Choosing the Right AI Tool

Why this lesson avoids naming a winner

Any ranking of AI assistants has a shelf life of a few months. Providers leapfrog each other constantly; a lesson that told you “X is best for writing” would be wrong by the time you read it, and confidently so.

So this lesson teaches the criteria instead. Those hold.

The main options, in shape rather than in rank

Four broad categories, distinguished by what they are attached to rather than which is strongest today:

• General assistants (ChatGPT, Claude, Gemini and others) — standalone chat, strong across writing, analysis, coding and general questions. This is the category most people want.
• Assistants built into a suite (Microsoft Copilot, Google’s Workspace features) — weaker in isolation, but they sit inside the documents and mail you already have open. The integration often outweighs the model.
• Search-first tools (Perplexity and similar) — retrieve and cite sources by default. Better for questions with a factual answer; less suited to open-ended drafting.
• Specialists — coding assistants inside your editor, image tools, transcription tools. Narrow, and typically much better than a general assistant within that narrow band.

Free tiers across the general assistants are now genuinely capable. For most people they are enough.

Six criteria that stay true

1. Where does your work already happen? A slightly weaker assistant inside the app you have open beats a slightly stronger one behind a detour. Friction decides whether you actually use it.
2. What do you do most? Long documents, code, quick factual lookups and image work have genuinely different winners. Optimise for your commonest task, not your most impressive one.
3. Does it show sources? If most of your questions are factual, a tool that cites and links is worth more than one that is a bit more eloquent.
4. What are the data terms? For work material, this can be the deciding factor regardless of quality. Check whether your inputs are used for training and whether that can be switched off.
5. Does it handle your languages well? Quality varies considerably outside English. Test with your own material rather than assuming.
6. Can you leave? Can you export your history, or are you accumulating something you cannot take with you?

Conspicuously absent from that list: parameter counts, benchmark tables, and funding rounds. None of them predict whether a tool will be useful to you.

A fair comparison in twenty minutes

Do this rather than reading reviews:

1. Pick three tasks you genuinely repeat. Real ones from your week, not clever test questions.
2. Write each prompt once, properly — role, task, context, format.
3. Paste the identical prompt into two or three tools. Identical, or the comparison is worthless.
4. Judge on: did it understand? Would I have to rewrite it? Did it invent anything?
5. Repeat once a few months later. The answer will have changed.

Twenty minutes on your own work tells you more than any amount of reading, because it measures the only thing that matters — fit with what you actually do.

Free versus paid

Free is enough if you use it a few times a week for drafting, summarising, explaining and brainstorming, and you rarely hit a limit.

Paying is worth it when you hit usage limits during work you were going to do anyway, or you need a specific capability the free tier withholds — larger file uploads, longer documents, image generation, connected tools.

Paying does not make answers more factually accurate. Hallucination is not a premium feature that gets switched off. Higher tiers buy capacity and capability, not truth.

One economical pattern: keep one free general assistant for everyday work, and pay for a single specialist in whatever you do most.

The trap worth naming

The biggest waste is not choosing the second-best tool. It is switching constantly and never getting good at any of them.

Prompting skill transfers between tools, but tool-specific knowledge — how it handles files, where its settings live, what it is quietly bad at — takes weeks to build and does not transfer. Pick something adequate and stay put for a few months. Revisit deliberately, not every time a launch goes viral.

Try this: Take one prompt you have already written and run it through two different assistants side by side. Do not judge which reply sounds better — judge which one you would have to rewrite less. That is the only comparison that predicts anything.