The short answer
Before you book AI training, check six things: you know the problem you're solving, the trainer uses AI for real work, the session uses your team's own tasks, it covers data safety and checking output, the price is fixed and clear, and there's support afterwards. Below: 30 questions, red flags and a scorecard to compare providers.
Buying AI training in 2026 is a bit like buying a used car in 1986. Plenty of choice, lots of shiny paint, and a worrying number of sellers who learned what's under the bonnet about a fortnight ago.
The UK government's own employer survey found the biggest barrier to AI training wasn't cost or time. It was that half of employers didn't know what training was relevant. So here's the checklist I'd want in your hands before you talk to anyone, us included. Tick as you go. It saves in your browser, and it prints.
What should I check before buying AI training for my business?
Check the problem you're solving and your team's current skills, then the trainer's real-world experience, whether the content uses your own tasks and covers data safety, the group size and format, the price and terms, and the support afterwards. The 30 questions below cover all six stages in the order they're worth asking.
Before you shop
Know what you're buying before anyone sells you something.
"Get the team using AI" isn't a brief. Hours lost to admin, inconsistent proposals, data risk or a stalled Copilot rollout each need different training.
A good answer: You can finish the sentence "after this, people will be able to..." with a real task.
Paying to teach confident users what a prompt is wastes everyone's afternoon. Ten minutes with the AI skills assessment tells you the weakest skill and who's below the safety line.
A good answer: You have a score per person, or at least a rough split of beginners and regulars.
Mixing the MD who has never opened ChatGPT with the intern building custom GPTs makes both of them miserable.
A good answer: Groups of similar ability, or a provider with a clear plan for mixed rooms.
Training people on a tool they can't access at work is driving lessons with no car.
A good answer: Approved tools are agreed and accounts work before the session, not during it.
Training without rules teaches people to paste client data faster. Read AI policy before training staff.
A good answer: A one-page policy exists, or the training includes writing one.
The trainer
This is where most of the quality lives, and where most buyers check least.
Someone who learned the slides last month teaches features. Someone who uses AI daily teaches judgement, which is the bit people actually need.
A good answer: Specific, recent examples from their own week, including what went wrong.
Accountants, estate agents and marketing teams use AI very differently. Generic examples don't transfer.
A good answer: Examples from your sector, or a plan to build the session around your own tasks.
The impressive founder on the sales call isn't always the person who turns up.
A good answer: A named trainer you can look up, and ideally speak to first.
A reseller has a reason to say their tool does everything. Sometimes it does. Often it doesn't.
A good answer: They'll tell you when ChatGPT, Claude, Copilot, Gemini or Perplexity is the better fit, and why.
Reviews are easy to collect and hard to fake at volume. A reference call is better still. Ours are on our reviews page.
A good answer: Recent, named reviews on Google or Trustpilot, and a client happy to take your call.
The content
What people will actually learn, and whether it'll still be true in March.
People remember the proposal they rewrote in half the time. They forget the case study about a fictional bakery.
A good answer: The provider asks for examples of your work before the session.
AI is learned by doing. Watching someone else prompt is like watching someone else swim.
A good answer: At least half the time with people on their own keyboards.
AI is confidently wrong often enough to matter. Anyone who skips this is training people to publish errors faster. See judging AI accuracy.
A good answer: A clear method for checking facts, figures and sources.
UK GDPR still applies when the tool is clever. This is the one topic that keeps a training budget from turning into an incident report. Our guide to keeping information safe covers the basics.
A good answer: Practical rules about approved accounts, settings and anonymising, not a scary slide.
The tools change monthly. A course built last year will confidently teach you features that have moved, renamed or vanished.
A good answer: Updated within the last three months, with current tool names and screenshots.
Delivery and format
The practical stuff that decides whether people turn up and stay awake.
Too basic and the confident users switch off. Too advanced and the nervous ones decide AI isn't for them.
A good answer: A level check before the day and a plan for stretching or supporting individuals.
Past about eight to ten people, hands-on training turns into a lecture. Cheap per head, poor per outcome.
A good answer: A cap of around eight to ten per trainer for practical sessions.
In person suits nervous beginners. Online suits spread-out teams. Neither is automatically better. More in in-person vs online AI training.
A good answer: A recommendation based on your team, not on what's easiest for them.
Ninety minutes clears the fear. A half day builds habits. Two days is more than most people absorb before they've practised. See how long AI training takes.
