Usually you own whatever rights exist, but there may be fewer rights than you think. OpenAI, Anthropic, Google, Microsoft and Canva all say they don't claim your outputs, and OpenAI and Anthropic assign you their rights "if any". Copyright generally needs a human author, so purely AI-made text or images may be protected thinly or not at all.
The practical gap is this: owning an output doesn't stop a competitor using something similar, because vendors warn that outputs aren't unique, and it doesn't protect you if the output echoes someone else's work. The edits you make are what turn a draft into something that is clearly yours, so the more a piece matters to your brand, the more of it should come from you.
What the terms actually hand you, vendor by vendor
Ownership language is surprisingly consistent across the big vendors. The differences sit in two places: whether you get an indemnity (a promise to defend you if someone sues over the output) and what voids it. This summary comes from each vendor's current terms, checked in September 2026.
| Vendor and plan | What the terms say you get | Indemnity, and the catch |
|---|---|---|
| OpenAI, personal plans | You own the output; OpenAI assigns its rights "if any"; output may not be unique | None for personal plans |
| OpenAI, business and API | The customer "owns all Output", again with an "if any" assignment | Output indemnity is listed for API and Enterprise, Edu and Healthcare customers; it excludes modified output and trade mark claims from commercial use |
| Anthropic, Free, Pro and Max | Anthropic assigns its rights "if any", subject to following its terms | None |
| Anthropic, commercial (API, Team, Enterprise) | The customer owns its outputs, with an "if any" assignment | Defends paid, authorised use; excludes modified outputs, combinations with other technology and trade mark claims |
| Google (Gemini) | Google won't claim ownership of content you generate; it may generate similar content for others | Check your Workspace or Cloud agreement |
| Microsoft Copilot (business) | Microsoft doesn't claim ownership of output | Its Customer Copyright Commitment defends commercial customers who used the built-in guardrails and content filters |
| Canva | You own your output; Canva makes no copyright ownership claims | Check Canva's AI terms for your plan |
| Midjourney | You own the images and videos you create | Businesses grossing more than $1,000,000 a year need a Pro or Mega plan for commercial use |
| Adobe Firefly | Adobe's own models are trained on licensed content | IP indemnity only for enterprise entitlements; partner models and beta features excluded |
Two details deserve a second look. First, "if any" is doing real work: the vendor hands you whatever rights it has, while making no promise that there are any. Second, OpenAI's service terms list the output indemnity for API and Enterprise, Edu and Healthcare customers, and ChatGPT Business isn't in that list. A small firm on Business or a personal plan should assume it carries the risk itself.
The Midjourney rule catches growing firms. An illustrative removals company turning over about $1.4 million a year, using Midjourney's entry plan for social ads, is outside the terms for commercial use. The fix is a plan change, not a lawyer, but it's the kind of detail nobody checks until a dispute starts.
Why "you own it" can still mean thin protection
Copyright protects original work by human authors in most legal systems. Registries and courts in several major jurisdictions have refused protection for material generated entirely by AI, while accepting the human contributions around it: the text you wrote, the edits you made, and your selection and arrangement of the whole. A few legal systems have a specific rule for "computer-generated works" that treats the person who made the arrangements as the author, sometimes with a shorter term of protection. The detail varies, and it is still moving.
For a small business, that translates into three working assumptions:
- Unedited AI output is weak property. You can publish it, but you may not be able to stop someone copying it.
- Your edits are the protectable part. Specific knowledge, real examples and your own structure add value for readers and make the work more clearly yours.
- Ownership doesn't equal clearance. If an output closely resembles a photographer's image, a brand's mark or another firm's copy, owning your version doesn't make using it safe.
When the stakes are high, such as a logo, a product whose value is the content itself, or a client contract that promises originality, that is the point to ask an IP lawyer, because the answer depends on where you trade and what exactly was made.
