Choosing AI tools for ordinary office work.
There are thousands of AI tools and a new "essential" one every week. For ordinary office work you need roughly one and a half of them. Here is the map, the decision rule, and the four questions that filter everything else.
The AI tools question generates more anxiety than any other part of using AI at work - the sense that everyone else has found some tool you have not, and that choosing wrongly means falling behind. The truth is calmer: for everyday office work the tools have converged, the differences are smaller than the marketing suggests, and the skill transfers between them.
What AI tools do you actually need for office work?
For most office workers: one general assistant, whatever transcription is built into your meeting software, and any AI already inside your office suite. Start with what your employer has licensed, and add nothing else until a specific recurring task demands it. Then ask whether it beats the assistant you already have.
This is a guide for individuals choosing what to use day to day, not an enterprise procurement framework - if you are selecting AI for a whole business, our business adoption guide covers that. For one person with an inbox, the landscape has three layers.
What kinds of AI tool are there?
General assistants.
ChatGPT, Claude, Gemini, Copilot as a chat window. Drafting, summarising, explaining, cleaning, thinking. The Swiss army knife - and where the method lives.
Built-in AI - check the licence.
Gemini inside Docs, Sheets and Gmail is bundled with Workspace Business Standard and above, and Enterprise; transcription comes with Teams, Zoom and Meet. Copilot inside Word, Excel and PowerPoint is the exception: whether you have it depends on your tenant's Microsoft 365 Copilot licence label.
Point tools.
One job each: meeting notetakers, slide generators, image tools, research tools. Occasionally brilliant, usually duplicating layers one and two for a monthly fee.
The pattern that saves you money: capability flows downhill. Whatever a point tool does today, the general assistants and built-in AI have a strong record of absorbing - meeting summaries, image generation and document chat all started as standalone products. Every point tool subscription is a bet against that flow.
Which AI tool should you choose first?
If your organisation runs Google Workspace on a Business Standard or above plan, or Enterprise, Gemini is already included in Docs, Sheets and Gmail. Microsoft is messier, and this is the thing most people get wrong: your Microsoft 365 licence gives you Copilot Chat, a web chat window, and Copilot in Outlook, but Copilot working inside Word, Excel and PowerPoint now depends on your tenant's Copilot licence label - some organisations keep standard in-app access without the add-on, others lost it entirely in April 2026. So "do we have Copilot?" has more than one answer, and it is worth asking IT which label applies to you. Your meeting software probably transcribes. And every major assistant has a capable free tier. So the starting stack for almost everyone is: one general assistant, used well, plus whatever is already switched on in your office suite. Which general assistant? The honest answer is that for drafting emails, summarising documents and untangling spreadsheets, ChatGPT, Claude, Gemini and Copilot are all good and trade places with every release. Pick on placement and permission: the one closest to where your work happens, and the one your employer allows. The briefing method in our starting-point guide works identically in all four, which means switching later costs you nothing - the prompts in your notes file move with you.
Pay for a subscription when you hit the free tier's limits a few times a week. A personal plan for ChatGPT, Claude or Gemini runs at roughly £19-20 a month, with cheaper entry tiers now available from about £5-7, and against even 20 minutes saved per day that arithmetic is not close. Microsoft's in-Office add-on costs more again - about £23 per user per month at list - which is why it is normally an employer's decision rather than yours. Either way, paying before you have a daily habit buys you a subscription, not a capability.
How do you evaluate a new AI tool?
- 1. Does it serve a task I already do every week? A tool adopted for a real recurring task gets used; a tool adopted because a video called it a game-changer joins the graveyard by week three.
- 2. Is it allowed - by policy and for this data? Any tool that reads your inbox, joins your calls or connects to company files is a data decision, not a personal one. If IT approval exists, use it; if it does not, ask. The safety rules apply to tools even more than to prompts, because tools have standing access.
- 3. Does it clearly beat my general assistant at this task? Run the task both ways for a week. "Slightly more convenient" is not worth a login, a subscription and another place your data lives.
- 4. Does the price match the time saved? Minutes saved per week, times your rough hourly value, versus the monthly fee. Most point tools fail this test the moment you do the multiplication.
A no on any of the first three ends the evaluation. The fourth is where honest trade-offs live.
Red flags, and one honest caveat.
Be suspicious of: tools whose main feature is a library of prompts (you can build a better one free); "AI-powered" versions of things that were fine without it; anything demanding full inbox or drive access for a narrow job; and lifetime deals for products younger than your milk. The caveat: point tools genuinely win where the job is specialised and constant - heavy transcription across organisations, serious research workflows, design production. If a tool passes all four questions, adopt it without guilt. The category most worth watching rather than buying is the new wave of AI "agents" that operate software for you - explained here without the hype.
Escape the evaluation loop.
The most expensive tool habit is not paying for the wrong one - it is perpetual evaluation: trying a new tool every fortnight, migrating prompts and workflows, never staying long enough for a habit to compound. One assistant, used daily for three months, will change how you work. Twelve assistants, used for a week each, will change nothing except your password manager. Choose adequately, not perfectly, and put the energy into briefing and checking - the skills that survive every tool cycle.
Frequently asked questions
What AI tools do I actually need for office work?
For most office workers: one general assistant (ChatGPT, Claude, Gemini or Copilot), whatever transcription is built into your meeting software, and possibly the AI already inside your office suite. Most people need to add nothing else until a specific recurring task demands it.
Should I pay for an AI subscription for work?
Pay when you hit the free tier's limits often enough to cost you real time, roughly a few times a week. A paid assistant is among the cheapest professional tools you can buy, but paying before you have a working habit buys nothing.
What is the difference between ChatGPT, Copilot and Gemini for work?
Capability differences are modest and shift with every release. The practical difference is placement and licence: Gemini is bundled with Google Workspace Business Standard and above, and Enterprise; your Microsoft 365 licence includes Copilot Chat and Copilot in Outlook, while Copilot inside Word and Excel depends on your tenant's Copilot licence label; and ChatGPT and Claude are standalone. Pick whichever sits closest to where your work already happens and your employer allows.
How do I evaluate a new AI tool before adopting it?
Four questions: does it serve a recurring task you already do; is it approved for your data and by your IT team; does it clearly beat the general assistant you already have for that task; and does the price make sense against the time it saves? A no on any of the first three means don't adopt it.
The tool decision rule.
- Start with one general assistant plus whatever AI is already in your office and meeting software.
- Pick on placement and permission, not benchmarks - the skill transfers between tools.
- Pay for a subscription only once you hit free-tier limits a few times a week.
- New tools must pass four questions: recurring task, allowed for this data, beats your assistant, price matches time saved.
- Three months deep with one tool beats a week each with twelve.
Put AI to work in your own job.
AI at Work is Clerq's practical, hype-free guide to getting real work done with AI - written for people with inboxes, deadlines and meetings, not developers. PDF and EPUB, launching soon.