---
title: "Effective prompting on Evonx"
description: "Write requests the AI builds right the first time - real examples of vague vs. precise prompts."
url: https://evonx.ai/docs/evolve/effective-prompting
locale: en
updated: 2026-09-04
---

# Effective prompting on Evonx

Write requests the AI builds right the first time - real examples of vague vs. precise prompts.

The quality of what Evonx builds tracks the quality of what you ask for. This page
is a short field guide: five habits, each with a weak and a strong version of the
same request.

## Name the user and the outcome

The AI designs around whoever you name. "Someone" produces generic screens; "my
front-desk staff" produces the right ones.

> **Weak:** Add a search feature.

> **Strong:** My front-desk staff need to find a customer's vehicle by plate number
> in under five seconds, from any screen.

## One change per request

Bundled requests build worse than sequenced ones — plans blur, and reviewing the
result gets harder. Threads are cheap; send three.

> **Weak:** Add photos to repair jobs, SMS reminders, and a monthly revenue report.

> **Strong:** When staff open a repair job, they can attach photos of the vehicle.
> Photos show in the job detail as a small gallery.

## Describe behavior, not implementation

Say what should happen; let the AI choose how. Prescribe technology only when you
actually require it.

> **Weak:** Add a WebSocket service with a Redis queue for job updates.

> **Strong:** When a job's status changes, everyone viewing the workshop overview
> should see it update without refreshing the page.

## State the edge cases you care about

You know your business's exceptions; the AI doesn't until you say them.

> **Weak:** Let customers book appointments online.

> **Strong:** Customers book a time slot online. Slots are 30 minutes, no
> double-booking per mechanic, and same-day bookings close after 16:00.

## Point at what you mean

When a change concerns something you can see, show it: attach a screenshot, or
[annotate the preview](/docs/preview-and-debug/annotate-preview) to mark the exact
element. "The third card on the dashboard" beats a paragraph of description.

> **Tip**
> Before a big request, switch to plan mode and ask the AI to plan first — you
> review the plan before any building starts. See
> [Models, plan mode & chat sessions](/docs/evolve/models-and-sessions).

## When the result misses anyway

Reply in the thread with what's wrong, concretely: "the gallery shows on the list
page, but I wanted it on the job detail". Follow-ups are part of the loop, not a
failure of it. If the app won't even start after a change,
[Analyse & Fix](/docs/preview-and-debug/analyse-and-fix) diagnoses it from logs.

## Related

- [Evolution Threads explained](/docs/evolve/evolution-threads) — where these
  requests go.
- [Build your first app](/docs/get-started/build-your-first-app) — a full first
  prompt, worked through.