Alex is Sprintlaw’s co-founder and principal lawyer. Alex previously worked at a top-tier firm as a lawyer specialising in technology and media contracts, and founded a digital agency which he sold in 2015.
Launching an AI web app is exciting, but your customer-facing terms can feel tricky once prompts, outputs and model errors enter the picture.
Generic website terms usually don't go far enough for AI products. They might cover payment, accounts and refunds, but miss the clauses that matter most when users are uploading content, generating outputs and making decisions based on what your tool produces.
The good news is that your terms don't need to be intimidating. They just need to be clear about what your AI tool does, what users can and can't do with it, and where responsibility sits if something goes wrong.
Do You Need Separate Terms Of Service For An AI Tool?
Yes, you usually need terms that are tailored to your AI tool, rather than relying on generic website terms of use.
Your terms of service are the contract between your business and your users. For an AI app, that contract has to deal with a few things that ordinary SaaS terms often don't cover well, including:
- what users are allowed to upload as prompts or inputs
- who owns or can use the outputs your tool generates
- whether outputs can be inaccurate, incomplete or similar to other content
- what types of use are banned, such as fraud, harmful content or unlawful surveillance
- what happens if a user relies on an output and suffers a loss
- how you can suspend accounts or remove content that breaches your acceptable use rules
If you're selling to UK consumers, your terms also need to sit within the Consumer Rights Act 2015. For non-subscription distance sales and digital content contracts, the Consumer Contracts (Information, Cancellation and Additional Charges) Regulations 2013 may also be relevant. That means your terms must be fair, transparent and not take away statutory rights.
If your AI tool is sold to consumers as a monthly, annual or other repeating subscription, you also need to factor in the new subscription contracts regime under the Digital Markets, Competition and Consumers Act 2024 and the Consumer Subscription Contracts (Cancellation and Information) Regulations 2026. Official guidance is now available, and the regime is due to apply from 1 January 2027. In practice, B2C subscription terms and checkout flows need to address pre-contract information, renewal and reminder notices, and cancellation requirements.
If you're selling to businesses, you have more room to negotiate limits on liability, but those clauses still need to be drafted carefully. Under the Unfair Contract Terms Act 1977, some exclusions and limitations need to be reasonable to be enforceable.
Templates can be a helpful starting point, but AI tools often need extra drafting around prompts, outputs, misuse and liability. This is where tailored terms can save you a lot of back-and-forth with customers later.
What Should Your AI Terms Say About Prompts And Outputs?
Your terms should clearly explain what users are responsible for when they enter prompts, and what rights they get in the outputs your tool generates.
Prompts are often user generated content. A prompt might include text, code, documents, images, personal data, trade secrets or someone else's copyright material. Your terms should make it clear that users are responsible for having the right to upload that content.
For prompts, consider covering:
- Permission: the user must own or have permission to use anything they upload.
- Restrictions: users can't upload unlawful, confidential, infringing, harmful or misleading material.
- Personal data: users must not upload personal data unless your product and privacy documents allow it.
- Security: users are responsible for keeping login details secure and controlling who can submit prompts through their account.
- Model improvement: if you use prompts or outputs to improve your system, say so clearly and make sure this matches your privacy notices and data processing arrangements.
For outputs, be upfront that AI-generated material may not be accurate, complete, unique or suitable for every purpose. If your app helps with legal, financial, medical, employment or other high-impact decisions, you should be especially careful about how the product is described and what review obligations sit with the user.
For example, if your tool generates marketing copy, your terms might say the user must review outputs for accuracy, legal compliance and brand suitability before publishing. That matters because UK advertising rules still apply even if an AI tool wrote the first draft.
How Should You Deal With Misuse And Prohibited Uses?
You should deal with misuse by setting clear acceptable use rules and giving your business the right to suspend, restrict or terminate accounts that break them.
