Talking to people is stressful—even and especially when it comes to talking about networking.
That was users’ feedback on a new feature being piloted by FutureFit AI, a workforce platform connecting employers and jobseekers. The team recently built an AI networking coach they called SideBar, designed to help users mine their networks for new job opportunities.
“I thought that working with this agent [SideBar] felt nearly stress-free compared to having a talk with an in-person mentor,” one user in the pilot said. “I know that I mentioned I haven’t had a real career counselor before. But I do have a non-work-related mentor and sometimes conversations like this would cause me anxiety in everyday life.”
His feedback sits in the messy middle of what it means to build self-service tools that help students and jobseekers navigate a stressful social world. Can an AI coach help users overcome that stress and actually start turning to people? Or, in an attempt to alleviate anxiety and fill in support gaps, will it ironically end up being a crutch that pulls people further toward technology and away from each other?
Those were some of the questions my colleague Anna Arsenault and I set out to explore in a recent project with FutureFit AI and nine other edtech platforms. The project aimed to embed “prosocial” designs—designs that put tech in service of human connection, not in place of it—into state-of-the-art navigation and guidance tools.
Last fall, we launched a small fund to support these platforms in developing and measuring functionalities geared toward developing students’ and jobseekers’ social connections and confidence. Each platform tested features like SideBar directly with users, soliciting feedback through surveys or focus groups. Our new report, “Prosocial Possibilities: Building Tech that Builds Students’ Connections,” documents what we learned.
The project builds on our prior research, which revealed a concerning trend in the edtech market: most college and career guidance tools are built for dispensing information, not fostering connection. They give students real-time guidance and information about colleges, training programs, and careers. But most aren’t designed to deepen and diversify users’ networks.
That’s a blind spot given that these platforms are ultimately trying to successfully launch students into jobs—half of which come through social connections.
The good news? A growing community of edtech leaders is trying to address that gap head-on.
Engineering Networking Skills—and Mindsets
Among the 10 edtech platforms we worked with, five built functionalities to explicitly teach students the art and science of building a network. For example, Roadtrip Nation, which just released an AI-enabled virtual road trip tool, piloted a career conversations module that taught students about how to reach out to and talk to professionals about their career journeys.
REACH Pathways, a platform that helps high school and college students explore careers, build real-world skills, and access internships, piloted a quest where students identified “hidden helpers” in their networks and drafted outreach notes to get in touch.
Their designs went deeper than “social” features in edtech typically do, directly integrating empirical research on how to develop networking self-efficacy. For example, both platforms embedded evidence-based messaging aimed at cultivating prosocial mindsets.
The messages came from professor Ko Kuwabara’s studies, which document the importance of growth mindsets in networking. As Kuwabara writes, “If you believe that networking is a manipulative, false, exploitative enterprise, then it will be for you. But if you choose to exercise your agency and bring your best to each new person that you meet, the possibilities for personal and career growth may be endless.”
His research illustrates where tech-enabled efforts to “nudge” students toward networks could fizzle out: if a student has made up her mind that she’s bad at networking, all the nudges in the world might not inspire her to build the network she needs to get the job she wants.
Instead, before prompting students to reach out to people, platforms like Roadtrip Nation and REACH Pathways used statements directly from Kuwabara’s research to prime users to approach networking with a growth mindset, likening it to learning a new language or going to the gym.
They also integrated reflection questions and activities that prompt users to identify their existing networks, or what researchers dub their webs of support. Studies have shown that helping students map their networks can actually increase the likelihood they tap those connections when they need help.
Their efforts to wire research on network-building directly into tools are showing early promise.
For example, 75% of FutureFit AI users said they were likely or very likely to use SideBar again. And in a post-survey, 92% of REACH Pathways users reported increased confidence reaching out to someone in their network for support or advice. Similarly, using pre-post surveys, Roadtrip Nation users reported a 4.2 percentage-point increase in confidence leading a career conversation with a real professional and a 6.7 percentage-point increase in confidence booking a career conversation after using the career conversations practice tool.
Those boosts weren’t necessarily about eliminating students’ fears and anxieties, but about matching those fears with courage. As one student put it, “I learned that it’s not easy talking to someone from a certain field you’re interested in. But once you get used to it, you can be braver in your interviews.”
A Critical Counterpoint to Emotionally Supportive AI
These pilots were small in scope, but they mark a radical departure from traditional AI tools on the market. They’re showing a path toward using AI not just as a student self-help machine but as a human-help multiplier.
That comes at a crucial moment, when research suggests that confidence and willingness to turn to people instead of AI may be moving in the wrong direction.
For example, researchers from Imperial College London, MIT, and Harvard recently released a study on what causes users to turn to AI for emotional support and how that tendency compounds over time. Their findings? Users often stumble into moments of emotional support in the course of using AI for tasks. When that AI-generated support resonates, it draws users in. “These incidental encounters are path-dependent,” the study concludes. “Positive experiences of AI emotional support update people’s beliefs about AI’s emotional capabilities and redirect their choices for future emotional support, increasing preference for AI and decreasing preference for humans.”
