AI Tool Contract Checklist for Small Business: 7 Clauses to Check Before You Subscribe

Most small businesses sign up for an AI tool the way they buy a coffee: card in, click, done. The terms behind that click decide whether your client files can train someone else's model, who pays if generated content triggers a copyright claim, and what happens to your price at renewal. You do not need a lawyer to check the basics. You need an AI tool contract checklist and about fifteen minutes.

This guide gives you seven clauses to check, where to find each one, and what a red flag looks like. It also covers what a small business can realistically change, because on most self-serve plans you cannot negotiate a word. This is general information, not legal advice, so for anything high-stakes, such as client contracts or regulated data, have a lawyer read the actual terms.

The AI tool contract checklist at a glance

Open the vendor's terms of service, privacy policy and pricing page. Then look for these seven things.

The rest of this guide takes them one at a time, in order of how often they cause real trouble.

What a small business can and cannot change

Be realistic. A $20 monthly plan with a pricing page is a take-it-or-leave-it contract. What you control is smaller but still valuable:

• Which plan you buy. The same tool often has different terms on personal and business plans, which is the single biggest lever (see clause 1).
• The settings. Training opt-outs and retention controls are usually a toggle.
• The billing period. Monthly keeps you free to leave; annual saves money but locks you in.
• A two-line email. On a larger or annual deal, asking for a price cap or a different term costs you nothing.

Everything else on this list is about knowing before you commit, so the choice is yours rather than an accident.

Clause 1: Can the tool train on your data?

This is the check that matters most. Many AI tools train on what you type unless you are on a business plan or switch it off. Take ChatGPT, whose training rules OpenAI documents in its help center:

A thumbs up or thumbs down on a reply can allow the whole conversation to be used for training even after you opt out. For context, the law firm Proskauer wrote in 2023 that it had become "off-market" for a provider to train on customer data without at least offering an opt-out (Proskauer, October 2023). That is a law firm's view from a few years ago, so verify each vendor's current terms and note the date you checked.

If a tool touches client-confidential material, move to a business plan or turn training off before the first file goes in. The same rule is the heart of our shadow AI audit, where we cover how to find the tools your team already uses without telling you.

Clause 2: Who is covered if the output triggers an IP claim?

Some vendors promise to defend paying customers if AI output is claimed to infringe someone's IP. Most do not extend that to free or consumer plans. The firm Proskauer summarized the main announcements at the time. These are dated reports, so check the vendor's current terms before relying on any of them:

Every one of these has carve-outs, for example claims tied to your own prompts, modifications you make or combining outputs with other material. Proskauer's note on Anthropic also says Google, Adobe, Shutterstock and IBM have rolled out similar protections. The takeaway is simple: if you publish AI-generated content under your name or in paid ads, check this clause for the tool and the plan you actually use. Internal drafts matter far less. If you generate images, our guide to the best AI image generators compares the tools, and the cover question belongs next to the price.

Clause 3: Price increases and automatic renewal

Search the terms for "adjust", "index", "increase" and "renewal". You are looking for two things: whether the vendor can raise the price during your term, and whether the plan renews automatically at a new price. Annual plans are where this bites, because the notice window can be as short as a few weeks.

If you find an increase clause on a plan you care about, ask for a fixed cap or for the clause to be removed. Vendors say no often enough on small plans, but the request costs nothing, and on larger annual deals it sometimes works. Put the renewal date in your calendar 30 days ahead either way. Our guide to the hidden cost of choosing the wrong AI tool shows how these small charges add up.

Clause 4: Usage limits and overages

Many AI tools have a monthly allowance of messages, credits or minutes, and charge more when you go over. Check three things: what counts toward the limit, what happens at the limit (hard stop, slower service or a surprise bill), and whether you can set a spending cap. Tools that bill per use deserve extra care, because a busy month can double the invoice. If a tool offers no cap and no alert, treat that as a reason to start on the smallest plan.

Clause 5: Term, cancellation and getting your data out

Start monthly, not annual, for anything you have not used through a full billing cycle. The saving from annual billing rarely justifies a lock-in on a tool you have not stress-tested. Enterprise advisory data points the same way: the advisory firm Redress Compliance reported from its own engagement files that 20 to 30 percent of platform commitments went unconsumed at the end of the term, and that seat-based assistants settled at 40 to 60 percent of initial projections within two quarters (Redress Compliance). That is a vendor-advisory sample of about thirty enterprise deals, so read it as a pattern rather than a statistic about small businesses. The pattern is still useful: people buy more than they use.

Before you pay, check how you cancel (button, email or phone), whether you can export your data in a usable format, and what happens to your content after you leave. A law firm's checklist for AI contracts also suggests agreeing up front on your right to keep using AI-generated content after termination (Harris Beach Murtha, August 2024).

Clause 6: Data retention and deletion

Find out how long the vendor keeps your inputs and outputs, whether you can delete them, and whether deleting your account removes them from backups or logs. Vague language like "as long as necessary" is a yellow flag, not a deal-breaker, but it means you should keep sensitive material out of the tool. If you must handle regulated data, ask directly whether the vendor can document its controls. A compliance platform such as Vanta is aimed at companies proving those controls, which is a sign of what larger customers will ask you for in return.

Clause 7: Liability and who owns the outputs

Two questions here. First, who owns what you put in and what comes out? The Harris Beach alert recommends that the company own its input data and, preferably, the outputs. Second, how much can the vendor be held responsible for? Most terms cap liability tightly and disclaim output errors. That is normal, and it is why you should never publish or send AI output without a human check, especially for legal, medical or financial content.

A 15-minute review routine

Run this for any tool that touches client work or goes on a company card.

1. Find the documents (2 minutes). Terms of service, privacy policy, pricing page and any data-processing page.
2. Search the text (5 minutes). Use your browser's find box for: train, retain, indemn, renew, increase, cancel, liability.
3. Check the plan (3 minutes). Confirm the terms you read apply to the plan you will actually use, not just the enterprise one.
4. Set the switches (3 minutes). Turn off training if you can, set billing to monthly and note the renewal date.
5. Write it down (2 minutes). One line per tool in a shared sheet: plan, date checked, what you changed.

If a clause worries you, send a short email before you pay. Something like this is enough:

Hello, before we subscribe to the annual plan, can you confirm three things in writing? 1) Is our content excluded from model training on this plan? 2) Is the price fixed for the 12-month term, and how much notice do you give before renewal? 3) Can we export our data if we cancel? Thanks.

A written answer you can screenshot is worth more than a verbal one, and a vendor's response tells you how they will behave when something goes wrong.

Which clause matters most for which tool?

The riskiest clause depends on what the tool does with your content.

This table says where to look first, not that these tools are good or bad. Their terms differ by plan and change over time, so check the live pages. Our guide to free versus paid AI tools covers what you give up on a free plan, and one answer is often exactly these protections.

Final thoughts

None of this takes a legal department. It takes the habit of reading six or seven lines before you subscribe and writing down what you found. The checklist protects you from three expensive surprises: your data training someone else's model, an IP claim with nobody to defend you, and a price rise you did not see coming.

Once your tools pass the check, the next question is which ones are worth buying at all. Our guide to the best AI tools for small business covers that, and how to choose the right AI tool for your business helps you decide between close options. If you want to see what your team is already paying for, start with the shadow AI audit.