How to use AI at work in a small business
To use AI at work in a small business, start with one task that eats your time, check what your office suite already includes, decide what data never goes into a chatbot, and use free plans before paying. Then write one page of rules for the team. Official guidance in the UK, Canada, Australia, and the US agrees on the basics: keep sensitive data out of public tools, and check every answer.
How to use AI at work when you run a small business
Your team is probably using AI already. In March 2026, about 56% of US workers told the Census Bureau they had used it on the job for at least one task, most often to look up information, write, come up with ideas, or summarize and translate. For a freelancer or a small business, the question is no longer whether to use AI, but how: which tool, on which account, with which data, and with what checks before anything reaches a client.
Business surveys show slower adoption, measured differently in each country. In the UK, 30% of businesses used at least one AI technology in September 2026, and 29% of those with fewer than 10 employees, according to the Office for National Statistics. In Canada, 19.2% of businesses had used AI to produce goods or services over the previous 12 months in Statistics Canada’s survey for the second quarter of 2026, and businesses with one to four employees were at 19.9%. The questions differ, so the figures do not compare across countries, but they agree on one point: the smallest firms are not far behind the rest.
This guide is for the person who has to decide. That might be a freelance bookkeeper choosing a subscription, or the owner of a five-person design studio who just realized that three people are pasting client emails into three different chatbots. It covers the five assistants most small businesses meet, ChatGPT, Claude, Gemini, Microsoft Copilot, and Perplexity, and what each means for a business without an IT department.
Throughout this series, we compare products only on what their makers publish, ask the same questions of each, and never rank them on quality without an independent test whose method is published.
Why it is harder than it looks
Choosing an AI tool is harder than choosing most software, for four reasons.
Names and plans change every few months. OpenAI renamed ChatGPT Team to ChatGPT Business on August 29, 2025. Microsoft stopped selling Copilot Pro on October 1, 2025, and now calls its Microsoft 365 Copilot app simply Microsoft Copilot. Google renamed NotebookLM to Gemini Notebook. Advice from a colleague last year may name products that no longer exist, which is why every plan name and price in this series is dated.
You may already be paying for AI. Google added Gemini to its Workspace Business and Enterprise plans on January 15, 2025, and raised their prices at the same time: in Google’s own example, Business Standard went to $14 per user per month, $2 more than without Gemini. Microsoft added Copilot to Microsoft 365 Personal and Family on January 16, 2025, with a $3 monthly increase in the US. In Australia, the ACCC took Microsoft to court on October 27, 2025, alleging that it misled about 2.7 million subscribers about the existence of cheaper “Classic” plans without Copilot. These are allegations, and we found no outcome as of November 10, 2026.
The account matters as much as the tool. On their personal plans, ChatGPT and Claude may use your conversations to train their models unless you turn that off; their business plans do not by default. With a personal Google account, some Gemini conversations are read by human reviewers, including contractors, while conversations from Workspace accounts are not reviewed or used to improve the models, according to Google. Microsoft says the interactions of its business users are not used to train foundation models. The same question typed into the same assistant is handled differently depending on which account you are signed in to.
They make things up. The US National Institute of Standards and Technology calls it confabulation: “confidently stated but erroneous or false content.” Its generative AI profile explains that this is a natural result of how the models work, by predicting the next word, and that the risk grows with long answers and specialized topics. OpenAI, Anthropic, Google, and Microsoft all warn, in writing, that their assistants can be wrong.
In practice, treat an AI answer like a first draft from a quick, capable new hire: useful, fast, and in need of checking before it goes out under your name. Facts, figures, quotes, and anything legal or financial need a source you have opened yourself.
Where to start
Start with a task, not a tool. Look at the tasks the Census survey found most common, research, writing, ideas, and summaries, and pick the one that costs you the most hours each week. Time it for a week without AI, then for a week with it. If the second number is not smaller, the tool is not the answer for that task, whatever its reputation.
Check what you already pay for. If your email runs on Google Workspace Business, Gemini is part of your plan, in Gmail, Docs, Sheets, Meet, and its own app. If it runs on a Microsoft 365 business plan, Copilot Chat is included with eligible licenses: it answers from the web and from the files you attach. Microsoft’s paid Copilot license goes further and reads your organization’s email, files, and meetings to answer. Testing what you have costs nothing.
Decide what never goes in. The authorities of all four countries give the same advice. The UK’s NCSC said in 2023 not to include sensitive information in queries to public chatbots. Canada’s privacy commissioners, in their principles for generative AI, say to enter personal, sensitive, or confidential information only where authorized. Australia’s OAIC recommends not entering personal information into publicly available AI tools, and the US Small Business Administration warns against sharing sensitive or proprietary data. For a small business, that usually means no client personal data, no passwords, no bank details, and nothing covered by a confidentiality agreement, unless you are on a business plan whose terms you have read.
Use the free plans first. Every assistant covered here has a free plan, and web search with sources is now standard. Some free plans go further: Gemini’s includes Deep Research reports and image generation. Free plans also have limits that change often, and some show ads: ChatGPT may show them on its Free and Go plans in some regions, while Anthropic has committed to keep Claude ad-free. Pay when a limit starts to cost you time, not before.
Write one page of rules. Canada’s Cyber Centre recommends a plan and a policy for AI use, and the SBA suggests having another person review AI-generated content before it goes out. One page is enough for a team of five: which tools, which accounts, what never goes in, and who checks.
Your AI starter sheet
Copy this into a note and fill it in before anyone signs up for a new tool. It takes 20 minutes and answers the questions that otherwise come up one awkward client email at a time.
| Question | Your answer |
|---|---|
| Which task do we try first, and how long does it take today? | |
| Which tool, and on which account: work or personal? | |
| What must never be pasted in? | |
| Who checks the output before it reaches a client? | |
| What does it cost per month, and when do we review that? | |
| Who decides when someone wants to try a new AI tool? |
Choosing your assistant
ChatGPT, Claude, Gemini, Copilot, and Perplexity overlap more than their marketing suggests, and differ in a few ways that matter: images, ads, team plans, and how they fit your office suite. Our guide to choosing an AI assistant for your business compares them on published facts, with a decision table and a one-week test.
What the plans cost
Prices differ by country, and every company now sells several tiers between free and enterprise. Our comparison of what AI assistants cost in each country sets out the plans in the US, the UK, Canada, and Australia, explains the team plans, and shows how to cancel.
Your data and the rules
Training settings, business and personal accounts, privacy law in each country, who owns what AI produces, and a template for a team AI policy are covered in our guide to what AI does with your data.
AI for everyday tasks
Emails, summaries, spreadsheets, and prompts that produce usable work, along with the checks that catch mistakes, are the subject of our guide to getting real work out of AI.
Meeting notes and transcription
AI note takers are built into Teams, Google Meet, and Zoom, and sold as separate apps. Recording a meeting also raises consent rules that differ by country and, in the US, by state. Our guide to recording and summarizing meetings covers both.
Automation and AI agents
Tools such as Zapier, Make, and n8n, and agents that act on your computer, can take routine work off your hands. They also act with your permissions, which deserves care. Our guide to automation tools for a small team compares the platforms and suggests a few safe first automations.
What to do this week
- Pick one task and time it.
- Check whether your Google Workspace or Microsoft 365 plan already includes an assistant.
- Fill in the starter sheet above with your team.
- Read the joint guidance on engaging with AI published by Australia’s ASD with CISA, the NCSC, and others: it is short, practical, and written for organizations of any size.