A practical workshop on AI and Microsoft 365 Copilot for library managers. Keep this handout next to your keyboard.
Contents: 1. Jargon glossary · 2. Prompt recipe cheat sheet · 3. Five golden rules · 4. Exercise sheets
1. Jargon glossary: 24 words, explained with library examples
The basics
| Word | What it means | Library example |
|---|---|---|
| Artificial Intelligence (AI) | Computers doing tasks that normally need human thinking, such as writing, recognising or recommending | A catalogue that learns what to suggest, instead of following fixed rules |
| Machine learning | Computers learning patterns from lots of examples, instead of following rules written by people | Learning which books go together by looking at thousands of loans |
| Algorithm | A fixed set of step-by-step instructions a computer follows | The shelving rules: the same steps and the same result every time |
| Generative AI | AI that creates new text, images or sound | An author, not a catalogue: it writes something new rather than finding something that exists |
| Large Language Model (LLM) | The "engine" behind chat AI, trained on huge amounts of text | A reader who has read millions of books but owns none, and can't say which page an idea came from |
| Training data | All the text and examples the AI learned from | The collection it read: if the collection is old or one-sided, so are its answers |
How AI works
| Word | What it means | Library example |
|---|---|---|
| Natural language | Normal human language, with no codes or commands needed | Asking at the desk in your own words instead of in catalogue codes |
| Chatbot / AI assistant | A program you talk to in everyday language | The reference desk: you ask in normal words |
| Token | The small pieces of text (often parts of words) that AI reads and writes. AI use is often measured in tokens. | Syllables in a reading lesson: words cut into small pieces |
| Context window | How much text the AI can keep in mind in one conversation | The reading table: only so many documents fit. Start a fresh chat for a new topic. |
| Multimodal | AI that works with text, images and sound together | A collection with books, audiobooks, films and photos in one place |
| Fine-tuning | Extra training that makes an AI better at one specific task | A general librarian taking a specialist course in local history |
Working with AI
| Word | What it means | Library example |
|---|---|---|
| Copilot | Microsoft's AI assistant inside Outlook, Word, Excel and Teams | A reference desk built into the tools you already use |
| Prompt | What you type or say to the AI: your question or instruction | The question at the reference desk: the better the reference interview, the better the answer |
| Prompt library | A shared collection of prompts that work well | A binder of standard letters that work well |
| Grounding | Giving the AI your own file or email so it answers from that | "Please answer using only this book" |
| Agent | AI that carries out several steps of a task for you | A volunteer who doesn't just answer, but does the booking, the email and the follow-up |
| Automation | Software doing a repetitive task by itself, the same way every time (not necessarily AI) | The automatic overdue reminder: the same rule every time, with no person needed |
Risks and responsibility
| Word | What it means | Library example |
|---|---|---|
| Hallucination | When AI invents something false that sounds convincing | A patron confidently citing a book that doesn't exist |
| Bias | Unfair patterns the AI picked up from its training data | A collection that only holds one point of view |
| Deepfake | A fake but realistic photo, video or voice made with AI | A convincing forgery in the archive |
| AVG / GDPR | The privacy law that protects people's personal data, including what you put in a prompt | The rules behind your members' privacy statement |
| Responsible AI | Using AI in a fair, safe and open way, with clear agreements | A collection policy: agreed principles about what we do and why |
| Human in the loop | A person always checks and approves what the AI produces | The manager signs off: Copilot drafts, you decide |
Tokens explained
AI doesn't read whole words. It cuts text into small pieces called tokens, and it writes its answer one token at a time.
- Common short words are usually one token: the · library · in · is · open
- Rare words, Dutch place names and long compound words are cut into several pieces, for example Na · ald · wijk or biblio · theek · con · ven · ant. (These are examples, because every AI system cuts slightly differently.)
- Rule of thumb: 100 tokens is about 75 English words. Dutch text usually needs more tokens for the same message.
