AI Disclosure

1. What the AI actually does#

KaiBuild builds websites with an AI agent. That is not a garnish on the product; it is the product. This page says plainly what the agent does, what leaves our systems to make it work, and where our responsibility ends and yours begins.

The agent:

• Reads your prompt and decides what kind of site you are asking for.
• Plans the pages and the sections on them, and searches our component library for components that fit.
• Writes the copy — headlines, body text, button labels, page titles and meta descriptions.
• Generates or selects images, where image generation is enabled.
• Edits what already exists when you ask it to change something, and reviews its own output against the brief before showing it to you.

Everything it produces is a draft. Nothing goes live until you publish it.

2. What data reaches which provider#

Two outside providers see content when the AI features run. Both are listed on the DPA & Subprocessors page.

• OpenAI — receives your prompts, your instructions to the agent, and the text of the site being built or edited, so that models can plan, write, and revise it. It also receives the text of our own component library, once, to build the search index the agent uses to choose components.
• Google (Gemini) — receives an image prompt when you ask for a generated image. Only the prompt. If image generation is not configured, nothing is sent and you get a placeholder instead.

What does not reach a model#

• Form submissions from your published sites. Nothing a visitor types into your contact form is ever sent to an AI provider.
• Your billing details. They are not in the agent's reach at all.
• Other customers' sites. The agent works on the site in front of it.

Training#

We do not use your prompts or your site content to train models, and we do not license them to anyone who would. What a provider does with data sent to its API is governed by that provider's own terms; we contract for API use, not for a training corpus.

What is kept#

Your conversation with the agent is stored alongside your site so you can see how it got there and pick up where you left off. It is deleted when the site is deleted. See the retention table in the Privacy Policy.

What not to type into it#

Do not paste passwords, API keys, card numbers, or anyone's sensitive personal data into a prompt. It will be sent to a model provider and stored in your transcript, and neither of those is where a secret belongs.

3. The output is AI-assisted and unverified#

Say it plainly: a machine wrote it and nobody checked it. The agent does not know your business. It knows what you told it and what patterns it learned. Everything it produces should be read as a confident first draft by someone who has never met you.

Specifically:

• Facts may be invented. Founding dates, staff numbers, awards, certifications, service areas, opening hours — if you did not supply it, treat it as fiction until you have checked it.
• Numbers may be invented. Prices, percentages, response times, guarantees.
• Text may resemble something that already exists. Generated copy, layouts and images can echo public material the model learned from. We do not warrant that output is original, or that publishing it infringes nobody's rights.
• Tone may miss. A model reaches for the average phrasing of your industry, which is often exactly the phrasing that makes a business sound like every other one.
• It is not advice. Nothing it writes about health, law, finance, tax, safety or medicine is professional advice, and it should not be published as though it were.

4. Testimonials, reviews and endorsements#

This has its own section because it is the mistake that turns a careless page into a legal problem.

A quotation attributed to a named person who did not say it is a fabricated endorsement. Not a placeholder, not a stylistic choice — a false statement about a real person, made in an advertisement. Advertising regulators effectively everywhere treat it as deceptive, and the person quoted has a complaint of their own.

The rule we ask you to follow:

• If the agent drafts a testimonial section, every quote in it is sample text. Replace it or delete the section.
• Publish a quote only if a real customer actually said it and agreed to your using it, in words you could show someone.
• The same goes for star ratings, review counts, "as featured in" logos, client logos, and case-study numbers. If it did not happen, it does not go on the page.
• Where you use an illustrative example on purpose, label it as one.

5. Your responsibility before you publish#

You publish. That single fact settles most of what follows: an AI wrote the sentence, and you published it.

• Review everything. Read the page as your customer would, and correct what is wrong before it goes live.
• Business claims are yours. Prices, offers, guarantees, delivery times, qualifications, regulated statements. So are the consumer and advertising laws that apply wherever your site is read.
• Other people's rights are yours to respect. Trademarks, copyrighted text and images, likenesses, and privacy.
• Your own privacy notice is yours to write. If your site collects anything from visitors, tell them what and why.
• Prohibited content is prohibited however it was written. Using the agent to generate it is a breach of our Acceptable Use Policy just as uploading it would be, and the agent will refuse where it recognises the request.

6. Who owns the output#

As between you and us, the content of your site is yours. We claim no ownership over the copy, structure or images the agent produced for you, and you do not need our permission to edit it, move it elsewhere, or keep using it after you leave.

Two honest limits on that:

• We cannot grant what nobody owns. In several countries, material generated by a machine without meaningful human authorship may attract no copyright at all. Nothing here changes that, and it means you may not be able to stop someone else using similar text.
• We do not warrant non-infringement. See section 3. If output resembles someone else's work, that is a risk you take on when you publish it.

The components, templates, design system and the Service itself remain ours, licensed to you for use in your site rather than transferred.

7. Credits#

AI work consumes plan credits. A build costs more than an edit; an image costs more than a line of text. A credit is consumed when the work is done, whether or not you liked the result — the model ran and the provider charged us either way.

Consumed credits are not refundable. See Cancellation & Refunds. If credits were consumed by a genuine fault on our side — a failed run that produced nothing, a duplicate charge for one action — write to [email protected] and we will put it right. That is not the same as being unhappy with the draft.

8. Availability and change#

AI features depend on outside providers. When a provider is slow, rate-limits us, or is down, the agent is slow or unavailable, and we cannot promise otherwise.

We change which models we use as better or cheaper ones appear. That may change how the output reads. Where a change alters which provider receives your content, this page and the subprocessor list are updated first.

9. Our own marketing#

Marketing material on KaiBuild may include illustrative metrics or composite examples. Where content is illustrative it is labelled as such, and it is not a guarantee of results. We hold ourselves to the rule in section 4: a named person is quoted here only if they said it and agreed to it.

10. Contact#

Questions about how the AI handles your content: Krushang Vachhiyat at [email protected].