A roofing contractor's new website, piece by piece
One illustrative case shows how this plays out. A roofing contractor rebuilt its website over a month using AI for almost everything: 14 pages of copy drafted in ChatGPT, 30 images from an AI image tool, a logo concept, and a draft of its workmanship guarantee terms. Before launch, the owner went through each piece with one question: what do we actually own here, and what could go wrong?
| Piece | What was found | What they did |
|---|---|---|
| 14 pages of copy | 5 pages rewritten with real job details; 9 only lightly edited | Six hours adding specifics to the 9 thin pages; drafts and edit history kept |
| 30 AI images | 22 were presented as "our recent work" | Replaced those 22 with real job photos taken over four weeks; kept 8 diagrams, captioned as illustrations |
| Logo concept | A reverse image search found a close match used by another trade business | Dropped it; a designer drew a new mark using the AI concepts only as a brief; trade mark search before ordering van graphics |
| Guarantee terms | Ownership wasn't the issue; accuracy and legal effect were | Sent to a solicitor for review |
The images were the biggest risk, and it wasn't an ownership problem. Presenting AI pictures of immaculate roofs as the firm's own jobs misleads customers, whoever owns the pixels. The logo was the second, because it was about to go on two vans, 40 hi-vis jackets and site boards, and a reprint after a complaint would have cost far more than the designer's few hours. The copy was the cheapest fix and the one that paid back twice: pages with real detail were both more clearly the firm's own work and more persuasive to customers. The wider question of images in marketing is covered in whether you can legally use AI-generated images in your marketing.
The edit that turns an AI draft into your work
The edit that matters isn't swapping a few adjectives. It's replacing general claims with knowledge only your business has. Here is a before and after from the roofing example (illustrative):
AI draft:
We provide high-quality flat roofing solutions tailored to
your needs, using durable materials and expert craftsmanship
to ensure long-lasting protection for your property.
Edited:
Most flat roofs we replace are 15 to 20-year-old felt roofs
that have started to hold water near the outlets. We strip
back to the deck, replace any rotten boards (usually two or
three sheets on a single garage), and lay a single-ply
membrane. A typical single-garage roof takes one day.The first version could appear on a thousand websites, and probably does. The second contains the firm's own observations, process and numbers. It is more useful to a homeowner comparing quotes, and it's the kind of human contribution copyright rules look for. A quick test for any AI-drafted page: highlight every sentence a competitor could publish unchanged. If more than half the page is highlighted, it needs more of you in it.
You can get the AI to help with that test. A prompt like this works well on any draft page:
Here is a draft page for our roofing website. Mark every
sentence a competitor could publish unchanged. For each one,
ask me one question whose answer would make it specific to
our business. Don't rewrite anything yet.
[paste draft]A typical reply (illustrative) flags the generic lines and asks things like: "How old are the roofs you usually replace, and what failure do you see most?", "How long does a typical job take, and what affects that?" and "What guarantee do you actually offer, and who backs it?" Answer those in your own words and you have the raw material for a rewrite. What to watch: the AI sometimes suggests answers of its own in brackets. Delete them, because a plausible invented detail is worse than a generic sentence.
Keep the evidence while you work. Save the prompt, the first draft and the edited version in the same folder, with dates. If ownership is ever disputed, a record showing what the AI produced and what your team changed is far more persuasive than a recollection.
Where the indemnity stops: edits, logos and trade marks
There's an awkward tension in the terms. Editing AI output makes it more clearly yours, yet both OpenAI's and Anthropic's indemnities exclude claims arising from outputs you have modified. In practice the indemnity is aimed at a narrow case, where the model reproduced someone's protected work and you published it unchanged, and it only applies on the plans that carry it. For most small firms, the protection that matters is your own checking, not the vendor's promise.
Both vendors also exclude trade mark claims arising from using output in trade, which is exactly the logo situation. That's why logos deserve their own process, set out in whether you should design your logo with AI. The cleaning-company version of the mistake is common enough to describe (illustrative): a firm generates a house-and-sparkle mark, loves it, wraps three vans and orders uniforms, and only then runs a reverse image search that turns up a franchise using a near-identical symbol. Nobody sued, but the re-wrap and new uniforms cost more than a designer would have charged at the start.
