The skills AI work actually needs
The AI in marketing pillar lists the skills AI rewards: briefing, evaluation, data literacy, platform literacy, experiment design and stack awareness. Most are skills good marketers already had, applied more deliberately, which is the first thing to tell a team worried about being replaced or left behind. The table translates them into what a manager can observe.
| Skill | What it looks like in AI work | How you can tell it is missing |
|---|---|---|
| Briefing | A prompt that names the audience, the one question, the sources and the voice | Generic output; long chains of follow-up prompts to fix the basics |
| Evaluation | Judging output against a source, a brand standard and a reader's need | Fluent drafts published with errors or blandness intact |
| Fact and claim checking | Tracing every figure, name and claim before it is used | Invented facts reaching review or customers |
| Data literacy | Knowing what a metric means and whether an answer about it is plausible | Generated analysis accepted without checking the query or table |
| Data handling | Sending only what the task needs to approved tools | Exports pasted into whatever tool is open |
| Experiment design | Testing generated variants with a hypothesis and a fair comparison | Variants launched and winners declared on thin evidence |
| Knowing when not to | Choosing manual work where AI adds review cost or risk | AI used everywhere because it is available |
Two of these are new in emphasis rather than in kind. Evaluation has moved from a final step to most of the work, because producing a first draft is no longer the bottleneck. And data handling, once the concern of a few analysts, is now a daily decision for everyone with a chat window open. The skills map places these alongside the wider marketing skill set.
Finding the baseline
Before training anyone, find out where people stand, and do it by looking at work rather than by asking for self-ratings. Confidence with AI tools and skill with them are poorly matched: the most enthusiastic users are not always the most careful, and the most sceptical are sometimes the best reviewers.
A practical baseline takes one session. Give each person a real task from their own area, the approved tool and the source material, and ask them to produce something usable and to say what they checked. Review the outputs together against the skills table. The exercise shows who briefs well, who checks, who knows the data rule without looking and who pastes everything. It also shows the team that the standard is the work, not the tool.
Record the result per person and per skill, in the same place as other development notes. The hiring and developing marketers guide covers how development plans are kept and reviewed; AI skills belong in the same plan, not a separate one. A marketer who wants to show AI-assisted work outside the team, in a portfolio, a CV or an interview, can use the tutorial on describing AI-assisted work to separate the tool's part from their own and a reviewer's.
Training on the team's own tasks
Tool demonstrations show what a tool can do on an ideal example. Training that changes practice works on the team's actual tasks, with its actual sources and standards, and ends with output a reviewer accepts. If the tools are in daily use and the team is no quicker, the page on AI tools that are not saving time checks whether training built on demos is one of the reasons, beside review depth, prompts and tool choice.
Run training as practice sessions around a task the team does often: product descriptions, campaign briefs, report summaries, ad variants, email drafts. Each session follows the same shape. Someone who does the task well shows the brief they would write and the sources they would supply. Everyone produces an output. The group reviews the outputs against the brand standard and the facts, and discusses what made the strongest ones strong. The session ends with an improved prompt or template that goes into the shared library.
Tie every session to the AI use policy. The data rule, the review tiers and the disclosure rules make more sense practised on a real task than read on a page, and practice is where people find the cases the policy did not anticipate.
Practising review, not just prompting
Most AI training concentrates on prompting, because prompting is visible and satisfying. The skill that protects the brand is review, and it needs as much deliberate practice.
Review practice can be designed. Prepare drafts with known problems seeded in them: an invented figure, a misquoted source, a superlative the product cannot support, a sentence that contradicts the brief, a tone that drifts off brand. Ask the team to review them under realistic time pressure and compare what each person caught. The first round is usually humbling, which is the point. Repeat with new examples until the team catches the common failure modes reliably. The failure modes listed in the accuracy and review workflows guide make a good syllabus.
