Read vs write actions

The Consultad AI assistant can both read your campaign data and take actions on your ad accounts. These two types of operations are handled differently to keep you in control at all times.

Read actions

Read actions query your data without making any changes. They run automatically when the assistant needs information to answer your question.

Examples of read actions:

• Querying your campaign performance metrics
• Fetching account structure (campaigns, ad sets, ads)
• Pulling spend and budget data across platforms
• Generating comparison tables and trend summaries

You do not need to approve read actions. They are executed as part of the assistant's normal response flow.

Write actions

Write actions modify something in your connected ad accounts. They never execute without your explicit approval.

Examples of write actions:

• Pausing or resuming a campaign
• Changing a campaign or ad set budget
• Creating a new campaign or ad set
• Updating targeting or bid settings

How confirmation works

When the assistant determines that a write action is needed, it follows a strict flow:

1. Proposal - the assistant describes exactly what it intends to do, including the target campaign or ad set, the platform, and the specific change.
2. Confirmation dialog - a confirmation card appears in the chat with the action details. You see Approve and Reject buttons.
3. Your decision - click Approve to execute the action, or Reject to cancel it. The assistant will never proceed without your click.
4. Result - after execution, the assistant confirms the outcome and reports any errors.

Write access per tenant

Write actions are gated per tenant. Not all tenants have write access enabled. If write actions are not available for your account, the assistant will let you know and can still help with read-only analysis and recommendations.

To request write access for your tenant, contact support.

What happens if something fails

If a write action fails after approval (for example, due to a platform API error), the assistant reports the error with details. No partial changes are applied silently - you always see the result.