This is not a summary. It is a real learning unit from this module — exactly as it appears in the booking portal, with a real chat tutor conversation. No login needed, it is here to try out.

T1 Token Awareness – Unit 4 of 9: Chat and context discipline (levers 1–3)

Module: T1 Token Awareness · Unit 4 of 9 · Topic block: The 7 behavioural levers in practice Your learning objective today: You have applied the first 3 of the 7 behavioural levers (new chat instead of an endless thread, summarising instead of the full text, targeted context instead of the whole document) to a real task of your own, and felt for yourself the difference from your previous habits.

Why this matters to you

Topic block 1 gave you the understanding. Topic block 2 now gives you the tools. The 7 behavioural levers were deliberately chosen so that each one is simple to apply on its own – and together they still produce a noticeable effect (mathematically up to about 40% with all 7 active, realistically more like 25–40% in everyday use). Today, the first three, all to do with your chat and context habits.

Concept: levers 1–3 in detail

New chat instead of an endless thread. A direct consequence of the context window principle from learning unit 2: start a new chat for each completed topic, instead of collecting everything in one endlessly growing history. Rule of thumb: is the topic finished, or are you switching topic completely? Then start a new chat.
Summarising instead of the full text. Instead of pasting in a long document or a long existing history in full, have it summarised (or summarise it yourself) briefly before you carry on working with it. The summary then becomes the new, shorter basis.
Targeted context instead of the whole document. When you give the AI a document to work on, consider: do you really need the whole document, or is the relevant section enough? Using a 20-page document for a question about page 3 wastes tokens on 19 pages that are not needed at all.

All three levers have one thing in common: they do not reduce the quality of your results, only the unnecessary context you are carrying along with them.

Your exercise – 4 rounds on a real task

Choose a real, current task where you would normally use a longer document or an existing chat (e.g. a report, a longer email thread, a project document).

Round 1: Document your current habit

My real task: ___________________________________________
How would you have approached this before (an ongoing chat? the whole document?)? ___________________________________________

Round 2: Apply lever 1 – a new, clean chat

Start a fresh chat just for this one task.

What did you deliberately leave out that would have automatically carried over in an ongoing chat? ___________________________________________

Round 3: Apply lever 2 – get it summarised

If your task is based on a longer text or document: first have a short summary produced, instead of continuing to use the full text.

My summarising instruction: ___________________________________________
Roughly how much shorter is the summary compared with the original? ___________________________________________

Round 4: Apply lever 3 – a targeted excerpt

Identify the section that is actually relevant to your specific question, instead of using the whole document.

Which section was actually relevant? ___________________________________________
How much of the whole document would you have sent along unnecessarily without this lever? ___________________________________________

Self-check – in conversation with the chat

Write your answers in the course chat:

Name the 3 levers from this unit, with one sentence each on why they save tokens.
At which point in your real task did which lever help the most?
What would you have done differently before (before this module)?

Bonus, if you have more time or finish early

Bonus A: Apply all 3 levers to a second, different task – does it work without much thought?
Bonus B: Consider which kind of tasks in your everyday work you most often use unnecessarily long ongoing chats or whole documents for – and what a good "switch-over point" for a new chat would be.

What you can do now – and what comes next

You have tried out the first 3 levers on a real task. In learning unit 5, two more are added, targeting your prompt and your desired answer length directly – with a direct before/after comparison on the token ticker.

Stuck? Ask the chat. If you are unsure whether a task counts as "a new topic" or still belongs to the current chat – when in doubt, start a new chat. A little too often is less of a problem than a little too rarely.

T1 Token Awareness – Unit 4 of 9: Levers 1 to 3: carrying less along

Practice sheet for this learning unit · T1 Token Awareness · Unit 4 of 9 Learning objective: Apply levers 1 to 3 to a real task of your own.

What this is about

The first three levers work on the same point: how much text you send along unnecessarily.

1. New chat instead of an ongoing thread – new topic, new conversation. 2. Summarise instead of full text – have the current state summarised and carry on from there. 3. Targeted context instead of the full document – send the section that matters, not all forty pages.

All three take one extra step and save more than any switch of model.

The three things that matter

One topic, one chat. The separation is not tidiness for its own sake, it is the lever.
Summaries are reusable. Once created, they replace the history permanently.
A targeted excerpt beats the full document – on quality too. A model given forty pages answers less precisely than one given the right page.

The most common mistake

Sending the whole document "so the context is right". The context does not get better this way – it gets bigger, and the essentials get buried in it.

Your prompt

Copy it into any AI tool. Fill in the MY DETAILS block first – without it you get a general answer that fits anyone and helps you with nothing.

You are a pragmatic guide who uses a real task to show what can be saved.

MY DETAILS
- A real task of mine, not a made-up one: [describe]
- What I have sent along for it so far: [documents, history, attachments]
- How often I do this task: [frequency]

TASK
Go through my task: what does the model really need to see, what can go, and what should I replace with a summary instead? Estimate the saving as a range. Finally, rewrite my prompt the way I should use it going forward.

RULES
- Do not invent anything about my organisation, my channels or my figures. Whatever is missing under MY DETAILS you ask about – you do not assume it.
- Do not state any figures, benchmarks, platform statistics or algorithm rules as fact unless they come from my details. Such information becomes outdated and does not carry over to my case.
- Work only with my real task. If my description is too thin, ask – do not invent a sample task.
- Do not name any prices, estimate in tokens.

FORMAT
Three sections matching the three levers, then the rewritten prompt.

How you can tell it is good

The new prompt is shorter, and you would use it exactly like that tomorrow.
It states what you deliberately do NOT leave out, and why.

One notch further

Verify. Apply levers 1 to 3 to the task you do most often – not to the biggest one. Frequency beats size when it comes to consumption.

Last updated: September 2026. Tools, platforms and conditions change without notice – the method on this sheet stays; check individual details against your own case.

Token Awareness (T1) — LU 4 of 9