Free AI tools can take on a lot in a small business: drafting posts, answering spreadsheet questions, summarizing short documents, brainstorming. What matters is how much you can use before you hit a cap, which changes often, and what happens to your conversations. "Free" also means something different for a ready-made app and for an "open" model you would have to install and maintain yourself. This guide sticks to what each company's official pages say and does not pick a "best" tool.
What free plans typically allow
The best-known assistants all have a free plan, and all of them cap it. The table uses the same criteria for each one, in alphabetical order, with no ranking. There are no message counts on purpose: the companies themselves say the numbers vary.
| Assistant (company) | What the official page lists for free | How the cap works |
|---|---|---|
| ChatGPT (OpenAI) | Web search, data analysis, image and file uploads, image creation. Ads may appear in some countries. | Everyday text chat has no fixed ceiling, with abuse-prevention safeguards. File uploads, images, voice and data analysis have separate limits. The page warns that the default model and the limits can change. |
| Claude (Anthropic) | Chat on web, desktop and mobile, web search, file creation, code execution and memory across conversations. | A session limit that resets every five hours, plus a weekly limit. The number of messages varies with demand, the model, chat length and file size. |
| Copilot (Microsoft) | For businesses, Copilot Chat is included with a Microsoft 365 subscription, with limited access to work data (uploaded files only) unless an add-on license is assigned. | Without an add-on license, priority is "standard" and may be limited at peak times. On individual subscriptions, limits depend on the plan and may change as new models arrive. |
| Gemini (Google) | A lighter model, image generation and editing, Deep Research, Gemini Live, Canvas and Gems. Video generation is on paid plans. | Limits are based on compute use and refresh every five hours up to a weekly cap. Google says limits may change without notice and that people without a plan may be limited first. |
Sources: the help centers and plan pages of OpenAI, Anthropic, Microsoft and Google, listed at the end.
Three catches come up again and again. The cap is about consumption, not "messages": big files, long chats and research features use more. When demand is high, free users can be limited first. And advanced features tend to sit behind a paid plan.
Free app or open-weight model: not the same thing
A free app is a finished service: you open the site or app, chat, and the company runs the servers. An open-weight model is a file a company publishes so anyone can download it and run it somewhere else. The download is free. Running it is not.
- Hardware. On its gpt-oss page, OpenAI says the larger model runs on a single 80 GB GPU and the smaller one on devices with 16 GB of memory, with weights to download under an Apache 2.0 license. An 80 GB card is specialized equipment, far from an office PC.
- Technical work. Someone has to install, update, secure and monitor it.
- License. "Open" does not mean "no rules": each model has its own license, so read it before commercial use. Under the Open Source Initiative's definition, open source AI must allow use, study, modification and sharing, with information about the training data, the code and the parameters. Publishing weights alone does not meet it.
- The upside. If you run it on your own equipment, the conversation does not pass through another company's servers. In return, security becomes your job.
For most small businesses, a ready-made app is the realistic route. If a service hosts an open model for you, that host's prices and data rules apply, not those of the company that built the model.
What happens to your data on free plans
This is the question that matters most. Personal accounts and business accounts usually follow different rules. According to the official pages:
| Assistant | Can chats improve the models? | How to change it |
|---|---|---|
| ChatGPT (OpenAI) | For individual services, OpenAI says it may use your content. For business services (Team, Enterprise and the API), it does not train by default. | ChatGPT data controls or the privacy portal, applying to new conversations. Temporary Chat stays out of history and training. |
| Claude (Anthropic) | On personal accounts, it depends on a privacy toggle. On Team, Enterprise and the API, chats are not used for training unless the customer joins a partner program. | Settings, Privacy tab, switch off the toggle that helps improve the models. It applies to new chats; data already used stays. Chats flagged by safety classifiers may still be used. |
| Copilot (Microsoft) | On personal accounts, Microsoft says it uses Copilot data for training, except for signed-out users, people under 18 and work or school accounts. For those, Microsoft says prompts and responses do not train models under enterprise data protection. | Signed-in users can turn training off for future conversations. The page warns that part of it applies only to the older app version, so follow the page for yours. Human review of some cases cannot be declined. |
| Gemini (Google) | With app activity on, Google uses chats to improve its services, including training AI. A subset of chats is read by human reviewers. | Turn off activity (Keep Activity in English): new chats stop training but are kept for 72 hours. Reviewed chats can be kept up to three years even if you delete them. Temporary Chats do not train. Work or school accounts may follow other rules. |
Turning training off does not solve everything: companies still keep conversations for a while and may review specific cases. Google even asks people not to enter confidential information they would not want a reviewer to see. OWASP's guidance on sensitive information disclosure makes the same point for anyone using these tools: avoid typing sensitive data, and favor products that let you opt out of training. And do not assume the terms stay put. The FTC has warned that quietly changing terms of service to use customer data for AI training may be unfair or deceptive, but the safer habit is to recheck the page yourself.
