Getting a clean site capture: what the AI reads on your page

Draft: owner to write. This post is an outline. The headings and bullets show what it will cover; the full text comes before launch.

What the capture does

• Loads your page in a browser at desktop width (1440px) and phone width (390px).
• Takes full-page screenshots and crops sections such as the hero, features and pricing.
• Reads the page title, meta description, headings and visible copy.
• Looks for logo candidates in the header and in the favicon.

What the AI does with it

• Summarises the product name, a one-line value proposition, the audience and three to six benefits, using only the copy it found.
• Builds a brand kit: colors from your CSS and computed styles, fonts from the computed font family, matched to the nearest Google font when needed.
• You confirm or edit all of it before a script is written.

How to get a clean capture

• Paste a public page. Pages behind a login can't be captured.
• Use real text for headlines, not text inside images.
• Use one clear main heading and landmark elements (header, main, sections, footer).
• Keep a meta description that says what the product does.
• Put an SVG or high-resolution logo in the header.

What the capture skips

• Pages marked noindex.
• Private and internal network addresses.
• Anything past the page size and time limits.

When the capture gets it wrong

• Edit the brand kit: swap the logo, fix a color, choose a different font.
• Upload your own screenshots or screen recordings.
• No site yet? Describe the product in a few lines instead.

To add before publishing

• A worked example: one sample page and the brand kit it produced.
• Check every detail above against the Phase 2 build.