Course resource
AI Tool Decision Tree
Which assistant should you open for the task in front of you? Work down the tree until something matches.
This deliberately names capabilities, not products. Model names and rankings change every few months; the questions below do not. Check the live comparison on the course tools page for which product currently leads each category.
Start here
1. Does the answer depend on something that happened recently?
Yes → Use an assistant with live web search, and make it cite sources. Models have a training cutoff and will answer confidently about events they never saw.
No → continue.
2. Are you working from a long document, or many documents at once?
Yes → Use the tool with the largest context window you have access to, and upload the source rather than pasting excerpts. If it is more than a few hundred pages, use a tool built for document grounding rather than a general chat.
No → continue.
3. Is the output something a person will read as your writing?
Yes → Use the assistant you judge best at prose, and give it a voice profile. Then edit. Nothing you have not edited should go out under your name.
No → continue.
4. Does it involve numbers, a spreadsheet, or a calculation?
Yes → Use a tool that actually executes code rather than predicting the answer. Language models are unreliable at arithmetic done "in their head" and reliable when they compute it. Ask it to show the calculation.
No → continue.
5. Does it involve an image — reading one, or making one?
Reading (a screenshot, a scanned table, a diagram) → any current multimodal assistant handles this. Verify extracted numbers against the original.
Making → use a dedicated image generator, not a chat assistant's built-in one, when quality matters.
6. Is the data sensitive?
Yes → Use a paid or enterprise tier with training disabled, an approved company tool, or a model running locally. See the AI Safety Checklist.
No → your default assistant is fine.
Choosing a default
You need one assistant you use for 80% of things. Decide it with a two-week trial, not a comparison article.
| If you mostly... | Prioritise |
|---|---|
| Write documents, emails, long-form | Prose quality and steerability |
| Research and cite | Live search with real source links |
| Work inside Google or Microsoft files | Whichever integrates with the suite you already pay for |
| Handle sensitive or regulated data | Data controls, retention terms, local options |
| Work with very large documents | Context window size |
| Watch every rupee | Free-tier limits, then price per month |
When to use two
Most people need one. You need a second when:
- Your default lacks live search and your work depends on current information.
- You need a long-context tool occasionally but not daily.
- Your employer mandates one tool and you want a different one for personal work. Keep these accounts strictly separate.
You do not need a second because a benchmark says another model is 3% better.
When to pay
Pay when you hit the free tier's limit repeatedly in a week, doing work that matters. Not before.
A rough test: if a subscription costs roughly what an hour of your time is worth, it pays for itself the month it saves you an hour. Most people who use AI daily cross that line quickly. Most people who use it twice a week do not.
When AI is the wrong tool
- The task needs a guarantee of correctness — accounting entries, dosages, legal filings. Use AI to draft and a qualified human to sign off.
- The task is faster by hand. A three-word email does not need a prompt.
- You do not know enough to spot a wrong answer. Learn the basics first, or get a human reviewer.
- The input cannot leave your organisation and you have no approved tool.
Record your choice
| My default assistant | |
| Chosen because | |
| Second tool, if any | |
| Used only for | |
| Review this decision on |