A good answer: A length matched to the goal, with practice time afterwards.
Twenty minutes of password resets is the most expensive twenty minutes of the day.
A good answer: A short joining checklist: accounts, logins, laptops, sample work.
Price and terms
Where small print becomes big money.
Per group is usually simpler for small teams, per head for big ones. Compare like with like using the AI training price guide.
A good answer: A fixed, written price with the group size stated.
VAT adds 20%. Travel outside London can add a few hundred pounds. Neither should be a surprise on the invoice.
A good answer: Ex-VAT and inc-VAT figures, and travel stated in writing.
Diaries move. Sick days happen. A fair policy tells you a lot about the provider.
A good answer: A clear rescheduling window, without penalty if you give reasonable notice.
Slides, prompt libraries and recordings are how training survives the next new starter.
A good answer: You keep the materials for internal use, and you know whether sessions are recorded.
Apprenticeship units, including three AI leadership units, are levy-fundable. Short commercial workshops aren't. Better to know now than in week three.
A good answer: A straight answer, including "no, this one isn't".
After the session
Training is the easy bit. Keeping it going is the job.
The first fortnight is when habits form or die. Questions come up on Tuesday, not during the session.
A good answer: Resources to keep, and a way to ask follow-up questions.
If you don't measure the before, you can't prove the after. Here's how to measure the ROI of AI training.
A good answer: A baseline (time on key tasks, or skills scores) and a check-in date.
A prompt library, a cheat sheet or a few saved assistants turn a good afternoon into a working habit.
A good answer: Practical takeaways built from your own tasks.
Good providers tell you when you don't need more training. Bad ones always find another module.
A good answer: An honest view of what, if anything, comes next.
Without an owner, AI use fades after six weeks. One curious person with an hour a week makes a huge difference. See how to get your team to adopt AI.
A good answer: A named AI champion with time set aside for it.
What are the red flags when buying AI training?
Walk away from AI training that guarantees a specific productivity percentage, is mostly slides, can't name the tools the trainer used this week, pushes one tool they happen to resell, skips data protection, hides the price of a standard course or pressures you into a multi-session package before the first one.
- Guaranteed results. "Save 40% of your time, guaranteed." Nobody honest can promise that about your team before meeting them.
- Death by deck. Sixty slides and a short demo isn't training. It's a presentation with a coffee break.
- Vague about their own use. Ask what they used AI for yesterday. Waffle is an answer.
- One tool for everything, which they resell. Fine if it's the right tool. Worth a second opinion if it isn't.
- No mention of data protection. If they don't raise it, your team won't think about it either.
- Price only after a 45-minute "discovery call" for what is clearly a standard course.
- Last year's content. Old tool names, old screenshots, features that no longer exist.
- Jargon as a sales tactic. If the pitch needs a glossary, the training will too.
- Package pressure. A good provider is happy to earn the second session by delivering the first.
How do I compare AI training providers side by side?
Score each provider from 1 to 5 on seven weighted criteria: the trainer's real-world AI use, tailoring to your tasks, hands-on time, data safety and checking, follow-up support, price clarity and references. Multiply each score by its weight, divide by five and add up for a total out of 100.
Weights are my opinion, and I've put them where most training succeeds or fails. Change them if your priorities differ, but decide the weights before you see the quotes, not after.
| Criterion | |||
|---|---|---|---|
| Trainer uses AI for real workweight 20 | |||
| Built around your own tasksweight 20 | |||
| Hands-on timeweight 15 | |||
| Covers data safety and checkingweight 15 | |||
| Support after the sessionweight 10 | |||
| Clear price and valueweight 10 | |||
| Reviews and referencesweight 10 | |||
| Weighted total /100 | - | - | - |
Price clarity scores 10 out of 100 on purpose. A cheaper provider who fails on the top three criteria is the expensive option, because you'll be buying training twice. For actual numbers, check quotes against the AI training price guide.
Put us through the checklist
Seriously. Book a call, bring the thirty questions and ask the awkward ones. If we're not the right fit, I'll tell you who might be.
Who should be involved in buying AI training?
Involve whoever owns the budget, a manager from the team being trained, someone who looks after IT or data protection, and at least one sceptic. In a small business that might be two people. The sceptic matters most, because they'll ask the questions the enthusiasts skip.
The person who signs the cheque rarely sits in the session, and the person who sits in the session rarely chooses it. Get both in the same conversation for ten minutes. And add whoever's going to roll their eyes at the whole thing. If the training convinces them, it'll convince everyone.
