Over the last year it seemed like every major site added some sort of "now better with AI" type call to action to its homepage. These tended to take the form of a big input box asking the user to describe something on their mind and let AI create it.
There are some flaws to this approach, however.
First of all, you can rarely get a great outcome out of a few words. This might work for some of the suggestions that often accompanied this CTA, like "tell me how rainbows work," but if you expect a website generator to produce something that perfectly matches your mental image, you're likely to be disappointed.
Second, and probably as a result of the first, this pattern started to converge into a gimmick. It became a means of showing what the product could do, but couldn't effectively create an ah ha moment for the user.
Behind the scenes, it's also expensive to run the processors that power generative AI features. A month ago, you could generally demo these products at least once after putting something into the giant CTA, whereas today you eventually hit a paywall.
Regardless of the driver, this pattern seems to be evolving.
Many marketing sites still include some sort of big CTA to introduce the AI capabilities (and the monthly fee). Inside products, companies are experimenting with more familiar patterns like labeling it a "pro" feature or introducing it as an option within a user journey.
This helps the user discover it within the context that it is most relevant. Using the Typeform example above, we can presume that by the time you are starting a wizard to create a new form, you likely have an intent in mind, which is going to make it more likely that what you describe gets you fairly close to what you are actually needing.
In some cases, the AI feature can't be discovered until other data has been added to the system. Products that have this dilemma include audio recording tools like Grain or Otter, and database tools like Productboard. In these cases, AI functionality is often introduced through a dialogue or banner, but again, discoverable in context.
Finally, let's not dismiss the possibility that sometimes a gimmick is exactly what you need. As someone who isn't an audio engineer, for example, I fail to see the commercial use case for AI-generated music. Still, I find their open CTA positively delightful, as it combines elements like suggestions and templates (a la Figjam) with a willingness to take the job seriously despite utterly ridiculous inputs.
I don't know if I'll ever need a motown-esque song about how much I hate work, or an indie-rock ballad about a cat who fell in love with the bird it was hunting. If I ever do need this tool, sites like Udio that make the initial use easy because of their clear CTA will be the first on my mental list.
