Skip to main content
This walkthrough uses one page and three requests, from safe to more powerful. Use a test page, or a draft workspace, for your first try.

Step 1: Open the AI agent on your page

  1. Select the page in the page tree, for example in Content > Layout.
  2. Click Ask AI Agent in the top bar, or press Ctrl/Cmd+Shift+K.
  3. On first use, read the AI notice and confirm it.
  4. Check the context line below the title. It must show the page you want to work on.
  5. Choose an AI provider in the footer if needed (for example OpenAI).
If the context line shows the wrong page, close the agent, select the right page, and open it again.

Step 2: Start with a request that only reads

Type a request that does not change anything:
The agent runs the T3AA step Find images without alt text (marked completed) and answers with file paths and UIDs. Decorative images are left out. Nothing is written. This is the safest way to see how the agent works with T3AA. Agent answer listing images without alt text
The list can still include many fileadmin assets (for example PWA icons). Prefer a follow-up such as “Only list images in fileadmin/user_upload” or use storage UID 1 with a smaller limit when you work through MCP. See Troubleshooting.

Step 3: Preview a change, then cancel

You can also pass a file path instead of a UID, for example Mark this "/user_upload/hero.webp" image as decorative. Preview card for Mark an image as decorative with Cancel available
  1. The card is marked Preview — nothing changed yet.
  2. Check the Parameters: file UID or file path, language, decorative flag, and page.
  3. Read the hint at the bottom. In Live, it may say On confirm this goes live on the website.
  4. Click Cancel (or Discard) to write nothing. After cancel, the agent confirms that nothing was written.
After Cancel: nothing was written to the file You can click Apply instead when you are ready to save. For documentation and first practice, Cancel is the safe choice.

Step 4: Optional — ask about accessibility issues

This only reads the last scan result. It does not start a new scan. To run a new scan (and use scan quota), ask for it explicitly:
Those requests show a scan preview. Confirm with Apply only when you intend to scan. The screenshot below is the issues answer (read-only), not a scan preview. Accessibility issues response in the agent (read-only)

Next steps