Ask five small businesses what AI tools they use, and you get a list of a dozen subscriptions. Ask what they used last week, and you get two. That gap is not a tool problem. As of late August 2026 the tools are cheaper and better than ever — the failure is almost always sequence: teams buy tools before they pick the job. The teams that get real money back from AI choose one function, standardize how it is done, and only then add the next tool. This guide follows that order — by job, not by brand — and uses only what we could verify from official pricing pages and our own testing. Starting from $0? Our guide to free AI tools that are actually worth using covers the free tiers in depth; this one is about what changes when the work is for a business.
Writing: the fastest win in the stack
Writing is where most teams should start: it is the cheapest to fix and the easiest to measure. ChatGPT’s free tier handles the daily bulk — product descriptions, email drafts, proposal structure — and the paid lineup now runs Go, Plus, Business, and Enterprise, with the newest model, GPT-5.5, arriving on paid plans first (openai.com/chatgpt/pricing). The Business tier matters once more than one person shares an account: it adds admin controls and data terms for when customer information enters prompts.
Official image from Google.
Gemini is the least dramatic choice and, for a team that already lives in Google Docs and Gmail, the most useful one: “Help me write” sits inside the document, and summarizing a Gmail thread or drafting from a Doc happens without copying between apps. Claude earns its $20/month when the deliverable is long and has to sound like one person wrote it — a proposal, a handbook, a case study — because it holds a consistent voice across a whole document.
The habit that separates teams that get value from teams that get drafts: keep a shared prompt file — your house style, your product facts, three or four prompts that actually work. It turns AI writing from a personal toy into a team asset.
Support: let the bot take the first reply
Support is where AI has changed the economics, not just the typing. The biggest real change in 2026: Intercom’s Fin agent is now sold standalone (fin.ai), priced per resolution — from $0.99 per Fin outcome, plus seats (intercom.com/pricing) — and Intercom backs it with a “Fin million dollar guarantee.” The honest math for a small team: a bot that answers the same three questions forty times a day pays for itself; a bot that touches sensitive account data without a review workflow is a liability before a saving.
For a three-person team the right order is usually: shared inbox, ChatGPT drafts for replies, a human presses send. Only when the same questions repeat daily does a dedicated agent earn its cost — and then the real job is writing the answer library it learns from.
Data: the spreadsheet is the interface
The most underrated business use of AI is asking questions about data you already have. Upload a CSV to ChatGPT’s free tier and ask plain-language questions — which customers churned, which product line is dragging, what the refund pattern looks like. If your data lives in Google Sheets, Gemini works inside the document; on Microsoft 365, Copilot sits in Excel.
The rule that keeps this safe: use AI for questions, not for final numbers. A model that is confident and wrong in a forecast costs you a quarter. Ask it to explain the outliers, then check the totals yourself — it is a fast analyst with a hallucination problem, not an auditor.
Meetings: the notes write themselves
Meeting notes are the least glamorous business task and the one AI is best at. Otter’s free tier gives 300 transcription minutes per user per month, with a 30-minute cap per conversation, and it joins Zoom, Teams, and Google Meet to write summaries with action items automatically; paid tiers raise the limits and add custom vocabulary (otter.ai/pricing). For most small teams the free tier is enough to standardize one habit: a summary for every meeting, in the same place.
Official product image from Sony (sony.jp).
We tested this category on hardware too. In our hands-on review of the Sony WH-1000XM6, the headphones sat through a 45-minute mixed Chinese-English conference call and the app auto-generated the minutes — agenda, conclusions, and action items separated, key figures mostly accurate, one currency unit wrong. That detail is the whole lesson: even a good AI note-taker gets a number wrong now and then, so the summary is a starting point, not a record. Meeting transcription has commoditized so far that Otter’s pricing page now also sells an AI sales agent that runs personalized video calls with site visitors.
The pitfalls that sink AI rollouts
Three mistakes end more business AI projects than bad tools do.
Treating a consumer chatbot as a data processor. Paste customer names, contracts, or payroll into a consumer free tier and you hand data to a product you do not control. Business tiers exist for this reason. Before anyone puts a customer record in a prompt, someone should be able to answer where the data goes and what the vendor does with it. If the answer is vague, it does not go in.
Letting confident output ship unread. Hallucination is not a rumor, it is a feature of current models. In customer-facing output — support replies, product listings, meeting minutes — a human reads before it ships. The Sony test caught a wrong currency unit; a support bot that invents a refund policy is worse than no bot.
Subscribing before standardizing. The “used it last week” test is brutal and correct: any tool nobody used in seven days is misconfigured or unnecessary. Cancel it, fix the workflow, revisit.
A starter stack that costs about $0
| Job | Starter tool | Free entry | The limit that matters |
|---|---|---|---|
| Writing | ChatGPT or Gemini | Full free tier | Newest model is paid-first |
| Support | Shared inbox + AI drafts | Free assistants | Human reviews every reply |
| Data | ChatGPT uploads, Gemini in Sheets | Included in free tiers | Verify every number yourself |
| Meetings | Otter | 300 min/user/month | 30-min cap per conversation |
The implementation advice is deliberately boring: pick one function — writing is easiest — and run it consistently for two weeks before adding another. Make someone responsible for reading AI output before it reaches a customer. When a free tier starts costing you time every week, that is the moment to pay for a Business plan, not before.
FAQ
Is it safe to put customer data into AI tools? Not into consumer free tiers. Use a Business or Enterprise plan with admin controls, and check the vendor’s data terms first. If “where does this data go” has no clear answer, keep it out.
What is the cheapest way for a small team to start? One free writing assistant, one free meeting transcriber, and a shared document with your house style and working prompts. That is the whole stack until a limit costs you time.
Which AI tool should a small business pay for first? The one your team hits limits on weekly — usually a writing assistant like ChatGPT Plus or Business, or the meeting transcriber if you live in calls. Pay for the bottleneck, not the hype.
Do these tools replace staff? No. They remove typing, not judgment. Someone still writes the answer library, checks the numbers, and decides what ships. Teams that treat AI as a junior employee with a review process get the value; teams that treat it as a magic button get the invoices.
Free-tier allowances and prices checked against official pages (openai.com/chatgpt/pricing, otter.ai/pricing, intercom.com/pricing) in late August 2026. Plans change often — confirm current terms before subscribing.