AI tools can be used in ways you didn't intend. Some misuse is obvious, like generating scams, malware or abusive content. Other misuse is more subtle, like scraping personal data, impersonating another person or using outputs to make discriminatory decisions.
Your terms should include a prohibited use clause that matches your product's risk profile. Common restrictions include bans on using the service to:
- break the law or encourage unlawful conduct
- infringe copyright, trade marks or confidentiality obligations
- generate spam, phishing messages, malware or deceptive content
- harass, threaten, defame or discriminate against people
- impersonate another person or mislead users about whether content was AI-generated
- reverse engineer, scrape, overload or interfere with your systems
- use the tool for regulated or high-risk decisions unless you have expressly allowed that use
Your acceptable use controls shouldn't just sit in the contract. They should also be reflected in your product design where possible. That might mean rate limits, content filters, account verification for certain features, audit logs, user reporting tools or manual review for risky use cases.
If your users can publish, share or store user generated content through your platform, your terms should also say when you can remove content. This helps you act quickly if a user uploads infringing material, personal data they shouldn't have, or content that puts your business or other users at risk.
Who Owns The IP In AI Outputs?
It depends on your product, your model provider terms and the law that applies to the content, so your terms should avoid overpromising ownership of AI outputs.
This is one of the clauses generic AI terms often get wrong. Users usually want to know whether they can use outputs commercially. You may be comfortable granting that right, but you should be careful about saying an output is definitely protected by copyright or definitely owned by the user.
Under UK copyright law, protection usually depends on originality and authorship. AI-generated content can raise difficult questions, especially where human input is limited. Your terms can still grant users permission to use outputs, but they should explain the practical limits.
A sensible output IP clause might say:
- the user keeps ownership of their prompts, subject to any licence they give you to operate the service
- your business owns the platform, software, models, branding, documentation and underlying technology
- the user can use outputs for their business purposes, subject to the terms and any plan restrictions
- you don't guarantee that outputs are unique, protectable by IP rights or non-infringing in every case
- the user must review outputs and avoid using them in a way that infringes third-party rights
If you use a third-party model provider, check its terms before promising anything to your customers. Some providers allow broad commercial use of outputs. Others may have restrictions for sensitive sectors, safety rules or limits on training use. Your customer terms should line up with those upstream obligations.
If your brand name, logo or product name is central to your launch, you should also think about trade mark protection through the UK Intellectual Property Office. Your terms can protect how users interact with your service, but they don't replace a trade mark registration strategy.
How Do AI Liability Clauses Work?
AI liability clauses work by explaining what your business is responsible for, what the user remains responsible for, and what losses are limited or excluded where the law allows.
For an AI tool, the key is to be honest about the role of the product. If your product is a drafting assistant, your terms shouldn't make it sound like a final decision-maker. If your tool analyses documents, your terms should explain that the user must check the output before acting on it.
There are some liabilities you can't exclude under English law. For example, you can't exclude liability for death or personal injury caused by negligence, fraud or fraudulent misrepresentation. Consumer terms also can't unfairly remove rights under the Consumer Rights Act 2015.
A Worked Example
Say Maya runs a UK AI copywriting app for small eCommerce brands. Customers pay GBP 29 per month under subscription terms. One business user generates an advert saying a skincare product is clinically proven to remove acne in 7 days. The claim isn't true, but the user publishes it without checking and receives a complaint.
Maya's terms could help by saying:
- outputs are draft content and may contain errors
- users must review claims before publication
- users are responsible for advertising compliance and product claims
- the service must not be used to create misleading advertising
- for business users, liability is capped at fees paid in the previous 12 months, so a GBP 29 per month customer would have a contractual cap of GBP 348, if that cap is enforceable in the circumstances
That doesn't magically remove every risk. A regulator or customer could still challenge misleading advertising. But the terms make the risk allocation much clearer between Maya and her customer.
For business contracts, limitation clauses and business terms and conditions often cap liability by reference to fees paid over a set period, insurance cover or a fixed amount. For consumer contracts, caps and exclusions need extra care because unfair terms may not be binding.