Everyday tasks: is free enough?
This table is practical guidance, not test results. It depends on volume and on what you put in the chat.
| Task | Free is enough when | Consider paying when |
|---|---|---|
| Captions and posts | You write a few per week. | You produce daily and keep hitting the cap. |
| Standard replies to customers | Templates are generic, with no customer names or orders. | Customer data goes in: prefer a business plan with written data terms. |
| Summarizing documents | Short, non-confidential text. | Long files, contracts or confidential documents. |
| Spreadsheet formulas | One-off questions with made-up data. | Frequent analysis of large sheets, which eats file limits. |
| Names, slogans and campaign ideas | Almost always. | Rarely. |
| Translating short texts | Messages and product descriptions. | Official or legal documents: hire a translator. |
| Promotional images | A few images per week. | High volume or video, which tends to have its own limits. |
| Customer chatbot on your site or in a messaging app | Not a use case for a free app. | Always: it needs a business product with its own terms. |
| Financial, HR or health data | Never with real data. | Only with a business plan and an internal policy; sometimes, no AI at all. |
Checklist before you adopt
- Define the task and the volume. Test for a week on the free plan and note when you hit the cap.
- List what never goes in the chat. Passwords, cards, customer data, confidential documents.
- Read the official privacy page and set the training option before first use.
- Check the terms of use to see whether the free plan allows commercial use.
- Use a business account with two-step verification, not an employee's personal one.
- Decide who checks the output. Numbers, prices, deadlines and legal text go past a person.
- Have a plan B. Keep your prompts in a document so you can switch tools if limits tighten or a plan changes.
- Write a one-page rule for your team and revisit it now and then, because terms change. See also more free tools for small businesses.
Frequently asked questions
Is there really free AI a business can use?
The main assistants all offer a free plan, but with usage caps that change and some features reserved for paid plans. Also check the terms of use to see whether the free plan allows commercial use.
Can I put customer data into a free AI tool?
The safest answer is no. On personal accounts, companies may use conversations to improve models unless you switch that off, and some data may still be stored or reviewed. For customer data, prefer a business plan with written data terms.
Is an open-weight model better than a free app?
Neither is better, they are different things. The model file is free, but running it takes hardware, technical work and responsibility for security. For most small businesses a ready-made app is the more realistic route.
When is it worth paying for an AI plan?
When you hit the cap often, when you need data protection spelled out in a contract, or when you rely on features that only exist on paid plans. Compare the cost with the time the tool really saves.
Sources: OpenAI Help: Using ChatGPT's Free plan · OpenAI: How your data is used to improve model performance · Anthropic: About free Claude usage · Anthropic: Plans & pricing (Free plan) · Anthropic Privacy: How do I change my model improvement privacy settings? · Anthropic Privacy: Is my data used for model training? (commercial products) · Google: Gemini Apps limits & upgrades for Google AI subscribers · Google: Gemini plans (subscriptions page) · Google: Gemini Apps Privacy Hub · Microsoft: Privacy FAQ for Microsoft Copilot · Microsoft Learn: Frequently asked questions about Copilot (Copilot Chat at work) · Microsoft: How AI usage works across Microsoft 365 subscriptions for individuals · OpenAI: Introducing gpt-oss (hardware requirements and license) · Open Source Initiative: The Open Source AI Definition · FTC Office of Technology: Consumer Facing Applications, a Quote Book from the Tech Summit on AI · OWASP GenAI Security Project: LLM02:2025 Sensitive Information Disclosure
