The gap between people who use AI daily and people who refuse to touch it is widening fast. A Guardian guide published this week, drawing on three AI experts, offers practical advice that maps surprisingly well onto challenges UK businesses are wrestling with right now.
One-third adoption is a workforce problem, not a technology one
Pew Research data from 2025 shows that one-third of US adults have used ChatGPT, including 58% of adults under 30. That figure has roughly doubled in two years. UK adoption patterns track similarly, and the split is now visible inside organisations: some teams run AI-assisted workflows daily while others in the same building have never logged in.
This is not a technology readiness problem. The tools are accessible and mostly free. It is a confidence and clarity problem. People do not know what AI is good for, where its limits sit, or whether using it at work will make them look lazy or incompetent.
Strategic Reality: The adoption divide within organisations is often wider than the divide between organisations. A company where 60% of staff use AI tools and 40% refuse them has an internal consistency problem that no technology purchase can fix.
The Guardian’s expert panel, comprising Timothy B Lee (Understanding AI newsletter), Catherine Goetze (@askcatgpt on TikTok), and Ella Hafermalz (Vrije Universiteit Amsterdam), frames their advice around personal use. But every recommendation they make has a direct workplace parallel.
What the experts actually recommend
The guide breaks AI usage into four categories. Here is how each one translates to an organisational context.
| Expert recommendation | Personal use case | Workplace parallel |
|---|---|---|
| Brainstorming partner | Coming up with ideas for projects | Breaking down complex briefs, generating options before meetings |
| Research assistant | Getting a lay of the land on topics | Competitor analysis, policy review, market scanning |
| Learning accelerator | Picking up new hobbies and skills | Onboarding, cross-training, professional development |
| Information organiser | Structuring notes and findings | Synthesising meeting notes, project documentation, reporting |
Implementation Note: These four categories work as a starter framework for internal AI training. Most employees can grasp “brainstorm, research, learn, organise” faster than abstract capability lists.
What makes this framework useful is that every category includes a built-in constraint. Goetze puts it directly: the best tasks for AI are those “where you know what the right answer looks like.” That single sentence should be printed on every internal AI policy document.
The conversation model beats the command model
One of the more practical insights in the Guardian piece comes from the shift away from prompt engineering. Lee notes that crafting the perfect prompt “is getting less and less important over time.” Goetze goes further: “You really want to think about it as chatting. The magic actually comes from the back-and-forth.”
This matters for businesses because many early AI training programmes taught employees to write prompts like database queries. Precise. Structured. One-shot. That approach still works, but it misses the bigger opportunity.
Success Factor: Employees who treat AI as a conversation partner rather than a search engine get better results. Training should emphasise iterative dialogue, not prompt templates.
Goetze’s “reverse-prompt” technique deserves attention. When she hits a creative block, she asks ChatGPT to generate questions that would help her think through the problem. The AI prompts the human, not the other way round. For UK businesses running workshops or training sessions, this is a concrete exercise that takes five minutes and demonstrates the conversational model immediately.
The implications are worth spelling out:
- Staff who struggle with “what should I ask it?” can instead ask it to ask them questions
- Teams can use reverse-prompting to prepare for client meetings, project kickoffs, or strategy sessions
- The technique lowers the barrier to entry because it requires no prompt engineering skill whatsoever
Where the expert advice gets uncomfortable
The Guardian’s experts are unanimous on one point: never trust AI output without checking it. Lee compares AI to Wikipedia. Goetze says to “check your sources, check those links, check the dates.” Hafermalz warns against staying “in a feedback loop with AI” because “you will end up in dark places.”
This is where most workplace AI policies fall apart. They say “always verify AI output” without specifying what verification looks like, who is responsible for it, or how much time it should take.
| Verification challenge | Why it matters for UK businesses |
|---|---|
| Who checks the output? | If the person using AI lacks domain expertise, verification is meaningless |
| How long should checking take? | If verification takes longer than doing the task manually, the productivity case collapses |
| What counts as verified? | Without clear standards, “I checked it” means different things to different people |
| Who is liable for errors? | UK professional services firms need clear accountability chains |
Critical Context: The instruction to “always verify” is necessary but insufficient. Organisations need verification protocols that specify what good checking looks like for each use case. A legal team reviewing an AI-drafted contract needs different verification standards than a marketing team checking an AI-generated social media caption.
