From Keywords to Context: How Real AI Is Reshaping Programmatic Targeting

Most contextual targeting still clings to keywords. But our brains don’t work that way, and neither should your media strategy.

At Mavern, we’ve moved beyond the outdated model of matching words on a page. We’re tapping into how people actually understand content: by grasping meaning, not just terms. That’s where sentence embeddings come in, an approach rooted in real AI and inspired by how the brain recognizes similarity.

Why Keywords Are Broken

Keyword targeting was built for a simpler internet. Today, it’s too literal, too brittle, and too easy to fool.

• No nuance: “Calcium” on a geology page gets the same treatment as a supplement guide.

• No meaning: “Heart health” and “cardiovascular wellness” are invisible to each other.

• No scale: You’re stuck managing endless lists that can’t keep up with the pace of content.

  
  

It’s a system that’s rigid when it should be flexible—and blind where it needs insight.

The Embedding Revolution: How AI Understands Language

Modern AI models like OpenAI’s and Cohere’s don’t look for keywords, they look for meaning. They translate entire sentences into mathematical representations that capture intent, emotion, and relevance.

Examples of semantic matching:

• “Running shoes for bad knees” ≈ “Supportive sneakers for joint pain”

• “TikTok trends dominating fall wardrobes” ≈ “What Gen Z is wearing back to school”

• “Navigating sleep issues after 45” ≈ “How menopause affects your rest”

  
  

These aren’t just close matches, they’re spot on.

Why This Works: Measuring Meaning

Once you’ve mapped meaning, you need a way to compare it. That’s where cosine similarity comes in.

It doesn’t care how long the content is, it cares how aligned the ideas are.

• Long articles don’t drown out short blurbs

• Paraphrased language isn’t a barrier

• Relevance is measured by intent, not token count

  
  

It’s fast, scalable, and it works the way humans think.

A Telling Side-by-Side: Keywords vs. Embeddings

Feature	Keyword Targeting	Embeddings + Cosine Similarity
Relevance	Spotty	Consistently strong
Handles Synonyms	Rarely	Automatically
Scalability	Manual, high-touch	Automated, AI-powered
Brand Safety / MFA Filters	Weak	Strong (semantic-aware)
Customization	Static lists	Dynamic, campaign-specific
Integration	Basic	Full DSP/SSP compatibility

Real Campaign Results

Car: Automotive

Goal: Promote a luxury SUV to affluent families

• Keywords matched: “SUV,” “lease,” “test drive”

• Misses: Crash reports, teen driving tips

• Embeddings found:

  • “Best weekend getaways within 3 hours of NYC”

  • “How one dad uses his third-row SUV for soccer and school runs”

    
    

Hospital: Healthcare

Goal: Reach women managing menopause

• Keywords matched: “menopause,” “hormones”

• Misses: Teen puberty content, irrelevant wellness posts

• Embeddings surfaced:

  • “Top supplements for women in midlife”

  • “Managing sleep changes after 45”

    
    

Shopping Bags: Retail

Goal: Back-to-school fashion for Gen Z

• Keywords matched: “school,” “supplies”

• Misses: News about strikes, school funding

• Embeddings nailed:

  • “TikTok fashion trends for fall”

  • “Best backpacks under $60, ranked by teens”

    
    

The Bottom Line: Smarter Context = Better Performance

If you’re still relying on keyword lists, you’re not just behind—you’re invisible in the moments that matter.

Mavern’s semantic targeting platform is built on real AI. We don’t guess at relevance—we measure it. And we do it at scale, across every major SSP.

Let us show you how it works—live, with your next campaign.

With Mavern Media: Cleaner supply. Smarter decisions. Performance that scales.