What are the minimum skills for AI use?

AI is already transforming the workplace, but does today’s workforce have the necessary skills to use it productively, appropriately, and safely?

The short answer, experts agree, is no. According to Forrester, most employees lack the baseline understanding, skills, and ethics to use AI tools like Microsoft Copilot or Google Gemini in this way. Among information workers in five countries, the analyst firm found that High AIQ only grew from 12% to 16% between 2024 and 2025, and front-line workers are likely further behind still.

“While we don’t have a survey of the AIQ of frontline workers at this point, we assume that they lag further due to gaps in education and immersion into tool usage,” notes J.P. Gownder, vice president and principal analyst on Forrester’s Future of Work team.

While natural language input and multimodal models can lower the operational barrier of AI tools, ease of input isn’t the same as the competence of use. Vendors consistently underplay the human side of AI, says Gownder, which includes the skills, motivation, ethics, and behaviors employers need to cultivate for employees to use AI successfully at work.

What’s the worst that can happen?

Lack of ‘AIQ’ can lead to misuse and mistakes, and so a business opens itself up to a lot of risk if users aren’t properly trained, notes Holly Chate, coo of FutureDotNow – a coalition of UK-based leaders working to close the digital skills gap.

Untrained users might unwittingly share customer or company data: “If you put confidential information into an open source GPT, suddenly that data’s now publicly available,” she points out. They may also pass along incorrect information to customers or prospects, or generate confidently wrong answers about their next best step on the job.

There’s a further risk beyond individual error, Luckin adds, noting that people who don’t understand the tool are the most likely to outsource their judgment to it, accepting what it produces because questioning it feels beyond them. “Over time, that erodes the very capabilities they need to use it safely. The gap isn’t only about access, it’s about who retains the judgment to stay in control of the tool and who gradually hands that control over.”

The four key AI skills employees need to have

In reality, the use of AI tools can widen the digital confidence gap between employees, with the confident getting more capable, while the hesitant fall further behind. The solution to this, and to reducing the other related risks, is to ensure your staff gets the necessary training. But what exactly does this entail?

FutureDotNow developed evidence-based AI skills recommendations to embed into the UK Department for Science, Innovation and Technology’s (DSIT) Essential Digital Skills Framework. Its research found four key AI skills that every worker needs to have. These are AI literacy and understanding, effective communication with AI, critical evaluation of AI outputs, and responsible use of AI.

In the case of effective communication, prompting matters, as this is something that doesn’t come naturally to most people, notes Luckin. “The better the prompt, the larger the pattern you give the model to match, so structured prompts with clear context produce better output.

“My caution is about sequence and proportion,” she continues. “I once sat through a training session that spent 45 minutes on prompt engineering and never once explained what a large language model (LLM) does or why it hallucinates. That’s the wrong way round. Teach it after understanding, not instead of it.”

The ability to critically evaluate AI output is the skill most worth investing in, Luckin says, which has three parts. She recommends starting with the why, as people evaluate output far better once they understand that the model generates plausible text rather than checking facts. Next is to build habits.

“I describe this as the ‘wobble test’: ask for sources, challenge the detail, press on the claims and see whether the answer holds together or collapses,” Luckin says.

“Practical checks help too, such as confirming that cited references actually resolve, since invalid digital object identifiers (DOIs), broken links, and fabricated citations are common signatures of generated text.”

The third part is domain grounding. Critical evaluation depends on knowing enough about the subject to sense when something’s off, so training is most effective when it’s contextual, practiced on real tasks the learner cares about, and uses genuine examples of failure, rather than delivered as abstract drills, Luckin explains.

Getting training right

A 2025 survey from McKinsey found that while 88% of organizations use AI, only 6% are generating meaningful returns. Work by BCG explains why, says Luckin: “Algorithms account for 10% of AI success – people and processes make up roughly 70%.”

This highlights the importance of staff training, but it’s just as critical to get that training right. Gownder notes that too few organizations currently invest in sufficient training: “It’s typically one-and-done.” AI requires iterative training, he says, as it changes all the time, and employees need practice and reinforcement. He adds, however, that social learning and practice are even more important.

Chate recommends designating ‘digital champions’ within teams, as “humans learn from humans. If you identify a handful of people to give extra training to, and incentivize them to support their colleagues, that’s been proven hugely successful.”Another idea is to build AI training into the day-to-day, she notes, such as adding 10 minutes to the end of a weekly meeting where people share their top tips or ask questions. “Having that little bit of support and knowing where to go is really helpful,” she adds.

Keri Allan is a freelancer with 20 years of experience writing about technology and has written for publications including the Guardian, the Sunday Times, CIO, E&T and Arabian Computer News. She specialises in areas including the cloud, IoT, AI, machine learning and digital transformation.