlist candidates for her by summarising and scoring each CV overnight.

**What good looks like:** Priya recognises this as exactly the kind of decision about people where bias in AI output matters most, and where candidates' personal data is involved too. She speaks to HR and checks the acceptable use policy before doing anything. If AI is used at all, it is an approved tool, its output is treated as a first draft rather than a judgement, and a human reviews every assessment, with Priya remaining accountable for the shortlist she produces.

Scenario 4: Whose work is it?

Tom used an approved AI assistant to produce most of a research summary that is going out to a client. The client will naturally assume the document was researched and written by the team they are paying.

**What good looks like:** Tom checks the client agreement and his organisation's policy on disclosure. Because AI did a substantial share of the drafting and research, he makes sure its role is transparent where the policy or agreement requires it. He verifies the content line by line before it leaves, and puts his name to it knowing that he, not the tool, is accountable for what it says.