Why does writing your AI acceptable-use policy in plain English matter for psychological safety — and what did simplifying it actually achieve?
Charlotte Halligan's organisation started its AI rollout with an acceptable-use policy seven pages long, packed with terms like "LLM", "large language model", and "prompt engineering" — jargon that immediately put employees off. They stripped it down to two plain-English paragraphs with a clear pointer to where staff could get more information or ask specific questions. The result was direct: the less confident people feel about the rules, the less willing they are to experiment safely, so making the policy readable was inseparable from building genuine psychological safety.
The logic behind the change is straightforward. A seven-page document full of technical acronyms does not communicate authority or rigour to most employees — it communicates inaccessibility. When workers cannot quickly understand what they are and are not allowed to do, uncertainty fills the gap, and uncertainty breeds caution or avoidance. By reducing the policy to something anyone could read in under a minute, Halligan's team removed a concrete barrier to safe experimentation.
This move reflects a broader principle running through the rollout: policy and communication are not separate workstreams. The way rules are written is itself a communication act, and in the context of AI — a technology that already carries significant anxiety for many employees — every unnecessary layer of complexity chips away at the confidence people need to engage. You can follow the full conversation on Listenly to hear how this fitted into the wider change management approach.
"It's not necessarily that AI is going to take your job, but somebody who knows AI is going to take your job. So you kind of have to get on the bus or get off the bus."
— Charlotte Halligan, Comms Leader, speaking on Unprompted: Real AI in ICAbout Charlotte Halligan
Charlotte Halligan brings over 20 years of experience in internal communications, having navigated multiple technology implementations and organisational change programmes throughout her career. Her perspective on AI rollouts is not that of an outside consultant parachuted in — she was living the adoption journey herself at exactly the same time as she was leading it for thousands of employees across a complex global organisation with diverse role types, legislation, and regulatory environments.
Halligan has been candid about being a slow personal adopter of generative AI at first, which gives her a particular credibility when addressing employee resistance: she understands the hesitation from the inside. Once she engaged with tools like ChatGPT directly, she recognised their genuine potential and took on a role with enterprise-wide AI enablement as a central responsibility. That combination — lived scepticism turned advocate, plus real-world delivery accountability — makes her voice on topics like policy accessibility and psychological safety grounded rather than theoretical.
See also
Charlotte found that production-line staff were far more terrified of AI than, for example, finance teams, because they saw automation as an existential threat to their roles — and communicating with them required a distinctly different approach that acknowledged those fears directly.
Charlotte found that the traditional know-feel-do sequence does not work well for AI adoption, and that starting with the 'do' piece was far more effective — getting people hands-on with the tools first, rather than trying to build knowledge or sentiment beforehand.
Unlike rolling out a tool such as Salesforce — where success is clearly defined and adoption is eventually guaranteed — generative AI is totally unbounded: there is no fixed endpoint, no single correct way to use it, and the technology itself keeps changing, which makes the communication challenge fundamentally different.