Is AI-written content plagiarised or duplicate content?
Not by design: these models generate text statistically, predicting what comes next rather than retrieving and pasting from a source document. But resemblance to third-party material is possible, and our terms say so plainly rather than pretending otherwise. We do not warrant originality. People review what the agents produce, and you review and approve it before it is published.
- Varsuite does not warrant that AI output is original or that it does not resemble material produced by or for a third party. That exclusion is written into our terms, not buried in a footnote.
- Language models predict text statistically rather than copying from a source document, so output is not lifted from a specific page by design. Design intent is not a guarantee, which is why the warranty is excluded.
- Every output is reviewed and refined by people before it reaches you, and the work is written to your brief and to the business detail gathered in discovery.
- Under our terms you are responsible for reviewing and approving what we deliver before it is published or relied upon. Where similarity to existing material would be a concern (brand names, slogans, distinctive content) you should carry out your own checks.
- Plagiarism and duplicate content are two different problems, and neither is settled by an AI detector. We do not treat a detector result as a pass mark or cite one as part of our process.
How the text is actually produced
A language model does not hold a library of documents and paste from it. It generates text one piece at a time by predicting what follows, based on statistical patterns learned during training. There is no source document being copied out, which is why the output is not, by construction, an extract of someone else's page.
That is a statement about how the technology works, not a promise about every sentence it produces. The mechanism makes wholesale copying unlikely. It does not make resemblance impossible, and we would rather explain the distinction than let it do quiet work in our favour.
Where the risk actually sits, in our own words
Our terms are explicit: "We do not warrant that AI output is accurate, complete or fit for a particular purpose, or that it does not resemble material produced by or for a third party." They give no warranty of non-infringement or originality in respect of AI-assisted output, save for our obligation to perform the Services with reasonable care and skill.
We could soften that. We have chosen not to, because the alternative is a promise we cannot keep. A guarantee of originality, or of a clean result from a checker, is not a promise our terms would support, so we do not make it.
What we do about it
Our agents draft; people review and refine before anything reaches you. That review is the point of the model: agents accelerate the work, and the team decides what is good enough to ship. Our terms put it plainly: AI can make mistakes, and while we review output and build in checks, you remain responsible for reviewing and approving what we deliver before it is published or relied upon. The human pass is where we look for generic phrasing and for anything that reads as borrowed.
The other defence is upstream. Before we build, we come to you, sit down with the people who do the work, and, with your consent, record the discovery conversation. The copy is then written to your brief and to that specific business detail. This is the practical reason the output comes out particular to you rather than assembled from stock phrases: it is describing your business, not a category.
What you need to do
Review and approve before publication. Under our terms that is your responsibility, and it is not a formality: you are the only person who knows your market, your competitors' language, and the phrases your sector treats as someone's property. Before anything goes live we share a watermarked preview or a staging version for you to review, and your approval is your confirmation that the work is acceptable.
Where similarity to existing material would be a particular concern, for example brand names, slogans, or distinctive content, our terms say you should carry out your own checks before relying on or publishing it. The same applies to anything you supply us: you confirm you own, or have permission to use, the logos, images, text and data you give us. We do not offer an indemnity against third-party infringement claims, and we will not pretend otherwise.
Plagiarism, duplicate content, and the question underneath
These are two different things. Plagiarism is presenting someone else's work as your own: a rights and honesty problem. Duplicate content is an SEO concept about the same or near-identical text appearing at more than one address, which is usually a technical or syndication issue rather than a copying one. Neither is settled by an AI detector. Those verdicts are probabilistic, and we do not treat one as a pass mark or cite it as part of our process.
The real question most people are asking is quieter: will I get generic boilerplate that reads like every other site in my sector? That is a fair thing to worry about and it is the one we can actually address, through discovery, a specific brief and human editing. What we will not do is attach a guarantee to it. Our terms are clear that we do not warrant any particular ranking, revenue or other outcome.
Will the content pass a plagiarism checker or an AI detector?
We do not claim it will, and we do not use a pass mark from any checker or detector as part of our process. Those verdicts are probabilistic. Our terms exclude any warranty of originality, so promising a clean result from a tool would contradict the agreement you are signing.
If someone claims my content infringes their rights, where do I stand?
Our terms give no warranty of non-infringement or originality for AI-assisted output, save for our obligation to perform the Services with reasonable care and skill, and we do not offer an indemnity against infringement claims. You are responsible for reviewing and approving what we deliver before publication, and for your own checks where similarity to existing material would concern you. Rights in the deliverables are governed by your own agreement with us: see our Terms of Service.
Does a human actually read every word, or is that a figure of speech?
Agents draft, people review and refine before it reaches you, and then you review it again at preview or staging before anything goes live. That two-stage review is how the model is meant to work. What we will not claim is that review makes output error-free. Our terms say AI can make mistakes and that we do not warrant output is accurate or complete.
How do you stop it sounding like every other site in my industry?
By not asking the model to invent your business. We visit you, sit down with the people who do the work, and record the discovery conversation with your consent. The copy is written to that detail and to your brief, then edited by people. The more you tell us about how you operate, the more specific the result. There is no guarantee attached to this, just a method.
Still want to ask it in person?
Tell us what you are building and we will give you a straight answer, and a plan, within 72 hours.