Hafermalz’s advice to “set a clear goal or intention every time you use ChatGPT” sounds simple. In practice, it means that before an employee opens an AI tool, they should be able to articulate what they are trying to achieve and how they will know if the output is good enough. That is a skill many people have not been taught.
The hidden cost of the confidence gap
The Guardian piece describes two camps: people who refuse AI entirely and people who use it daily. For UK businesses, there is a third group that causes more problems than either: people who use AI tools but do not tell anyone.
Shadow AI usage is already widespread. A 2024 Microsoft Work Trend Index found that 78% of AI users were bringing their own tools to work. The figure will be higher now. These employees are uploading client data, internal documents, and sensitive information to consumer AI tools without any governance framework.
Hidden Cost: The employees who refuse AI are visible. The employees who misuse it are not. Shadow AI is a data governance risk that most UK businesses have not addressed.
The Guardian’s experts sidestep this issue because their advice targets individuals. But for any organisation reading their guide, the unasked question is: what are your staff already doing with AI tools, and do you know about it?
Four challenges that UK businesses should address:
-
Data leakage through consumer tools. Employees pasting confidential information into free-tier AI tools have no guarantee about how that data is stored or used. Enterprise AI agreements exist specifically to address this, but many SMEs have not explored them.
-
Inconsistent quality standards. When some employees verify AI output carefully and others do not, the organisation produces work of unpredictable quality. Clients and regulators will not distinguish between AI-assisted errors and human ones.
-
Skills erosion in junior staff. Hafermalz warns individuals about becoming reliant on AI. The organisational version is more concerning: if junior employees use AI to skip foundational learning, they may never develop the expertise needed to verify AI output later in their careers.
-
Training that does not match reality. Many UK businesses offer generic AI training (“here’s how ChatGPT works”) when what employees need is task-specific guidance (“here’s how to use AI for the specific work you do, and here’s where to stop”).
Building an AI adoption framework that works
The Guardian’s expert recommendations, properly translated, give UK businesses a practical starting framework.
Week 1-2: Audit and categorise. Map existing AI usage across the organisation. Find out who is using what, for what tasks, and whether they are using enterprise or consumer tools. Do not frame this as surveillance; frame it as understanding.
Week 3-4: Define use cases by the four categories. For each team, identify specific tasks that fit brainstorming, research, learning, and organising. Be concrete. “Marketing can use AI for brainstorming campaign themes” is actionable. “Teams should explore AI opportunities” is not.
Month 2: Establish verification standards. For each approved use case, define what good verification looks like. Include time estimates. If a task takes 30 minutes manually and 10 minutes with AI plus 25 minutes of verification, that is still a net negative. Be honest about where AI actually saves time and where it does not.
Month 3: Train on conversation, not commands. Build internal training around the conversational model. Include Goetze’s reverse-prompting technique. Make training task-specific rather than tool-specific.
SME Advantage: Smaller organisations can move through this framework faster than enterprises. A 50-person company can audit AI usage in a week. A 5,000-person company needs months. Speed of adoption is a genuine competitive advantage for UK SMEs.
The real strategic question
The Guardian’s experts frame AI as a tool for individuals to expand their capabilities. That framing is correct but incomplete. For UK businesses, the question is not “should our people use AI?” but “are our people using AI well, and do we know about it?”
Three factors will separate organisations that get this right:
-
Clarity over capability. The tools are already good enough. What matters is whether employees understand which tasks benefit from AI assistance and which do not.
-
Verification as a skill. Checking AI output is not intuitive. It needs to be taught, practised, and built into workflows with specific standards.
-
Governance that enables rather than blocks. Blanket AI bans push usage underground. Thoughtful policies that provide approved tools, clear use cases, and defined boundaries keep the organisation both productive and safe.
Take Action: Start with a simple internal survey: who on your team uses AI tools, for what tasks, and using which platforms? The answers will likely surprise you, and they are the foundation for every decision that follows.
Hafermalz’s closing advice in the Guardian piece applies as much to organisations as it does to individuals: “Use it where you can verify it yourself, in the real world.” For UK businesses, that means starting with tasks where the organisation already has the expertise to judge quality, and expanding from there.
Analysis based on “We asked experts about the most responsible ways to use AI tools”, published by The Guardian on 18 March 2026, featuring insights from Timothy B Lee, Catherine Goetze, and Ella Hafermalz.