How long should buying AI training take?
For a small business, two to three weeks is plenty: a week to agree the problem and assess skills, a week to speak to two or three providers using this checklist, and a few days to agree dates and get accounts ready. Longer than six weeks usually means the brief isn't clear yet.
If you're stuck on the brief, start smaller. A 90-minute session for one team will teach you more about what the rest of the business needs than a month of scoping documents.
How does The Oxford AI School answer these questions?
The Oxford AI School publishes fixed group prices, caps practical sessions at eight people (ten for the AI Policy Workshop), builds sessions around your team's own work rather than case studies, teaches across ChatGPT, Claude, Gemini, Copilot and Perplexity, and delivers online or in person across the UK from Witney, Oxfordshire.
| Question | Our answer |
|---|---|
| Does the trainer use AI for real work? | Every day. I run the school, its marketing and this website with the tools we teach. |
| Real tasks or case studies? | Your work, not case studies. We ask for examples before the session. |
| Tool-neutral? | Yes. ChatGPT, Claude, Gemini, Copilot and Perplexity, and we'll say which suits which job. |
| Group size? | Up to 8 for team courses, up to 10 for the AI Policy Workshop. |
| Fixed price? | Yes. Intro To AI £1,199 + VAT, half-day modules £2,299 + VAT, AI Policy Workshop £2,999 + VAT. |
| Online or in person? | Both, across the UK. Half-day modules cost the same either way. |
| Reviews? | Five-star reviews on Trustpilot and Google, on our reviews page. |
Key takeaways
- Agree the problem and assess your team's skills before you speak to any provider.
- The trainer's own day-to-day use of AI is the strongest single signal of quality, and the one buyers check least.
- Insist on sessions built around your own tasks, at least half hands-on, with data safety and checking covered.
- Get a fixed written price with group size, VAT and travel stated.
- Walk away from guaranteed results, slide-heavy sessions and package pressure.
- Score providers on weighted criteria you set before you see the quotes.
Frequently asked questions
What questions should I ask an AI training provider?
Ask whether the trainer uses AI for their own work every day, whether the session uses your team's real tasks, how much is hands-on, how it covers data protection and checking output, the maximum group size, whether the price is fixed with VAT and travel stated, and what support you get afterwards.
How do I choose the right AI training provider?
Agree the problem first, assess your team's skills, then score two or three providers on weighted criteria you set in advance: the trainer's real-world AI use, tailoring to your tasks, hands-on time, data safety, follow-up, price clarity and references. Our guide on choosing between AI training programmes goes further.
What should a good AI training course cover?
Prompting, checking output for mistakes, data protection and safe use, choosing the right tool, and applying AI to the team's own everyday tasks. Better courses add working with documents and data, plus simple automation. See what a good AI training course should cover.
How much should AI training cost?
For a UK team of around eight, hands-on in-house training usually costs £1,000 to £3,500 a session, or about £125 to £450 a head. Free government courses cost nothing, and university programmes run from £1,700 to £7,000 a place. Full details are in our AI training price guide.
What are the warning signs of poor AI training?
Guaranteed productivity percentages, mostly slides, a trainer who can't describe their own recent AI use, pushing one tool they resell, no mention of data protection, hidden pricing for standard courses, outdated content and pressure to buy a package before the first session.
Should we write an AI policy before training staff?
Ideally yes, or at least alongside it. Training without rules teaches people to use AI faster without knowing what's safe to share. A one-page policy is enough to start. Our AI Policy Workshop builds one with your team in a day.
Is online AI training as good as in person?
It can be. In person works better for nervous beginners because the trainer can spot who's stuck. Online suits teams spread across sites and confident users. What matters more is group size, hands-on time and whether the session uses your own work.
Can I print this AI training checklist?
Yes. Use the Print checklist button and your ticks print with it. The ticks are also saved in your browser on this device, so you can come back to them later.
About Harry Lang
Harry runs The Oxford AI School in Witney, Oxfordshire, teaching UK teams the basics of AI in a fun, easy and practical way. He's a former CMO who has scaled B2C brands globally, and the author of Learn To Fly With AI and Brands, Bandwagons & Bullshit.
Last reviewed 2 November 2026. Prices and figures change; if you spot one that has moved, email Learn@TheOxfordAISchool.com and I'll fix it.