Review also covers quality, not only errors. A draft can be accurate and still generic, the kind of content that lacks the experience and expertise summarised as E-E-A-TExperience, Expertise, Authoritativeness, Trust.. Practise spotting where a draft needs the team's own knowledge, examples and opinions, and adding them. For generated variants that will be tested, practise the discipline of the A/B TestControlled comparison of two variants. too: a hypothesis, a fair comparison and a stopping rule decided in advance.
Sharing what works
Skills spread through a team when good practice is easy to find and copy. Two mechanisms do most of the work.
A shared prompt library. Prompts, templates and source packs that produced accepted work, stored where the team works, with a note on what each is for, who owns it and when it was last checked. The library turns one person's good brief into everyone's starting point, and the prompt library tutorial covers how to build one, and it gives review findings somewhere to go: a prompt that keeps producing the same defect gets fixed once, for everyone.
Champions. One or two team members with a small amount of protected time each week to pair with colleagues on real tasks, curate the library and bring recurring problems to whoever owns the policy. Choose champions for judgement as much as enthusiasm, rotate the role so it does not become one person's identity, and recognise the time in their objectives. Champions are not a help desk and not tool salespeople; their job is to make the team's own work better.
Short, regular show-and-tell sessions, where someone walks through a task they improved and what they checked, complement both. They work best when people share failures as freely as successes.
Amberlode Coffee Roasters builds skills over a quarter
Amberlode Coffee Roasters, an invented specialty coffee business with a marketing team of 7, runs a baseline session in which each person drafts a product page for a new single-origin coffee from the roaster's tasting notes. 5 of 7 drafts include at least 1 invented tasting note or origin detail, and 2 people paste the wholesale customer list into a chat tool to "personalise" a launch email. Over 10 weeks the team holds 5 practice sessions on its own tasks and 3 review drills with seeded errors; catch rates on the drills rise from 4 of 12 seeded errors on average to 10 of 12. Two champions get 2 hours a week each and build a library of 14 prompts with source packs. At the end of the quarter, the product page defect log shows 3 invented details in 24 pages, against 9 in the 12 pages before the programme, and the team's 2 newest members still write 1 product page a month without AI as part of their development plans.
Protecting judgement and junior development
The hardest skills risk in AI adoption is not that people fail to learn the tools. It is that junior marketers learn the tools instead of the craft. A junior who has never written a product page from scratch, built a media plan by hand or synthesised interviews without help may struggle to tell a good AI draft from a plausible one, and that is the judgement the whole review system depends on.
Managers can protect it deliberately. Keep some tasks manual for learning, chosen for what they teach. Ask juniors to explain their edits: why they kept a paragraph, cut a claim or rewrote an opening. Give feedback on that reasoning as well as on the output. Pair juniors with experienced reviewers so they see judgement exercised, not only its result. And include judgement in development conversations explicitly, so that speed with tools is not the only visible progress. The marketing careers skills section covers how individuals build their own skills; the manager's job is to make sure the team's work still teaches them.
Measure the programme by the work, not by attendance. Defect rates in review, the share of library prompts in active use and the outcomes of review drills are better signals than the number of people trained. Pick one as the programme's KPIA metric tied directly to an objective., and review it with the team each quarter. The academy capstone on an AI operating model for an online course provider practises setting a skills plan with a baseline and champions inside a wider operating model, beside grounding, incident handling, agency rules and measurement.
Common mistakes
- Equating training with tool demos. Demos show what a tool can do; practice on real tasks changes what the team does.
- Measuring confidence instead of skill. Self-ratings miss the gap between enthusiasm and care.
- Training prompting and ignoring review. Review is where most of the work now sits and where the brand is protected.
- Keeping good prompts private. Without a shared library, every person rediscovers the same fixes.
- Turning champions into a help desk. Without protected time and recognition, the role burns out.
- Letting juniors skip the craft. Judgement comes from doing the work, and the review system depends on it.
- Training separately from the policy. The data rule and review tiers stick when practised on real tasks.
- Counting people trained. Attendance is activity; defect rates and drill results show skill.