Why it matters to you
- It writes piece by piece. AI predicts the next token, again and again. That's why it sounds fluent, and why it can be fluently wrong.
- Its memory is counted in tokens. In a very long chat, AI can lose track of the beginning (the context window). For a new topic, start a new chat.
- Costs are counted in tokens. Many AI services charge per token. Microsoft 365 Copilot has a fixed licence price per user, but when a supplier offers an AI tool, ask what the price is based on.
- Names can go wrong. Words that are cut into pieces are easier to misspell, so always check names like 's-Gravenzande or Honselersdijk.
2. Prompt recipe cheat sheet
A good prompt is a good briefing for a new assistant. Use four ingredients. Only the goal is required, and every extra ingredient makes the answer better.
| Ingredient | Ask yourself | Example |
|---|---|---|
| Goal | What do I want? | "Write a reply to Sandra Koppenol…" |
| Context | Why do I need it, and who is it for? | "…policy adviser at Gemeente Westland, who asked for our subsidy figures…" |
| Source | Which file, email or meeting should Copilot use? | "…use the figures agreed in this email thread…" |
| Expectations | Length, format, tone? | "…formal but friendly, under 200 words, no promises." |
Fill-in template
Goal: Please [write / summarise / list / explain] …………………………… Context: This is for [who] because [why] …………………………… Source: Use [this file / this email thread / this meeting] …………………………… Expectations: Keep it [length], as [bullets / a table / an email], in a [friendly / formal / simple] tone.
Three habits of good prompters
- Keep talking. The first answer is a draft. Reply with "shorter", "friendlier" or "add the opening hours".
- Let it ask you. Not sure what to say? End your prompt with: "Ask me any questions you need first."
- Write normally. Use full sentences and plain words. Typos are fine, and no magic words are needed.
Where to find Copilot
- Outlook: "Summarise" at the top of an email, and Copilot when you reply
- Word: the Copilot button on the Home tab
- Excel: the Copilot button on the Home tab. The file must be saved in OneDrive with AutoSave on.
- Teams: Copilot in a meeting (with transcription on) or in a chat
- Copilot Chat: the Microsoft 365 Copilot app, or Copilot in Teams
Microsoft sometimes moves or renames buttons. If you can't find one, look for the Copilot icon.
3. Five golden rules for using AI safely
- You are the author. Read everything before you send or share it.
- Check facts, names, dates and numbers, every time.
- Protect people's data. Don't put in more personal details than the task needs. Never include absence records, performance reviews or salaries of staff, or the addresses and reading histories of members.
- Work tools for work. Use your work Copilot, not free public AI apps, for library documents.
- People decide about people. Decisions about hiring, performance reviews and absence stay human.
4. Exercise sheets
Activity: AI or not AI?
Sort the 10 cards into three piles: AI, Not AI and Not sure.
The clue: does it learn from examples, or does it just follow a fixed rule?
One card I'd like to discuss: ……………………………………………
Activity: Token cards
Part 1: Cut the words. With a pen, mark where you think the AI cuts each of the 8 word cards into tokens. There is no exact answer, because every AI system cuts a little differently.
| Word | My guess: how many tokens? |
|---|---|
| the | |
| library | |
| Copilot | |
| subsidy | |
| Naaldwijk | |
| bibliotheek | |
| Honselersdijk | |
| 's-Gravenzande |
Which word had the most pieces, and why do you think so? ………………………………………
Part 2: How big is it? Put the 6 text cards in order from smallest to biggest, then match each one to a token number card.
| Order | Text card | Token number |
|---|---|---|
| 1 (smallest) | ||
| 2 | ||
| 3 | ||
| 4 | ||
| 5 | ||
| 6 (biggest) |
What does this tell you about how much Copilot can "keep in mind" at once? ………………………………………
Activity: Fix this prompt
Rewrite your prompt using the four ingredients. Then try both the old and new versions in Copilot Chat.
| Prompt A: "Summarise this." | Prompt B: "Write a vacancy text." | Prompt C: "How are our numbers?" | |
|---|---|---|---|
| Goal | |||
| Context | |||
| Source | |||
| Expectations |
What changed in the answer? ……………………………………………
The practice files are about Bibliotheek Westland, but all people, emails, figures and decisions in them are invented.
