Landing pages, proposal pages and ROI calculators built by an AI agent

Template page builders are fine for a generic page. They struggle when you want something specific: a page for one prospect with their numbers on it, a calculator that converts currencies, a form that creates a CRM contact with the right source and consent. A coding agent builds those as real pages, so the limit is your brief, not the template.

Sales pages are one of the two best starting points we suggest in the guide to building software with an AI coding agent: self-contained, quick to review and directly tied to revenue. Here's how to brief them, capture leads properly, read the numbers honestly, and approve them before they go live.

Three kinds of page worth building

A published Cortex customer combined the last two:

Individual results vary. The proposal itself came from research turned into a decision brief, a process we describe in from market research to client proposal.

A landing page brief that fits on one screen

Use the same five-part thinking as any change request; see how to brief an AI coding agent. The 'proof' line is where pages get into trouble. Tell the agent explicitly: no invented statistics, no made-up testimonials, no 'trusted by thousands'. If you don't have proof yet, a clear explanation of what you do and a direct way to talk to you is better than borrowed credibility.

Briefing a useful ROI calculator

A calculator persuades because visitors use their own numbers. It also backfires if the defaults are fantasy. Brief it carefully:

1. Inputs: three to five things the visitor knows, such as number of rooms, average booking value, hours spent on a task.
2. Formula: written in plain words in the brief and shown on the page. 'Hours saved per week × hourly cost × 48 weeks'.
3. Assumptions: every number you supply, visible and editable, with its basis.
4. Outputs: a range or a clearly labelled estimate, not a single promise.
5. Currency and tax: AUD by default; say whether GST is included; add a second currency only if your audience needs it, as in the AUD-and-yen example.
6. Action: what happens next. 'Email me these results' is a good one, because it's useful to the visitor and creates a lead.

Before release, check two or three results by hand with a calculator. If you can't explain a number on a call, it shouldn't be on the page.

Capturing leads into the CRM properly

A form that emails you a notification isn't lead capture. A form that creates a contact in the CRM, with the page as its source, a consent record and a next action, is. In Cortex, form submissions can land in Fulcrum CRM, where the contact gets a next action and the agent drafts the follow-up for your approval, as described in the CRM follow-up process.

Two Australian rules shape the form. Under the Spam Act, you need consent before sending marketing messages; the ACMA lists ticking a box on a website as one way to give express consent, and says to keep a record of who consented, when and how (ACMA: Avoid sending spam). And if the Privacy Act 1988 covers your business, APP 5 deals with telling people certain matters when you collect their personal information (OAIC: APP quick reference). In practice:

• Ask only for the fields you'll use. Name and email is often enough.
• Use an unticked box for marketing consent, separate from the main action.
• Link to your privacy policy next to the form.
• Store the consent choice, date and page on the contact record.

This is general information, not legal advice.

Using analytics without overstating what they mean

Page analytics tell you what happened. They rarely tell you why, and small numbers mislead easily. A few rules for reading them honestly:

• Count outcomes, not visits. Enquiries and booked calls matter; page views mostly don't.
• Respect small samples. If 3 of 40 visitors enquired one week and 6 of 45 the next, you don't have a trend. You have two weeks.
• Change one thing at a time. New headline and new offer in the same week means you can't tell which mattered.
• Check the source. A spike from one email send isn't the page getting better.
• Don't publish what you can't support. 'Conversion up 40%' on a page with 60 visitors isn't a claim you should put in a sales deck.

Reviewing campaign results

At the end of a campaign, ask the agent for a one-page result: visitors by source, form submissions, how many became real conversations, how many became deals, and what it cost. Ask it to mark anything below a sensible sample size as 'too early to tell'. The deal numbers come from the CRM, which is why the form-to-CRM link matters so much.

Approval before anything goes live

Public pages are outward-facing by definition, so they go live only on your OK. Run this check on the preview, on a phone:

1. Every claim on the page is true and you could defend it.
2. Any customer quote is used with permission and quoted exactly.
3. Prices are in AUD and GST treatment is clear.
4. The form works, creates the CRM contact and stores consent.
5. The privacy policy is linked.
6. The calculator's sample results check out by hand.
7. It loads quickly on mobile data and reads well on a small screen.
8. Proposal pages for a single client aren't indexed by search engines or linked publicly unless you intend them to be.

For changes to a live page, especially calculators and forms, follow the release checklist in reviewing AI-built software changes.

When a template builder beats an AI landing page builder

If you want a standard page, edit it yourself every few days, and don't need CRM capture or calculations, a template page builder may be simpler. The agent approach earns its keep when pages are specific, interactive or connected to your other systems, and when you'd rather describe a change from your phone than drag blocks around.

Cortex Small and Growing plans include website hosting; see pricing. To see how page requests are built, deployed and reported back from chat, look at Fulcrum Code.