What happens when a tool can do a million things and the user is only aware of a handful of them? 

Nudges use progressive disclosure to help users identify AI and use its capabilities in ways that that they didn't know existed. They also help people get back to important actions so they can easily access them the next time.

Think through the journey of your product or feature. What are the moments where AI is most likely to be helpful? Which of those moments are most difficult to discover?

Nudges are especially prominent in generative tools, like Writer.com, Jasper.ai, or Copy.ai, as well as knowledge management systems like Coda and Notion. They help users identify ways to use AI to create value on top of content that already exists (e.g. find action items), to take care of mundane tasks (generate a summary of your meeting notes), or to remix the content and improve it ("make it shorter").

Use this framework to identify the capabilities most useful to the user at any moment. If it's unlikely they will have discovered or mastered it yet, consider a nudge to show them how it works.

Different use cases

1. As in-app clues - The capabilities of these tools go well beyond what an average user might think to use them for. Open chat experiences may give someone full control of the possibilities, but the other side of this is users may feel anxiety to know where to begin. Nudges help users get a feel for how a tool works, and show them the basic features and capabilities to give them confidence to explore on their own

Sample nudges from Github Copilot, Adobe, Figma, and Grammarly

2. Where they serve as in-product marketing - Companies are constantly everboarding users into new features. This will be especially true given the fast-moving nature of this technology and the lack of clear product strategy. Nudges can help users identify new features in a way that gives them the most value - whether the feature itself is new or the user finally reached a moment in the journey where it makes sense.

Take MyMind's Serendipity feature, which is only available once you have saved 25 items. Not only does this make it more likely that a user will get benefit from the feature - it also helps the product developers ensure that the feedback and data they are responding to represents users who have reached a critical threshold of usage.

Hypotenuse and Writer both include a long list of user cases in their app, some of which require integrations or advanced features

3. Where they serve to sell you on the AI feature - In other cases, companies are taking a spaghetti development approach to AI–throw everything against the wall and see what sticks! In these cases, the prompted actions are not contextually relevant to the user, or the user is not in a place to use them to positive effect. Notion's suggested Summary action results in a blank purple box when plopped into an empty page, with no affordance about what to do next.

The blanket list of things Notion and Grammarly nudge the user with is exhaustive-and sometimes overwhelming

Details and variations

  • Nudges are shortcuts to actions that a user can take to interact with the AI
  • They tend to be generic at first but can become more personalized at the AI learns the user and the context
  • Few apps have keyboard shortcuts to common nudges/actions today, but that may change in time
  • Keep in mind the user's context in the product even if the AI hasn't learned their needs in the moment. Nudges in a blank state should be different from inline nudges



Contextual progressive disclosure
The best time to teach someone is when your assistance can help them achieve a goal they are actively striving towards. Affordances like nudges help people understand the technology by applying it to something they are already trying to do.

Develop power users
As AI makes it easier to perform difficult or labor-intensive tasks faster, these prompts can open up features of your product they never knew existed, demonstrating new value and helping people get more out of your experiences. I'm waiting for the action that will take data and develop a pivot table for me, and then show me how they did it. Progressive hand holding, combined with guidance through chat interfaces can help you continuously onboard your users well into their journey.

Potential risks

You don't need to show someone everything you can do on the first interaction. Make sure that the actions you are nudging don't lead to dead states. Consider setting guardrails to ensure people only unlock those features after they have hit critical mass or critical moments. These guardrails might be invisible, or examine how you might gamify them. Remember: AI is here to help, not just shine.

Use when:
The system can easily generate an AI response without requiring the user to write a response, such as an action to summarize information.


After submitting a search, Google informs you that you can use their AI tools to go deeper into the search results
Writer.com is full of small, contextual cues for how to use AI to complete basic tasks
Github CoPilot puts nudges inline to textboxes for easy discovery
Notional also provides a list of nudges after the user has put some text on the page. In fact, this appears to be table stakes for writing apps
Productboard offers actions to generate common documentation based on information already available in the system, significantly decreasing the time to synthesize these insights
Linear's AI-powered filters get a user to relevant results far faster (and predictably) than if they had tried to write the perfect prompt
Grammarly also makes it easy to perform basic tasks from your existing content, such as generating a reference
Grammarly offers nudges for how to use its tool to improve your writing
Copy provides the nudge "improve" which shows the user how to write better prompts
Midjourney puts action tools immediately under the image. Many of these relate to perameters that the user would have to know exist
No items found.