What AI can — and can’t — do
Before you use any AI tool at work, you need to know what it’s good at, where it goes wrong, and when to trust it. This module gives you that judgement.
🎬 Watch the video lesson, then work through the full written module below.
Your goals
- Explain, in plain English, what a generative AI tool really is.
- Name three tasks AI does well at work — and two it does badly.
- Decide when you can trust an AI answer and when you must check it.
What AI actually is
A generative AI tool — like ChatGPT, Copilot or Gemini — is a very advanced word-prediction engine. It has read a huge amount of text and answers by predicting, word by word, what most likely comes next.
The key idea
AI predicts what sounds right. It doesn’t know what’s true and isn’t searching verified facts. It can be confident and wrong at once — a hallucination.
When an AI tool answers, what is it really doing?
What it’s great at — and bad at
✅ Great at
- Drafting emails and posts
- Summarising long text
- Rewording and changing tone
- Brainstorming and first drafts
- Explaining things simply
âš ï¸ Bad at
- Exact facts and figures
- Recent events — unless its web-search mode is on (then check the links it cites)
- Precise maths
- Anything you’re accountable for
- Confidential data
Use AI to shape the words, never as your source of truth.
Which is riskiest to hand to AI without checking?
When to trust it, when to check
The “put my name on it?†test
Before you send anything AI helped with, ask: would I be happy to put my name on this and be responsible for it? If it has a fact, figure, name or rule you haven’t verified — check first.
Always verify: facts, stats, legal/financial details, names, anything time-sensitive. Usually safe: tone, structure, wording, first drafts.
The main AI tools — and which to use
You don't need them all. Four cover almost everything at work:
General chat + writing
- ChatGPT — all-rounder
- Claude — long documents, careful tone
- Gemini — tied to Google apps
Inside your work apps
- Microsoft Copilot — Word, Excel, Outlook, Teams
- Gemini for Workspace — Docs, Sheets, Gmail
Free versions are fine to learn on. Paid tiers add better models, file uploads and bigger limits. Rule: use whichever your employer has approved — that matters more than which is 'best'.
How to spot a hallucination
AI is most dangerous when it's confidently specific. Watch for these tells:
Red flags
- Very precise figures, dates or statistics with no link you can check
- Named 'sources', studies or laws you can't find
- Quotes attributed to real people
- Anything about very recent events
Pressure-test it: ask “are you sure? give me a link I can verifyâ€. If it can't, treat the claim as unproven and check it yourself.
Test AI's confidence vs accuracy
Your task (5–10 min)
See the gap between how sure AI sounds and how right it is.
- Open any free AI tool.
- Ask it something you already know well (your job, your town). Judge the answer.
- Now ask something obscure and factual (a niche date or figure).
- Ask “are you certain? give a source.†Notice if it backpedals or invents one.
- Takeaway: confidence is not accuracy. Always verify facts.
Priya’s client email
Priya works in an estate agency. She asks AI to draft a warm email to a buyer and to include the current stamp-duty rate for a £450,000 home.
What should they do? Think, then reveal.
Reveal the answer
Use the draft — check the number.
The wording is perfect AI work. But the stamp-duty figure is a live, legal, financial fact — she must confirm it from GOV.UK/HMRC before it reaches a client. Draft with AI; verify the fact herself.
End-of-module quiz
6 questions. You need 4 to pass. Retake as often as you like.
Question 1
Why can AI sound confident but be wrong?
Question 2
Which is safest to use AI for with light checking?
Question 3
The quickest test before sending AI-assisted work?
Question 4
A colleague says 'the AI is never wrong, it gave a source'. Best response?
Question 5
Which is a GOOD job for AI at work?
Question 6
'Hallucination' in AI means:
Remember this
- AI predicts words, it doesn’t know facts — it can be confidently wrong.
- Great at wording, weak at facts — shape with it, don’t source truth from it.
- Use the “put my name on it?†test before anything goes out.
Finish all 8 modules and pass each quiz to earn your AI Skills for Work certificate of completion — shareable on your CV and LinkedIn. CPD accreditation is being arranged.