Skip to content

How the AI builder works

The chat panel is where you build. You type what you want; Genesis figures out what kind of job that is, does it, and updates the live preview. Understanding how it interprets your message is the single biggest thing you can do to get great results consistently.

For every message you send, Genesis picks one approach to handle it. For example:

  • A quick visual tweak (“make this heading bigger”) → a fast, surgical edit.
  • A bigger request (“add a pricing section”) → the full builder generates new content.
  • A question (“how do I publish?”) → it answers without changing your site.
  • A specialized feature (“a product catalog”, “a backend with login”) → a dedicated tool.

It decides this from signals in your message: whether you’ve selected an element on the page, whether you attached an image or pasted a URL, the words and verbs you used, and whether your message ends with a question mark.

Genesis is smart, but it can’t read your mind — and it picks one approach per message. Most of the time it guesses right. But for specialized, structured features, a vague message can land in the generic “just edit something” path instead of the dedicated tool that does the job properly.

You don’t have to learn a rigid template. You have two reliable levers:

  1. Be clear and specific — see Writing good prompts.
  2. Use a tag to lock the approach — when you know exactly which tool you want, type a bracket tag and Genesis routes straight to it, no guessing. This is the most important skill for advanced features like the Dedicated Cloud backend, product catalogs, and analytics. See Steering the AI with tags.

Remember you also have the visual editor alongside chat:

  • Chat — structural and content changes you describe in words.
  • Visual editor — click an element to change its text, size, color, or image directly.

Both write to the same site, so use whichever is faster for the change at hand.

When a request is simple, just say it plainly. When you want a specific tool — a backend, a product catalog, analytics, editable controls — tell Genesis explicitly with a tag instead of hoping it infers the right one.

Next: Steering the AI with tags →