What Else Should Your AI App Terms Of Service Include?
Your AI app terms should include the usual SaaS clauses, plus extra clauses that deal with outputs, prompts, acceptable use and AI-specific risk.
Here is a practical outline you can use as a starting point. A lawyer should tailor it before you use it, because the right wording depends on your product, users, data flows and risk level.
- Who The Terms Apply To: name your company and explain when users accept the terms, such as creating an account or using the service.
- The Service: describe what your AI tool does and, just as importantly, what it doesn't do.
- Accounts And Access: set rules for login details, authorised users, age limits and account security.
- Plans, Fees And Payment: explain subscriptions, renewals, trials, failed payments, VAT treatment and any usage limits.
- Consumer Subscription And Cancellation Rights: if you sell to consumers, deal with the 14 day cooling-off period and digital content consent rules where relevant. For B2C subscriptions, also cover the DMCCA 2024 and 2026 Regulations requirements around pre-contract information, reminders, renewals and easy cancellation.
- Prompts And User Content: say what users can upload, what they're responsible for, and what licence they grant you to provide the service.
- Outputs: explain rights to use outputs, review obligations, accuracy limits and any restrictions on commercial use.
- Acceptable Use: list prohibited uses and your rights to suspend, restrict or terminate access.
- Intellectual Property: protect your software, brand, models, datasets, documentation and platform content.
- Privacy And Data Protection: connect the terms to your privacy policy and, for business customers, any data processing agreement needed under the UK GDPR and Data Protection Act 2018.
- Disclaimers: explain that outputs may be inaccurate, incomplete, outdated or unsuitable without human review.
- Liability: set fair and legally workable limits on liability, exclusions and user indemnities.
- Suspension And Termination: say when you can suspend accounts, end subscriptions and delete or export data.
- Changes To The Service Or Terms: explain how you can update features, pricing or legal terms, and how you'll notify users.
- Governing Law: choose the law and courts that apply, usually England and Wales for a UK-based business.
For consumer-facing AI subscriptions, make sure the legal wording lines up with the customer journey. The information a consumer sees before they subscribe, the wording around free trials, the timing and content of reminder notices, and the cancellation process all need to work together. A cancellation right buried in terms won't help much if the product flow makes cancellation hard to find.
Don't forget the basics around company information. A UK limited company's website, business letters and order forms should show details such as its registered company name, registered number, registered office, place of registration and limited status. VAT-registered businesses also need to include their VAT registration number on VAT invoices. These requirements come from UK company, tax and trading disclosure rules, including the Companies Act 2006 framework.
If personal data is involved, your terms shouldn't try to do the job of a privacy policy. Under the UK GDPR, users have privacy rights, and some requests must be handled within one month. If a personal data breach creates a risk to people's rights and freedoms, the ICO may need to be notified within 72 hours. Your terms, privacy policy and any data processing agreement should work together.
Key Takeaways
- Generic website terms usually don't cover the AI-specific risks around prompts, outputs, misuse and reliance.
- Your terms should make users responsible for what they upload and clear about how they can use outputs.
- If you sell B2C AI subscriptions, prepare for the DMCCA 2024 and 2026 Regulations requirements on pre-contract information, reminders, renewals and cancellation.
- Acceptable use rules matter because misuse can include scams, infringing content, harmful outputs and unlawful decision-making.
- Be careful with IP promises. You can grant rights to use outputs, but don't overpromise that every output is unique or legally protected.
- Your terms should fit with your privacy policy, data processing terms, model provider terms and product controls.
If you'd like help with AI tool terms, subscription compliance, output and IP clauses, or data protection documents, you can reach us at 08081347754 or team@sprintlaw.co.uk for a free, no-obligations chat.
Get your customer-facing terms right
What should your privacy and online terms cover?
If you collect customer data, sell online or run marketing campaigns, your public terms and privacy documents should match the real customer journey.







