Which AI tool should I use?
There are dozens of AI tools, and every one of them claims to be essential. Most owners get stuck here — they compare features, watch reviews, and end up choosing nothing.
The honest answer: you don't pick the tool first. You pick the job — the one recurring task that eats your week — and then match a tool to that. Everything else is noise.
Name the job, not the tool about 30 minutes
Tool-picking fails when it starts from the tool and works backwards to a problem. Start from the problem — the weekly task you dread most — because a tool chosen for a real job either saves you hours or it doesn't, and you'll know which within two weeks.
How to find your job
- List the tasks you repeat every week. Drafting emails, answering the same customer questions, posting to social media, chasing invoices, booking appointments. Five to ten is normal.
- Score each one twice. A 1–10 for hours it costs you, and a 1–10 for how much you dread it.
- Pick the top scorer. That's your job. If two are close, pick the one with the clearest steps — the easier a task is to describe, the easier it is to hand to a tool.
Done when: you've written down one job and roughly how many hours a week it costs you.
Match the job to a tool type about 30 minutes
Every AI tool falls into a small number of types, and the type is decided by the job — not by the brand. Find your job in the first column, and you have your starting point. One job, one tool.
The map from job to tool
| Your job | The tool type that fits |
|---|---|
| Writing, drafting, summarizing | A general assistant — ChatGPT or Claude. Free tiers handle most of it. |
| Social posts, flyers, graphics | A design tool — Canva or Adobe Express. |
| Invoices, bookkeeping, chasing payments | The AI already inside your accounting software — QuickBooks or Xero. No new tool needed. |
| Missed calls, phone tag, after-hours lines | An AI receptionist — Rosie or Retell. |
| Documents, email, spreadsheets | The AI bundled with what you already pay for — Copilot for Microsoft 365, Gemini for Google Workspace. |
| Anything touching customer or financial data | AI that runs on your own machine — Ollama. Different rules apply; see the data guide. |
Two rules for reading the table:
- Prefer the tool already inside software you pay for. If the job lives in your accounting package or your email, the AI feature there beats a separate login — less setup, fewer passwords, no new data trail.
- Don't buy a specialized tool for a job a free general assistant can do. Try ChatGPT or Claude on the task first. Upgrade to a specialist only when the general tool clearly isn't enough.
Done when: you've picked one tool type — and one specific tool — for your job.
Test free, then measure 2 minutes a week
Run your real job through the tool for two weeks — not a tour of its features. Then answer one question: did it save you hours, yes or no?
How to run the test
- Sign up on the free tier. Don't enter a card. Nearly every tool worth using has a free option that covers a real test.
- Give it your real work. Paste an actual email you'd answer, an actual post you'd write. Generic prompts produce generic results, and then you'll blame the tool.
- Use it every time the job comes up for two weeks. Not once. Every time. The habit is the test.
- Count the hours at the end. How long did the job take before? How long now, including fixing its mistakes? If the fix time eats the savings, the tool hasn't earned anything yet.
If nothing changed after two weeks, the tool was wrong for the job — go back to the table and try the neighboring type. That's not failure, that's the process working.
Done when: you know the hours saved per week — or you've moved on to the next tool type.
Pay only when it earns it about 5 minutes
Free tiers are genuinely good now. A paid tier buys three things: privacy, volume, and better models under heavy use. Pay for the one you actually need — not for a badge.
When upgrading makes sense
- Your data is sensitive. Free consumer tiers often train on what you type. If the job touches client details, that alone can justify paying — or going local.
- You're hitting limits weekly. Message caps and usage quotas that interrupt real work are what paid tiers exist to remove.
- The hours are proven. Three saved hours a week at $75/hour is over $11,000 a year. A $20–30 subscription against that is an easy call. A subscription against unproven hours is not.
If the free tier quietly does the job, keep using it. Nobody awards points for spending money. For the full decision — the three things paying actually buys, the math, and when to skip the subscription and go local instead — see when to pay for AI.
Done when: your tool choice costs exactly as much as it earns — often nothing.
Two things to keep in mind
Some jobs change the rules. The moment a task touches customer records, health information, or financials, the question stops being which tool and becomes where does the data go. That's covered properly in the data safety guide — read it before putting client data into any cloud tool.
One tool beats a stack. Resist the urge to adopt three tools at once. Get one job running on one tool, prove the hours, and add the next only when the first one is boring you. That's the whole method.
Name the weekly job you dread most → match it to a tool type → test the free tier on real work for two weeks → count the hours saved → pay only if it earned it. The tool is the last decision, not the first.