Habits that reduce the risk, whichever plan you're on:
- Don't prompt with living artists' names, other brands' names or "in the style of" a specific company.
- Run a reverse image search on any image that will carry your brand or appear in paid advertising.
- Treat AI logo output as a sketch for a designer, followed by a trade mark search, never as the finished mark.
- Use the vendor's safety and citation features rather than switching them off; the OpenAI and Microsoft protections both depend on it.
- Check your plan: if an indemnity matters to you, it lives in enterprise and API terms, not in $20 subscriptions.
What you feed in shapes what you can claim
The terms also say something about inputs: you keep the rights you already had in what you upload, and you promise you have the right to upload it. OpenAI's business agreement makes the customer responsible for having "all rights, licenses, and permissions" for its input, and its indemnity excludes claims arising from customer content. Three everyday cases show why that matters:
- Your own material in, your own material out. A locksmith uploads a photo of its van and asks an AI tool to clean up the background. The original photo is the firm's, and the edited version builds on it. This is the safest kind of AI content there is.
- Someone else's material in. Pasting a competitor's service page and asking for "a version for us" produces a derivative of their work, and the vendor's indemnity won't cover it because the problem came from your input. Write from your own notes instead.
- Customers' material in. Reviews belong to the people who wrote them, and customer photos belong to the customer. Ask permission before you quote a review or feed a customer's photo into a before-and-after graphic.
When staff or freelancers make the content
Most AI content in a small business isn't written by the owner. Work an employee produces as part of their job usually belongs to the employer under the employment contract, and that doesn't change because AI helped. Freelancers are different: unless your contract assigns their rights to you, a freelancer may keep them, and "AI-assisted" adds a second question about which tools and plans they used.
A short addition to any freelance brief covers it: "You assign to us all rights you hold in the deliverables, including AI-assisted work. Tell us which AI tools you used, confirm your plan allows commercial use, and keep your drafts for 12 months." An illustrative cleaning company commissioning a set of staff training leaflets from a freelance designer added exactly that, and it cost nothing. Without it, the firm would have owned printed leaflets but possibly not the files needed to update them next year.
Client contracts that promise originality
If you deliver written or designed material to clients, their contract may ask you to warrant that everything is original and that you own all the rights, then assign them. With AI in the workflow, you may not be able to give that promise in full. An illustrative HVAC installer that writes operation and maintenance manuals for commercial clients hit this when a facilities company's supplier terms arrived with a blanket originality warranty.
The installer proposed wording along these lines (illustrative, to be checked by a solicitor):
Use of AI tools. We may use AI tools to help draft documents.
Our staff review, edit and take responsibility for all
deliverables. We assign to you all rights we hold in the
deliverables. We don't warrant that AI-assisted material
qualifies for copyright protection, but we warrant that we
have not knowingly copied third-party material, and we will
correct any deliverable found to reproduce it.The client accepted it with one change: a duty to disclose which tools were used on request. That exchange took two emails and removed a promise the installer couldn't honestly keep. If you're unsure whether to tell clients about AI at all, the reasoning is set out in how to write an AI content policy for your marketing, which also covers who approves what before it's published.
A publishing check for anything AI helped make
A short check before publishing catches most problems. Here it is filled in for an illustrative landscaper's spring brochure:
AI PUBLISHING CHECK: Spring brochure Date: [date]
Tools and plans used: ChatGPT Business (text),
AI image tool, Pro plan (images)
Plan allows commercial use? Yes (checked terms page, [date])
Text edited by a person? Yes: prices, planting lists,
aftercare steps rewritten
Drafts and prompts saved? Yes: /Marketing/2027-spring/AI
Real photos of our work? Yes: all "our gardens" images
AI images labelled? Yes: 3 illustrations captioned
Names or brands in prompts? None
Reverse image search: Cover image: no close matches
Logo or brand marks by AI? No
Facts checked by: [first name], prices and dates
Approved by: OwnerTen minutes per piece is a fair price for publishing something you can stand behind. Labelling rules for social platforms are covered separately in how to label AI-generated images and video on social media.
