Phase 0 · Foundations
Choosing the right tool
By the end, you'll be able to…
- Use three questions to decide which AI tool fits a given task
- Match a task to a tool based on where the data lives and what the task needs
- Explain why the 'best' tool depends on the job, not a ranking
Why it matters
With four capable assistants and no single winner, 'which one should I use?' is a real, recurring question. This lesson replaces agonising with a quick three-question test, then walks one ordinary task through all four tools so you can feel how the choice actually plays out. The goal isn't loyalty to one tool; it's picking the right door each time.
There's no 'best', there's 'best for this'
By now you know the four assistants (ChatGPT, Copilot, Claude, Gemini) are four doors into a similar room. Any of them can chat, draft, summarise and analyse a file. So "which is best?" is the wrong question. The right one is "which fits this task, today?", and that usually has a quick answer once you ask three things.
The three questions
Run any task through these, in order. Most of the time the first one alone decides it.
1. Where does the data live?
The strongest pull is your existing tools. If the material you're working with already sits in Microsoft 365 (a Word doc, an Excel sheet, an Outlook thread), then Copilot, which lives inside those apps and (with the right work licence) can read them directly, saves you all the copy-pasting. If your world is Google Workspace (Gmail, Docs, Sheets), then Gemini is right there beside it. Working with loose text you'll paste in yourself, with no app allegiance? Then ChatGPT or Claude on the web are both natural. Following the data avoids friction and keeps you from shuttling sensitive content between systems.
2. Does it need current information?
If the task depends on up-to-date facts (today's prices, this week's news, a figure that changes), you need a tool with web search on (any of the four can, when enabled) rather than an answer from frozen training. If it's self-contained (rewrite this, summarise that, brainstorm names), currency doesn't matter and any tool will do. This question mostly decides a setting (search on or off) more than a product, but it's worth asking every time so you don't quote something stale.
3. Does it need to act, or just answer?
Some tasks only need words back: a draft, an explanation, a summary. Others want the tool to do something: pull data out of your live spreadsheet, work across your actual emails, take an action in a connected app. "Just answer" is the easy case any assistant handles. "Act on my real systems" points you towards the tool wired into those systems (Copilot in Microsoft 365, Gemini in Workspace) and, because it touches real data, towards an employer-approved account, per the privacy lesson.
A quiet fourth consideration rides alongside: style and strengths. For careful, sensitive writing many people reach for Claude; for a fast general workhorse, ChatGPT. Small preference, real difference on the day.
One task, four tools
Let's make it concrete with a single, ordinary job: turn a page of rough meeting notes into a structured summary and an action list. Picture the Fernway meeting notes: the messy weekly ops sync, half-finished actions, a couple of owners missing. Here's how each door handles the same task.
- ChatGPT. Paste the notes in, ask for a summary plus an actions table. A fast, flexible general assistant, the obvious pick when the notes are just text on your clipboard and you've no strong reason to be elsewhere. This is the sensible default for a self-contained job like this.
- Claude. Same approach: paste and ask. Claude tends to be strong with longer documents and produces careful, well-judged prose, so if the summary is going somewhere that needs a considered tone, it's a fine choice. For plain notes, the result is very close to ChatGPT's.
- Copilot. The story changes if those notes live in a Word document or a Teams meeting in your Microsoft 365. Then Copilot can summarise them in place, with no copy-pasting and no moving the content out of your work environment, then drop the actions straight into a document. Where the data lives makes Copilot the low-friction winner here.
- Gemini. Likewise if the notes are in Google Docs or arrived over Gmail. Gemini can work on them right where they are, inside Workspace, and help you turn the summary into a shareable doc. Same logic as Copilot, different ecosystem.
Notice what decided it: not which model is "smartest", but where the notes already were and whether the output needed to slot back into your systems. That's questions one and three doing the work. And whichever door you pick, the actual prompt barely changes:
Turn these rough meeting notes into: (1) a four-sentence summary a manager could read at a glance, and (2) a table of every action with columns for Action, Owner and Due date. Mark any action with no owner as 'UNASSIGNED'. Notes: [paste, or point Copilot/Gemini at the file].
Why this works: A clear task plus an explicit output shape works across all four assistants; the tool choice is about where the data lives, not about needing a different prompt.
If the notes mentioned a time-sensitive fact you needed to confirm, question two kicks in and you'd add grounding:
Also, one action mentions the current UK VAT rate. Search the web, confirm today's rate, and cite the GOV.UK page so I can check it.
Why this works: Turning search on and asking for a source handles the 'does it need current information?' question, grounding a live detail instead of trusting frozen training.
And if it needed to act on your real systems, question three routes you to the right, approved place:
In this Word document of meeting notes, insert a summary at the top and add a formatted actions table below it, then flag the two unassigned actions for follow-up.
Why this works: A task that must read or update real work files belongs in the tool wired into those files and, because it touches real data, on an employer-approved account.
Try it now
Common mistakes
- Hunting for the one 'best' tool. There isn't one. The right choice shifts task to task; the three questions decide it faster than any ranking.
- Ignoring where the data lives. Copy-pasting a Word doc into a web chatbot when Copilot could work on it in place; more friction, and needlessly moving content between systems.
- Forgetting the current-info question. Using a tool with search off for a task that hinges on today's facts, then quoting something stale.
- Reaching for a personal tool when the task needs to act on real work data. "Act on my systems" means an approved business account, not a personal login; the privacy lesson still applies.
- Over-trusting a tool because it's your favourite. Familiarity isn't fit. Your preferred assistant can still be the wrong door for a task whose data lives elsewhere or that needs current facts; pick for the job, then check the output as always.
Keeping current
Which tool integrates with what, and how deeply, changes constantly as the four companies race to embed AI in their ecosystems. The three questions are durable; the specific integrations aren't. Check each tool's official product pages for what it connects to today before committing a workflow to it. Accurate as of 13 July 2026.