Beyond the Hype: 3 Content Lessons from Winning an AI Hackathon

Merriam-Webster’s Word of the Year for 2025 is ‘Slop,’ the A.I.-generated content that fills our social media feeds and is nothing but junk. And yet, agencies and brands are evaluating ways to incorporate artificial intelligence into their workstreams. As marketing leaders in the advertising industry, we want to make sure we’re leading those conversations for our clients, not following them. As the Integrated Solutions department, we set out to host an AI Hackathon to identify potential use cases for incorporating our AI systems into our work. 

At the outset of our internal AI Hackathon, my team discussed whether tools like Gemini and Jasper could help alleviate a major pain point in influencer marketing: the "identification bottleneck." In highly regulated sectors like Cybersecurity and Financial Services, we aren't looking for "influencers" in the traditional lifestyle sense; we are looking for subject-matter experts (SMEs) with the technical "chops" and verified credentials who can authentically speak about a product or service.

Winning the hackathon required more than technical skill when it comes to crafting the ideal prompt; it called for using AI to solve the "Needle in a Haystack" problem in finding the ideal creator with an authoritative voice, whose audience primarily resided across niche social media platforms like Mastodon, Discord, and specialized Substacks.

Most agencies use a hybrid approach, switching between different software tools like a Popular Pays, Tagger, or Influential, and conducting manual searches on the Internet and social media. This fragmentation slows down influencer marketing professionals and can lead to inefficiencies. For example, an influencer’s audience data lives in one place, their contact info in another, and content vetting happens in the feed.

As a team, we acknowledged that influencer marketing for Highwire’s clients, specifically those in cybersecurity and financial services, is fundamentally different from selling lifestyle products. We aren't looking for 'Influencers' in the traditional sense. We are looking for Subject Matter Experts (SMEs). In these industries, a high follower count doesn't matter if the person doesn't have the technical chops to back it up.

Here are the key lessons I believe our team uncovered from this AI Hackathon: 

1. Prompt Precision is the New Vetting Tool

A key insight from the hackathon was that the quality of output for expert identification depends on treating prompts as you would a standard creative brief. In highly-regulated industries, the "risk factor" is massive; recommending that our clients partner with a former analyst-turned-creator who you later find out left their role under bad circumstances is a brand nightmare. To improve the results of our prompts, we, as a team, moved away from broad requests and used specific persona constraints. An example of that would look like:

• Persona: "You are an influencer marketing professional who works at an advertising agency which specializes in B2B organizations with a strong focus in cybersecurity."

• Context: "Find influential voices and experts who are considered influencers that have a track record of creating sponsored content for cybersecurity decision makers.” 

• Constraint: "Exclude 'finfluencers' who focus on lifestyle; prioritize those shaping high-level technical conversations on LinkedIn and X."

2. Gemini vs. Jasper: Casting the "Wide Net."

As an agency, we have access to the enterprise versions of two AI-powered tools, Google’s Gemini and Jasper. This investment level grants us additional capabilities and, most importantly, keeps our clients’ proprietary information secure and prevents any outputs from our prompts from being accessible to everyday users. This removes a major concern for our clients who are hesitant about sensitive proprietary information being accessed by competitors or market analysts.

As an agency, we recognized that Jasper and Gemini serve distinct functions when ideating on how best to navigate a currently fragmented tool stack. This is exactly where we saw the opportunity; we needed a way to streamline these workflows. And the key thing to note is that our submission did not involve using AI to replace the human element, but rather to speed up the initial casting of a 'wide net.’ Helping us identify potential recommendations that empower us to scale our efforts and ensure we’re finding the right creators for our clients. So what does that look like in practice?

• Gemini for Discovery: Gemini excels at analyzing real-time data to find experts who are sometimes "invisible" to the most popular influencer discovery tools. During our tests with the varying prompts, Gemini successfully identified two creators that our teams had previously identified through hours of manual research.

• Jasper for Alignment: We found that Jasper is more effective at ensuring there’s alignment between our client’s brand voice and the creators when evaluating said content against the “always on” content we create on behalf of our clients.

3. The "Human-in-the-Loop" as a Risk Mitigator

For many advertising and influencer marketing professionals, teams are facing a massive identification bottleneck. And that's why it was the crux of our build for the AI Hackathon. The reality of agency life is that we rarely get cookie-cutter requests for any social media or influencer marketing work.  And on top of that, we often deal with timelines focused on major tentpole events, like a conference or a product launch.

One day, we’re looking for a cybersecurity analyst, the next it’s someone who can authentically speak about financial education. And because no two campaigns are the same, we can’t just rely on a static database of influencers we’ve researched in the past. We are effectively starting our influencer discovery efforts from scratch every time.

And I believe our submission was a success because we demonstrated that using AI to handle the initial "wide net" phase allowed our team to focus on the high-stakes verification process. While AI can quickly identify potential candidates for an influencer activation, it cannot verify whether or not a creator is the right brand fit. By letting AI handle the bulk of the identification, our team could scale our efforts without sacrificing the deep vetting required for brand safety and impactful results.

Based on my learnings, here are some things to consider as you look to implement AI into your workflows: 

• The 80/20 Rule: Let AI handle 80% of the initial expert identification and "wide net" research, but keep 20% of the effort focused on human credential verification.

• Data Validity: Always fact-check credentials surfaced by Gemini. While it can find the "needle," it can hallucinate technical details when prompts lack precision.

• Platform Diversity: Move beyond Instagram/TikTok; use AI to scan professional, gated platforms where SMEs actually reside.

Thanks again for spending time reading this latest update! If interested, you can check out more of my writing here.