The short version for a busy owner: the vendors aren't going to claim your content, but the law may not give you much to claim either unless you add real human work. Edit for substance, keep your drafts, keep AI away from your brand marks without a designer and a search, and save the lawyer's time for the pieces where ownership genuinely matters.
Ownership questions owners ask next
Can I register an AI-generated logo as a trade mark?
Trade mark protection comes from using or registering a mark for your goods and services, not from copyright, so an AI-assisted logo isn't automatically ruled out. The risks are that it resembles an existing mark and that its copyright is thin. Have a designer refine it, run a trade mark search, and ask an IP lawyer before printing it on vans and signage.
Do I have to label AI-generated content?
It depends on what it is and where you sell. If you sell to customers in the EU, the AI Act's transparency duties, which have applied since 2 August 2026, cover deepfakes and some AI-generated content. Platforms add their own rules for AI images. Whatever the law says, don't present AI images as photos of your real work.
Can a competitor copy my AI-written web pages?
They may be able to copy the parts no human wrote, because copyright generally needs a human author. The sections you rewrote with your own knowledge, and your selection and arrangement of the whole, stand a better chance of protection. Other remedies, such as unfair competition claims, vary by legal system, so ask an IP lawyer if copying is costing you work.
Further reads
- Should Freelancers Tell Clients They Use AI? — Disclosure to clients when AI helps with paid work.
- How to Check AI-Generated Images for Errors Before Posting — Catch warped hands, fake text and wrong details before you post.
- Brand Guidelines for AI Images: Colours, Style and Limits — Keep AI images consistent with the brand you're protecting.
- AI for Freelance Designers: Faster Concepts Without Rights Issues — How designers use AI concepts without handing over rights problems.
- Canva AI for Small Business: Which Features Save Real Time? — Canva's AI features, the tool many small firms publish from.
- ChatGPT Prompts for Small Business Owners: 50 Tested Examples — Prompts whose drafts you can then make genuinely your own.
- What to Check in an AI Tool's Privacy Policy and Terms — Seven checks for any AI tool's privacy policy and terms, the words to search for, and how ChatGPT, Claude, Gemini and Copilot compare.
- How to Mention AI Use in Client Contracts and Proposals — Sample proposal wording and contract clauses for disclosing AI use to clients, matched to how much AI touches the work, with an architect-practice example.
- AI Model Shots for Boutiques: Try-On Images Without a Photoshoot — How AI on-model images work, what tools like FASHN and Photoroom cost, which garments come out wrong, and how to check and label every image honestly.
- How Coaches Use AI to Build Workbooks, Exercises and Programmes — Turn your coaching method into a programme and workbook with AI doing first drafts, while the method, the edits and the claims stay firmly yours.
- AI Room Visualisation From Client Photos: A Designer's Workflow — Turn a client's room photo into an honest concept visual: shooting rules, a lock-list prompt, drift checks and the caption every image needs.
- How to Choose a Marketing Agency That Uses AI Responsibly — Seven questions that separate careful agencies from careless ones, a test brief for the pitch, and how a wedding planner compared three agencies.
- Who Owns the AI Workflows a Consultant Builds for You? — Legal ownership and practical control of AI automations: assignment vs licence, whose accounts they run in, platform transfer rules and sample clauses.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI Terms of Use (effective 1 Jan 2026), OpenAI Services Agreement (effective 1 Jan 2026) and Service Terms; Anthropic Consumer Terms and Commercial Terms of Service; Google Terms of Service and Gemini API Additional Terms; Microsoft Learn (Data, Privacy, and Security for Microsoft Copilot); Canva AI Product Terms; Midjourney documentation on commercial use; Adobe Firefly product description and legal FAQs. Checked September 2026. Not legal advice.