Can I legally publish a B2B ebook written by AI from our own content?

In most cases, yes. Publishing a B2B ebook that AI drafted from content your company owns is generally lawful, and most AI vendors' terms assign the output to you. The open questions are narrower: copyright may not protect the purely AI-written passages, vendors do not promise the output is non-infringing, and testimonials, product claims and customer data follow their own rules. A person who reviews, edits and approves the book reduces most of these risks. This is general information, not legal advice.
We make Nagent, whose long-form lead is FOLIO. Our section on where it fits is marked below. Sources were checked on 24 October 2026.
Is it legal to publish AI-generated content at all?
We know of no general rule in the major B2B markets that bars a business from publishing text an AI helped write. The rules that bite are the ones that already applied to marketing content, about what the text says and whose material it uses: misleading claims, fake reviews, personal data, defamation and copyright apply whether a person or a model wrote the sentence.
So the useful question is not "is AI content legal" but "is this particular ebook clean on each of those points". The rest of this post walks through them.
Who owns an ebook that AI wrote from our content?
Two separate questions hide here: who owns it between you and the vendor, and whether copyright law protects it at all.
Between you and the vendor. Read the terms of the tool you use. Two examples:
Jasper's terms say it claims no ownership over customer property, which includes output, though you grant Jasper a broad licence to use it, including to improve its services (Jasper terms, effective 30 May 2025).
Writesonic's terms say the customer owns outputs, that identical or similar outputs may be generated for other customers, and that AI-generated outputs may not qualify for intellectual property protection (Writesonic terms, updated 12 May 2026).
Under copyright law. In the US, the Copyright Office's January 2025 report on copyrightability concludes that AI-assisted works can be protected in whole or in part where a person determined the expressive elements, but that prompts alone are unlikely to be enough (US Copyright Office, Part 2). In its words: "Prompts alone, however, at this stage are unlikely to satisfy those requirements."
What that means for a B2B ebook:
Your source material (your docs, research, case studies) stays yours.
The human-written parts and your selection and arrangement can be protected.
Passages generated by AI with little human change may not be protectable, so a competitor copying them would be harder to stop.
Other countries take different approaches, so check where you publish.
For a lead magnet, thin copyright is rarely the business risk. Your edge is the knowledge and the distribution, not exclusive rights over sentences.
Can the AI output infringe someone else's copyright?
It can, mainly if it reproduces protected text or images closely. Vendors push this risk to you: Jasper disclaims warranties of non-infringement and says content is used at your own risk, and Writesonic does not warrant that outputs are non-infringing and asks customers to ensure outputs do not infringe when distributed (terms linked above).
Lower the risk by:
Drafting from your own documents rather than open web research.
Running long passages through a plagiarism checker before publishing.
Not asking for images "in the style of" a living artist or of a real brand.
Never uploading a third party's paid report as a source and republishing its content.
Jasper's blog makes a related point: output that substantially reproduces a protected work is where liability sits, more than training (Jasper).
Do we have to disclose that AI wrote the ebook?
For a typical B2B marketing ebook, usually no law requires a label today, but check two rules.
EU AI Act, Article 50. Deployers that publish AI-generated text meant to inform the public on matters of public interest must disclose it, unless the content has undergone human review or editorial control and a person or company holds editorial responsibility (Article 50 text). The same site notes Article 50 applies from 2 August 2026. A product guide is unlikely to be public-interest journalism, and human review is the exception anyway, but keep a record of who reviewed and approved.
Platform and store rules. Marketplaces and ad platforms set their own disclosure rules. Check them where you distribute.
Many teams add a short line such as "Drafted with AI from our own documents and reviewed by our team". It is honest and costs nothing.
Can we use customer reviews and testimonials in an AI ebook?
Yes, carefully. In the US, the FTC's rule on consumer reviews and testimonials prohibits creating reviews or testimonials that misrepresent that the reviewer exists or had experience with the product, buying reviews with a required sentiment, and undisclosed reviews by officers or managers (16 CFR Part 465). For an AI ebook that means:
Quote real reviews word for word. Never let the model "tidy" or invent a quote.
Keep the reviewer's consent and the source for every named quote.
Do not present a testimonial from an employee or manager as a customer's.
Do not imply the reviews you show are all of them if you picked only the positive ones.
What else can get a B2B ebook into legal trouble?
Unsupported claims. "Cuts costs by 40%" needs evidence. AI tends to add confident numbers, so check every figure against a source.
Statements about competitors. Cite their own pages, dated, and keep the tone factual.
Customer names and data. Use them only with permission, and do not upload personal data to an AI tool your data processing terms do not cover.
Confidential material. Strip internal pricing, roadmaps and partner terms from sources before drafting.
Trademarks. Use other companies' names to identify them, not to imply endorsement.
How much human involvement is enough?
There is no word count that settles it, but the direction is consistent: the more a person decides, the stronger your position on copyright, on the EU editorial exception and on accuracy. In practice, aim for these:
A person writes or approves the outline, which sets the selection and arrangement.
A person edits each chapter, not only the first.
An expert adds their own examples and opinions, which the model could not have produced.
A named person approves the final file and the date is recorded.
Keep the drafts, the edits and the approval record together. If a question comes up a year later, that trail shows who decided what.
What is a practical legal checklist before you publish?
Confirm you own or have permission for every source document.
Read your AI vendor's terms on output ownership, training use and indemnity.
Check every number and claim against a named source.
Verify every quote and testimonial against the original and keep consent.
Run a plagiarism check on long passages.
Review competitor statements against their own dated pages.
Record who edited and who approved the final file.
Add a short AI-assistance note if your audience or market expects one.
For regulated sectors (health, finance), send the final PDF to legal.
Where does Nagent fit?
Nagent is multiplayer AI for growth: people and AI agents in one shared context, delivering the number. FOLIO is the long-form lead on the Content team, which CREA, the AI Content Chief of Staff, leads. Several of its design choices map to the checklist above:
Your sources. A book is written from the Knowledge Hub documents a person picks, each chapter against the passages ranked for it.
Cited comparisons. In the Comparison book type, the writer is told to cite every factual claim to a numbered source, and a subject backed by your own document is never researched on the web. Other book types do not carry citations.
Rules first. The brand guideline's banned claims and mandatory items are placed ahead of tone.
A person approves. Publishing is never part of the chain that writes the book. A person presses Promote, and changes to what an agent may do alone are written to the audit log.
Security. Nagent is ISO/IEC 27001:2022 certified; SOC 2 Type II is in progress.
What it does not do: give legal advice, check quotes against their original source, or run a plagiarism check. Those steps stay with your team.
Frequently Asked Questions
Can I copyright an ebook that AI helped write?
In the US, the parts a person wrote, selected or arranged can be protected; purely AI-generated passages may not be. The Copyright Office's 2025 report says prompts alone are unlikely to be enough.
Can you publish an AI-written ebook on Amazon Kindle?
Stores set their own rules on AI-generated content and disclosure. Read the current publishing terms of the store before you upload.
Do AI vendors indemnify us if the output infringes?
Check your contract. The standard terms we read for Jasper and Writesonic disclaim non-infringement, and Writesonic's US copyright indemnity excludes claims arising from outputs. Enterprise contracts may differ.
Is it safe to upload customer case studies to an AI tool?
Only if you have permission to use them and the vendor's data terms cover that data. Remove personal details you do not need.
To see how FOLIO keeps a person in charge of every published book, book a demo.