That should be an alarming trend given what research has shown about help-seeking avoidance and social capital: the less willing students are to reach out to people for help, the less likely they are to develop the kinds of social capital that contribute to their academic and career success.
“Our longitudinal evaluations have shown that help-seeking avoidance is a pervasive barrier for incoming college students,” says Jean Rhodes, a leading scholar of mentoring and social capital development. Rhodes and her colleagues have been working to flip that script, with a curriculum called Connected Scholars. They’ve found that when students are equipped with the skills to reach out for help, it has direct, positive impacts on their sense of belonging, likelihood of recruiting mentors, GPA, and retention. Rhodes is particularly worried that students could lose those skills in the age of AI.
“When students retreat to chatbots, they risk atrophying the very relational skills they need to thrive,” she says. (Rhodes recently released a free, digital, single-session flash course aimed at building those relational skills and mindsets.)
Still, for edtech platforms, striking this balance between emotional support and dependence is tricky—these tools need to feel supportive to be sticky, but not so sticky that they make the kinds of human interaction needed to build a network less appealing.
And knowing whether a tool is successfully scaling meaningful human interactions is tricky unto itself. While tools managed to measure usage patterns and changes in confidence, offline behavior change is the holy grail of these experiments, and it lags self-reported confidence scores and engagement rates.
Online Solutions to Shift Offline Behaviors
Boosting that behavior change was what another platform in our pilot project, Climb Together, wanted to test. Climb Together’s chatbot, Goldi, is one of the more advanced tools on the market, drawing on years of social capital data and curriculum from the organization’s sister program, Climb Hire, a career training program to help overlooked talent break into upwardly mobile jobs.
At the start of the pilot, founder Nitzan Pelman was hoping to tackle a concerning trend she’d observed among users: after engaging with Goldi, the vast majority of students reported being significantly more confident telling their story and reaching out for informational interviews. But far fewer followed through with actually having a career chat that expanded their network.
“What we realized is that we needed to create a ‘forcing function’ where students are compelled to actually do something that feels deeply uncomfortable to them,” Pelman says. “The forcing function can be through a human coach or mentor asking them to practice with AI and then follow through with a human. Or a faculty member or teacher can offer credit in a college course or high school class.”
Pelman and her team tested various approaches to address that gap, from prompting users to reach out to existing connections (rather than new ones), to practicing with peers, to embedding the tool within an education-to-employment program.
The program context appeared to be the most significant driver of offline behavior change: when they piloted Goldi with the education-to-employment nonprofit COOP Careers, which requires participants to grow their networks over the course of the program, students’ rates of outreach and informational interviews soared. LinkedIn-specific outreach frequency increased 62.5%, and self-rated LinkedIn message quality jumped nearly a full point on a 5-point scale—the single largest magnitude shift in their entire dataset. Overall, network outreach frequency rose 36.6%.
Pelman’s takeaway? While programs with stronger relational contexts are great, scaling impact to as many students and jobseekers as possible will require more forcing functions. “Because most educational institutions don’t teach the value of relationship building, we need to create that understanding in classrooms and then find clear ways to obligate students to take action, meet people, and have real career chats,” she says. “When that happens, we see students really having a lightbulb moment and seeing the direct impact of relationship building as they actually unlock job referrals.”
For other platforms, encouraging connection was more about scaffolding asks than forcing them.
For example, in the course of student focus groups, Uprooted Academy founder Tiffany Green found that for young people, asking for help isn’t just daunting—it can feel directionless. “Many students had people they could turn to, like family, teachers, mentors. Fewer students knew exactly what to ask those people for or how to ask. There was a gap between ‘I have people who support me’ and ‘I know what to ask them for to move forward.’ … Specific prompts (e.g., ‘ask X for Y’) are more doable,” says Green.
Green and her team built out functionalities that asked students’ support networks to share their “superpowers” and nudged students to tap into those specific supports.
Deepening Understanding to Deepen Connections
The 10 pilots in our report offer a glimpse into what it will take to build edtech that connects. In a turbulent labor market, connections carry currency. In other words, making connecting with people more “doable” isn’t just a nice-to-have. It’s the difference-maker between platforms that lead students toward potential careers and those that successfully open doors to actual jobs.
That turns out to be a psychological challenge, not just a logistical one. Other humans make us anxious—especially in the parts of our education and career journeys that feel vulnerable and high stakes.
AI’s nonhuman solutions to quell that anxiety could be a fix or a crutch. We’re still trying to figure that out.
Julia Freeland Fisher is the director of education research at the Clayton Christensen Institute and the author of Who You Know: Unlocking Innovations That Expand Students’ Networks.