Jasper’s VP of Design Stephanie Mencarelli has built global consumer products and experiences for Spotify, Walmart, and Fandango, to name a few. “I've taken the ethos of designing for millions of people with various needs and segwayed into building a very opinionated, persona-first product for Jasper,” she said.
An opinionated product is one that strongly reflects specific design choices where there may be 100 different ways to complete a task, but a designer chose one. And that one choice is to cater to a specific audience rather than an infinite choice in the way a user can act. For Jasper, that specific audience is marketers, so every part of our product is built with them in mind.
However, the term “marketers” can be very vague. It can mean professionals with vastly different needs, workflows, areas of expertise, and leadership responsibilities — not unlike the innumerable types of listeners who listen to Spotify every day. Many of those marketing functions can also differ dramatically from business to business and industry to industry. This nuance raises the stakes for Jasper to be opinionated in all the right ways.
Mencarelli and her team have the tough job of appealing to all those marketers. Doing so requires finding a common thread: “It needs to be simple and easy to use, yet powerful,” she said.
Designing a simple yet powerful AI platform for marketers requires a number of aspects. I spoke to Mencarelli about what those aspects are and her efforts to make Jasper as opinionated, aka useful to marketers, as possible.
She discussed how her team always keeps a specific proto-persona in mind at the beginning of an initiative and where their needs are being served or underserved. She’s also passionate about fine-tuning models, so a lot of work is put into ensuring Jasper’s outputs are specific to a brand's voice and the marketer's task at hand.
Read on to hear Mencarelli’s thoughts and opinions on what makes Jasper so opinionated.
{{rule}}
What makes a product, particularly an AI platform, "opinionated”?
You obsess over your customer's pain points and you come up with creative solutions they never could dream possible. I view building a product like building a house. You set your concrete foundation: a simple-to-use, consumer-grade, first-time experience that gets to an “aha” moment in minutes.
Then you build the first level, which is understanding each specific persona's top jobs to be done. For instance, a content marketer writing a blog post like this one — but even that job is much broader. A content marketer is actually trying to generate leads, brand awareness, and website traffic. That’s why we moved Jasper from content creation to being a full end-to-end copilot so they can truly complete their primary goal.
From there, it’s understanding customers’ daily workflow and figuring out how we can tweak our system to accommodate them. I am a champion of always having a “next best action in the product.” This is where building an AI-first product gets interesting because I believe the UI must change based on a user's input. It starts with understanding what someone wants to do — context is the product. Over time, when we learn more about you, Jasper suggests things like low-performing content and how to make it better. It’s a cyclical relationship of tell, push, and pull.
With generative AI, an open form field where you can ask anything leads to a user doing nothing. AI is a paradox of choice, like scrolling through Netflix for hours only to fall asleep without watching anything. That is why I believe so strongly in building software that is persona-first. Marketing professionals will never be truly served by open-ended, broad tool meant to be a one-stop-shop for everyone.
Why were marketers chosen as Jasper’s target users rather than other demographics?
It started with figuring out the teams that would benefit most from AI. Across the AI industry, there are very specific buckets of individuals and needs that AI is best suited for, whereas some industries just aren’t well suited for it. Who has the use cases that fit best with AI? Instead of trying to force it into an area where it doesn't make sense, which I'm seeing a lot in our industry. That's where we get underwhelming AI products and those built in search of a customer problem instead of the other way around.
Building AI for marketing teams makes sense because they produce so much content. And that’s really where Jasper started in generative AI. This applies to the content they’re creating — whether it’s a single blog or a full campaign — all the way through seeing how that content is performing.
Now, with our recent launch, we’ve shown our customers and the world that we are a true end-to-end co-pilot. My team focuses on every phase of a marketer's journey, from research, creation, performance and more. And we love what we do because we get quotes from customers like, “I’ve fallen in love with my job again thanks to Jasper. I can spend more time on being creative and I met a recent deadline without having to work over the weekend.”
How have popular trends in AI UX informed your approach to designing Jasper?
Designing in AI is like looking at the night sky and seeing the stars: you're quite literally looking at millions of light years in the past, but you're just seeing it now. And we’re experiencing the same with how humans are engaging this new technology. It’s a very exciting time to be designing in AI but I also think that our industry is looking too much at what each other is doing. So we're getting this very bastardized experience with everything having a chatbot, all the tools being purple, and having a sparkle icon to indicate something is AI.
I always try to stress with my design team that we should be designing things that don't exist. Let's experiment and swing big and always think, “What is actually an AI-first version of this experience?”
For many years, designers have relied on guidelines and shared patterns from material design to human interface guidelines, or HIG. But designers and product folk all over the world are at the early stages of defining this new playbook. There will be a new AI copilot guideline that guides our industry; however, we are all developing it right now. Our product has been copied many times over, and it’s a big focus for our design team to define this future. And you can see the baby steps of that in our product today.
How are you designing the product around ethical, safe, and unbiased AI use?
I’m afraid to rely on some LLM creators. Because what they have shown thus far is that they do not actually care about bias or safety. We operate based on interoperability, so we can use tons of other LLMs to support amazing content creation. And we need to because there's a lot that could and does go wrong in this area.
Right now, we can flag something in the product and it goes through different safety protocols — but that's after creation. We actually want to take it a step further to pre-creation. It’s moving past, “How do you report something bad happening on the platform?” to “Let’s not have anything bad happen to begin with.” I would love for Jasper to be at the forefront of gating harmful content before it even comes out, without relying on what a model provider thinks is ethical or unethical.
LLMs base their generations off of human creations, which can be inherently unethical and biased. If we can filter out that problematic content as new things are created, I think that would be one of the biggest wins that AI could ever achieve. This has huge implications for the enterprise. You cannot create things that would harm a business. It's something that I feel very, very strongly about and that we're currently working on with our AI team.
[.blog-quote]"We should be designing things that don't exist."[.blog-quote]
We want customers to feel the ultimate trust in Jasper and that they are in control of the machine. Auto-filtering harmful content is one thing. We also have in our roadmap the ability to add specific words, phrases, and more to where an admin can say, “Do not write about this top-secret project” for instance.
At our last executive offsite, we wrote stories about where Jasper would be a year from now. Part of my story was to have our CEO Timothy Young speaking before Congress on the developments we’ve made in this area so others will contribute as well. Safety of companies' data, not producing harmful content, and leading with our moral compass are core to our values at Jasper.
What’s a recent feature you were excited to design with the team?
An insight I’ve learned is that humans expect to be in control of technology, so they want to see how it works. So all that “invisible” user experience needs to be accessible when a customer wants it or when they don’t trust it.
For example, when I was at Spotify, we prototyped the option to see your algorithm and edit it. At the time, it didn’t test well because it was overwhelming and too “techy.” I firmly believe now would be a great time to ship this feature. It would give parents around the world the ability to hide “Baby Shark” and other children’s music from their algorithm so their suggestions and year-in-review feel like them and only them.
Our version of this at Jasper is called X Ray Mode. This feature gives our customers the ability to look under the hood. They can see how we applied knowledge, style guide, their voice and more. Then if an output is right or wrong, they can give feedback to better tune their unique experience.
It’s super easy to use but it gives people full control over “the machine” when they want it. Most people may not want it and be happy with the outputs. But for the percentage that do — because they don’t trust an output or something came out wrong or they feel like the tool is not listening to them — they can “open the hatch” and see what happened.
Things are really opaque in how AI works. So we want to champion users having a high level of control over the LLMs and demystifying the process. Because at the end of the day, it isn’t the machines in control but the humans. LLMs wouldn’t exist without the creations of humans; let’s never forget that.