Practice 1 · Outlook: answer a request from the municipality
Practice file 1: the email thread "Input for 2026 subsidy report + questions about Maasdijk"
- Open the email thread and click Summarise at the top.
- Click Reply, open Copilot, and type the prompt below.
- Ask for one change, such as "make it shorter" or "make it more formal".
- Check the figures, dates and times. Don't send it!
Write a reply to Sandra Koppenol at Gemeente Westland. Use what Marieke and Jeroen agreed in this thread: when we send which figures, how we explain Maasdijk, and the meeting time. Formal but friendly, under 200 words, no promises about Maasdijk.
Check: Does it promise figures we don't have yet? Does it present anything about Maasdijk as decided? Is the meeting time right? ……………………
Practice 2 · Word: turn the MT meeting into actions
Practice file 2: transcript of the management team (MT) meeting
- Open the file and click Copilot on the Home tab.
- Type the prompt below in the Copilot pane.
- Then ask: "Write a short staff update about the decisions."
Summarise this MT meeting for the director, who missed it. Give me: the decisions made, a table of actions with who and by when, and the open questions for 1 December. Keep it to one page.
Check: Did Copilot find every action and deadline? Would you send the staff update as it is, or does it share things the MT isn't ready to share? ……………………
Practice 3 · Excel: ask questions about the branch figures
Practice file 3: Westland figures 2026 (sheets: Branch figures, Budget 2026, Staff)
- Save the file to OneDrive and make sure AutoSave is on.
- Click Copilot on the Home tab and ask these questions one by one:
- Which of our eight branches had the most visitors, and which the fewest?
- Show visitors per month for each branch as a line chart.
- Which branch is growing and which is shrinking? Explain in three plain sentences for the Supervisory Board.
- Extra: "Which budget lines have spent more than 80% so far?" (Budget 2026 sheet) and "Which teams have vacancies, and where is sickness absence above 5%?" (Staff sheet)
Check: Look at the table yourself to confirm one of Copilot's numbers. ……………………
Practice 4 · Word or Copilot Chat: from rough notes to a project proposal
Practice file 4: rough notes for "Digitaal Westland", help with the digital government in every village
- Open the file and Copilot, and type the prompt below.
- Then ask for a short press release for the local newspaper.
- Then ask for a LinkedIn post for our partners.
- Extra: "Shorten the proposal to half a page for the Supervisory Board" or "Add a short section on risks."
Using these notes, write a one-page project proposal for Gemeente Westland. Include: the need in our villages, what we will do, the results so far, the budget, partners, and how we measure success. Clear headings, convincing but factual.
Check: Is the budget still €48,000, and does the breakdown add up? Did Copilot add partners or results that aren't in the notes? ……………………
Activity: Spot the mistakes
This is the kind of draft Copilot could write. It reads well, but it contains four mistakes. Compare it with the email thread from Practice 1.
Dear Ms Koppenol,
Thank you for your email. We will send you the complete 2026 figures by 27 November. Visitor numbers in Maasdijk fell by 8%, so we have decided to move the Tuesday evening opening to Saturday morning. We look forward to seeing you on Tuesday 8 December at 10:00 at the town hall in Naaldwijk.
Kind regards, Jeroen van der Berg, Manager Public Services
The mistakes I found:
- ……………………………………
- ……………………………………
- ……………………………………
- ……………………………………
Activity: My first AI task
My task (one I wish took less time): ……………………………………………
| Ingredient | My prompt |
|---|---|
| Goal | |
| Context | |
| Source | |
| Expectations |
How I improved it after the first answer: ……………………………………………
Would I use this again? Why? ……………………………………………