Response times may vary depending on the size or complexity of the request. As a result, a pattern has emerged and is quickly standardizing to let a user control whether to stop or continue the output.

The most common control is the stop icon, which allows users to pause a request mid-fetch. We might presume that there is a business case here to reserve server time for requests that people want to make. From a UX perspective, this saves the user time by allowing them to stop and re-set their prompt if they are not getting the results they expected.

The fast forward button allows users to continue their request if it times out. The use most frequently seen here is where a model might return up to a certain amount of information, presumably to once again save server processing time. For example, if a user asks for 100 blog posts titles, the model might return 20, leaving it up to the user to click the next button in order to retrieve the final 80. There's a parallel here to the “Are you still watching?” prompt you might get after a few hours binging Netflix.

The play button is used to represent that a request is ready to run. It is interchangeable with a paper airplane or sparkles.

Details and variations

  • Use controls to let users manage the flow of information
  • The stop button is the most common control, which saves users time and companies processor costs when a result is not panning out mid-generation
  • For longer results, additional controls can allow the AI to pause and check in without disrupting the overall processing of the final result

Considerations

Positives

Gives users control over the flow of information
This may by the first moment where it “clicks” to them that they are directing the AI

Saves the user time and frustration
Avoids the equivalent to the “spinning beachball of death” effect caused by latency

Consider how you might combine other patterns to maximize user empowerment
Joining this with Tuner patterns can help users improve their prompt if they didn’t get what they were looking for the first time. For example, when a user clicks the stop sign, that could be a signal to show a nudge or suggestion on how to improve it. Parameters could also be introduced here.

Potential risks

Remain consistent
Users may get frustrated if they stop a prompt and start it again expecting to get a similar result. Variants can cause the second response to be quite different than the first, which is confusing.

Use when:
Users need the ability to stop the AI from fully running a prompt. This can be especially helpful when you anticipate more complicated results, such as data manipulation or large-scale generation.

Examples

⏹️ ChatGPT popularized the use of the stop symbol to let a user end a result before it is done
⏹️ Umso lets you stop a site generation mid-run
⏹️ Adobe Acrobat combines a loading visual with the clear action to "stop"
⏹️ Midjourney (Discord) allows you to cancel a job. This is especially useful in this type of environment, where each job takes X number of hours that the user pays for ahead of time
⏹️ Grammarly also combines the affordances of progress with the stop icon
⏹️ Jasper puts its controls front and center, along with a hot key
⏹️ Another stop icon by Julius
⏩ For longer results, Chat GPT will pause generation unless the user prompts it to continue
⏩ Hypotenuse prompts the user to continue generating their content after the initial run
⏩ Notion also gives the user the option to keep generating content. This is placed alongside options to remix.
No items found.