How should a product ask for user confirmation before AI automatically takes action?

First, distinguish suggestions from execution

AI can suggest changes, and it may also be designed to send messages, modify files, or submit requests. Users need to know whether the system is displaying a draft or preparing to carry out an action. If the distinction is unclear, customers may change data or notify others without understanding the impact.

Inspired by public discussions at CES about trust and transparency in intelligent content, this article offers independent interaction design recommendations. Whether specific authorization or other compliance measures are required should be reviewed by relevant professionals based on the actual context.

Determine confirmation points based on impact

Review the actions the system may perform, their targets, and their consequences, paying particular attention to sending information to external recipients, deleting or overwriting materials, making payments, and changing permissions. Confirmation design should reflect the actual impact and reversibility of each action, rather than using the same dialog for every operation.

For low-impact, reversible actions, provide clear status information and a way to undo them. For significant actions or those that are difficult to reverse, show key details and require explicit confirmation before execution. The specific classification should be determined by the product and business teams and the relevant professional leads.

Avoid frequent, meaningless confirmations that train users to click automatically. Each confirmation should help users understand what is different about this particular action.

Show the information users actually need to review

On the confirmation page, show the action type, its targets, the content to be sent or modified, and the main known effects. For example, before sending an email, show the recipients, subject, and body, and allow users to edit them. Before making bulk changes, explain which files are involved and the scope of the changes.

Distinguish between conditions customers have explicitly selected and details the system has inferred automatically. Require further confirmation for unverified targets, amounts, or permissions instead of assuming they are correct. When information is lengthy, provide an accessible way to view it in full. Key information must not be hidden solely in collapsed sections.

Use specific action labels for buttons, such as “Confirm and send” or “Go back and edit,” and ensure that canceling does not allow the action to proceed silently. Avoid vague or misleading options.

Report the actual results after execution

After confirmation, explain whether the request has been received, whether it is being processed, and its final completion status. If some actions fail, clearly state which were completed and which were not, and provide working options to retry or contact support.

If an action can be undone, explain the scope and conditions. If it cannot be undone, do not use “Canceled” to imply that an action that has already occurred has been reversed. Avoid displaying success without evidence when the network connection is interrupted, and handle duplicate execution caused by repeated clicks.

Record the necessary action and confirmation statuses so customers can review them or support staff can investigate issues. Handle data retention, access, and retention periods according to the company’s actual policies and applicable requirements.

Test the design against risk scenarios

Ask users to try reviewing, editing, canceling, and confirming actions, and to explain what they think will happen. Test scenarios involving an incorrect target, insufficient permissions, network interruptions, and partial execution failures, and observe whether recovery is possible.

Check that backend behavior matches the interface, and have the relevant leads verify the boundaries of business operations. Qingmei Future can help design interaction flows. The company is responsible for actual permissions, technical implementation, data processing, and service commitments.

For example, before a bulk send, ask participants to check the recipients and message body, then try canceling. Verify that no messages are sent after cancellation. If only some messages are sent, the results should state the actual status of each item rather than displaying only a generic failure message.

After testing, document user misunderstandings, the conditions that triggered them, and who is responsible for fixes. When revising interface copy, also check that the backend follows the same confirmation rules to prevent situations where the interface says an action has been canceled but the action still executes.

If the action’s targets or permissions change, present the information requiring confirmation again. Do not reuse a decision the user made about earlier content.

Discussion Background and Scope

CES · AI or Human? Trust, Transparency, and Co-Creation in the Age of Intelligent Content · 2026-01-07. Official page date: 2026-01-07. The source discusses authenticity, transparency and accountability in intelligent content. It is background reading, not evidence for each practical recommendation in this article. The recommendations are Qingmei Future’s independent analysis.

Related Services and Content

Explore digital interface design　Explore industrial design services　Discuss product experience design