Conversation and drafting
ChatGPT and Claude are broad starting points when you want to think, write, analyse or work with documents.
The practical AI chooser
Which AI makes sense for this job?
Choose a useful starting point now, then build a setup you can carry between tools instead of starting again every time the models change.
5plain-English questions
2tools in your shortlist
1portable way of working
The simple explanation
The model is only part of what you are choosing.
A model is the engine doing the language and reasoning work. The product around it decides how you chat, search, upload files, connect your work and approve actions.
That is why two products can use capable models but still suit different situations. It is also why your system should not live inside one product.
Choose a starting point, not a permanent home.Keep your business brief, voice, privacy rules, good examples, source material and review steps somewhere you control. Then you can take that setup from ChatGPT to Claude, Gemini or another suitable tool without rebuilding the way you work.
01
The taskWriting, research, workspace help or multi-step work.
02
Your portable systemYour brief, rules, examples, sources and review steps.
03
The product and modelThe interface, modes, connections, permissions and capabilities.
04
Your judgementPrivacy, checking, editing and approval still belong to you.
A useful first map
Different starting points for different situations.
These are not winners and losers. They are practical reasons someone might put a product on a two-tool shortlist.
Current research
Perplexity, or a verified research mode inside another product, makes sense when visible sources are central to the task.
Your existing workspace
Gemini or Microsoft Copilot may be the shorter path when the work already lives in Google or Microsoft 365 and the right access is available.
Agents are a different kind of choice. Manus and Kimi K3 can make more sense for defined multi-step work. Grok has a clearer reason to be considered when live X conversation is part of the task. More capability also means more decisions about permissions, credits, review and consequences.
Realistic examples
What a proportionate choice can look like.
A routine client email
A bookkeeper has five rough points and wants a warm appointment-confirmation email. ChatGPT or Claude is a proportionate first shortlist. An autonomous agent adds complexity the job does not need.
A changing software question
A consultant needs current guidance and must show where each claim came from. Perplexity or a verified deep-research mode belongs on the shortlist. The original sources still need checking.
A defined multi-step deliverable
A business owner wants a tool to plan and assemble a report from approved, non-sensitive source files. Manus or Kimi K3 may be worth testing if the owner can review the plan, credits, permissions and result.
Use it responsibly
Three checks that do not change with the model.
🔒
Share the minimum
Give the tool only the information needed for the task. Remove names and identifying details wherever possible.
🔎
Check the evidence
A citation or confident answer is not proof. Open important sources and make sure they support the exact claim.
🧑
Keep approval human
Drafting and organising can be assisted. Sending, publishing, spending and consequential decisions stay behind review.
If the choice still feels muddy
Four quick checks.
“Every tool sounds the same.”
Your task may still be too broad. Name the exact deliverable, audience, source need and consequence of getting it wrong.
“The product recommended itself.”
Do not ask a product to judge its own suitability. Use the situational questions here and verify product claims in official documentation.
“A feature is missing.”
Your plan, account, region, device or rollout may differ. Check current official help before changing settings or paying.
“The sourced answer looks wrong.”
Open the original source. Check the date, author, context and whether it actually supports the specific sentence.
Your turn
Your five small missions.
0 / 5 complete
A sensible next move
Choose a tool. Build a system you can move.
Use a safe task to test the shortlist. Keep the brief, rules, examples and review process outside the product, so changing models does not mean starting again.