How Unfunnel uses AI

Where AI comes in

Unfunnel uses a large language model in four places, and only when you ask it to:

• First draft. You describe your offer and audience; the model drafts the landing page, the questions, the lead form and the result pages.
• Edits. You type an instruction such as "make the headline shorter"; the model returns a changed funnel document, which you can undo.
• Drafting from your website. When you ask for a funnel built from a website, we read that site for colours, fonts and wording, and the model decides what a funnel in that brand should look like and say. The website step in onboarding reads the site the same way but uses no model.
• Importing a page. You give the address of a page you already have; we read it in a browser, the model rebuilds it block for block as a funnel you can edit, and its pictures are copied into your workspace.

After a draft or an import, the model also looks at screenshots of the result, beside your website or the original page, to check its design and fix what it finds. Templates, the editor, publishing, lead capture, tracking and reporting do not use AI. Nothing an AI writes goes live until you publish it.

Which provider

We use Anthropic's Claude models over Anthropic's commercial API. We do not run our own models and we do not train any. The model we call is set on our side and may change as better ones become available.

What is sent, and what is kept

• Sent to Anthropic: your instruction, the funnel document being drafted or edited (its copy, questions, settings and theme), and, when drafting from a website or importing a page, the text, colours, fonts and pictures read from it, plus screenshots of the draft and of that website or page for the design check.
• Never sent: your leads and visitor answers, your billing details, your password, your tracking credentials, or anything from another workspace.
• Kept by us: the funnel you save, plus a usage record per request: the time, the mode, the model, the token counts and the cost. With it, so Unfunnel staff can check and improve the quality of AI drafts, we keep what the request was given (the instruction you typed, onboarding's goal and answers, the website or page it read, and the design check's result) and, for a new draft, the funnel as the AI first made it. These are never shown to other customers or used for training, and are deleted with the workspace (deleting your account deletes the workspaces only you belong to), or when you ask. The editor's Edit with AI also keeps each request you type and the AI's short answer with that funnel, so your workspace can see what was changed; they are deleted with the funnel. We do not store the raw response.
• Kept by Anthropic: only what its commercial API terms allow for abuse monitoring, for a short period, and never for training.
• Training: neither we nor Anthropic use your instructions, documents, funnels or leads to train models.

What it can get wrong

The model writes plausible marketing copy. Plausible is not the same as true, original or allowed. It can invent statistics, promise results, describe features you do not have, make health or financial claims that need substantiation, echo a competitor's phrasing, or draft for you something close to what it drafted for somebody else. It does not know the advertising rules of your platform or your country.

So: read every page before you publish it, check every claim, and treat the draft as a starting point written by a fast, well-read stranger. Under our terms, what you publish is your responsibility, whoever wrote the first version.

Who can use it

The Free plan includes AI to build your first funnel: drafting, edits and rewrites until it is ready to publish. Paid plans, from Launch up, draft and edit every funnel, with a monthly and a daily AI budget that grow with the plan, so a runaway loop cannot run up a bill. You will see a message if you reach one. A paid plan's free trial has the Free plan's AI until its first payment, and importing a page follows the same rules as drafting.

AI assistants that connect to Unfunnel

Separately from the drafting above, every plan lets you connect an assistant such as Claude, ChatGPT or Codex to a workspace. That assistant is run by its own provider, under its own terms, and what it reads from your workspace, including lead details, is handled by that provider. Our privacy policy covers what we log about those connections.

Contact

Questions about how AI is used go to hello@unfunnel.io.