This month, Cognitiv announced a major upgrade to ContextGPT™, the advertising industry’s first chat-based contextual targeting tool. This upgrade makes our contextual product up to 40% more accurate in ad placement, all without relying on cookies, pixels, or user data. 

For the first time, advertisers can use Interactive Audience Exploration to describe entire audiences in plain language and instantly explore intelligent recommendations for contextual environments that will reach them. This moves contextual targeting beyond matching content alone, enabling custom and precise audience-level strategies. 

Our press release and product demo go into great detail about the product itself, and while I do recap some of those exciting features below, I wanted to take this opportunity to share the story behind the product upgrade, because frankly, I’m really proud of our team and what we’ve accomplished.

This project showed me how far curiosity can take us when paired with a team willing to challenge the obvious path. For me, it’s a testament to what happens when we lean into questions, explore bold ideas, and build the culture that makes breakthroughs possible.

Mohsen’s Big Swing at Top Golf

I’ll never forget the moment. We were at a company all-hands meeting in Las Vegas—Top Golf, of all places—when Mohsen, our staff machine learning (ML) scientist, pulled me and two of our team leaders, Jason and Heng, aside with an idea.

At the time, our science team in Bellevue, WA had been working on ContextGPT’s relevancy engine and mapping out how to expand its “sliders”—controls for sentiment, inclusivity, and other dimensions that helped connect ads with the right audiences. We fully expected to keep investing in this approach throughout 2025.

But Mohsen suggested something different. Rather than building more and more sliders, he had a clever idea for how to make the prompt itself smarter. If ContextGPT could better interpret what advertisers actually wanted, we wouldn’t need to keep creating one-off models for every new dimension.

His proposal forced us to reconsider the path we’d been on since releasing ContextGPT in 2023. Sliders had proven effective at capturing abstract qualities like sentiment on webpages, but each new slider required its own dedicated model, which added complexity. Mohsen was pointing us toward a more elegant solution: let the advertiser’s prompt define relevance directly.

From Idea to Proof of Concept

Of course, an idea, no matter how clever, is only the beginning. To prove it out, I asked Mohsen to run a head-to-head comparison to show how his prompt-based method performed against our slider models.

Within weeks, he returned with data. The new method not only held its own, it was just as accurate as dedicated slider models. More importantly, it would allow advertisers to define what “relevant” meant for their campaign, rather than forcing every plan into a fixed set of dimensions. 

That combination of accuracy and adaptability was a breakthrough. That being, a system that could match the performance of custom-built models while freeing planners to guide it with their own intent.

Why The Upgraded ContextGPT Is a Breakthrough

Most contextual tools still rely on fixed keyword segments, targeting pages that fit within broad buckets like “sports,” “travel,” or “finance,” regardless of whether the content truly fits the campaign’s goals. 

Sure, this method is straightforward, but it treats relevance as a one-size-fits-all and misses nuance: who a page is written for, how it’s written, or why it resonates with a particular audience. 

When we first launched ContextGPT in 2023, it was ahead of its time because it went beyond keywords. Instead of just keyword matching, we could represent both advertiser prompts and webpages as vectors in a high-dimensional space, then use cosine similarity to find pages “closest” to a given prompt. That was a leap forward over keyword-based methods, because it captured semantic closeness rather than just shared words.

But cosine similarity has a big limitation: it cannot distinguish which kind of similarity matters. For example, an advertiser looking for uplifting content might get results that literally contain the phrase “positive sentiment” rather than pages that actually read as optimistic. 

That’s why sliders became necessary in the original ContextGPT. They gave planners control to steer results toward intent dimensions when prompts alone weren’t enough.

The next evolution of ContextGPT removes that extra step. Instead of going straight from a prompt to cosine similarity, it now preprocesses the prompt with a reasoning model and applies a multi-layered scoring method to interpret intent directly. This makes “relevance” flexible, able to reflect audience, style, content type, and semantic meaning, rather than forcing every campaign into a single definition.

Bringing This Breakthrough to Market

Of course, showing that the idea worked was only step one. The early version of the new relevancy engine could take ten or fifteen minutes to process a single prompt. That might be acceptable in a research setting, but advertisers need answers in seconds. At that speed, the tool stops being a batch process and becomes interactive. Something planners can explore, test, and refine in real time as they shape their campaigns.

This is where our engineering team came in. While Mohsen and the science team had shown that the prompt-based approach was possible, engineering figured out how to make it interactive. They found clever ways to run tasks in the background and parallelize key steps, streamlining the pipeline so what once took minutes could now be done in just a few seconds.

Again, this collaboration, science proving what could be done, engineering making it fast and scalable, turned an exciting idea into a functional product. 

What ContextGPT Means for Advertisers

The upgraded ContextGPT vastly improves the campaign planning experience, making it more powerful, more performant, and easier to use.

Outcome-Driven Targeting in a Conversational Interface

If you’ve ever planned a campaign, you know that the hardest part is not deciding what you want to achieve, but translating that vision into the levers, toggles, and checkboxes a platform provides to you.

The upgraded ContextGPT changes that.

Featuring the industry’s first chat-based user interface, ContextGPT makes campaign planning conversational and intent-driven versus dependent on traditional campaign levers. 

What does this look like in practice?

Similar to ChatGPT’s user interface, advertisers can now chat directly with ContextGPT, allowing them to plan campaigns, powered by deep learning, in their own natural language. 

Want to reach consumers planning a vacation? Simply share your campaign goals, media brief, or target audience, and ContextGPT will provide robust content recommendations, no data required.

Transparent Reasoning and Smarter Suitability Controls

One of the most common frustrations with advertising tech is when it feels like a black box. You set your inputs, the system spits out recommendations, but you’re left wondering why those choices were made. 

ContextGPT makes these choices crystal clear. 

With the latest upgrade, every content recommendation comes with transparent reasoning, showing advertisers not only the suggested pages and content categories, but the logic behind our AI’s recommendations.

Check suggested content categories for ads themselves, refine those suggestions based on any slight nuances, and ultimately build trust in the system. 

Advertisers gain even more flexibility through smarter suitability controls. Whether that’s through applying inclusion and exclusion lists to ensure campaigns reach only the most relevant environments, using brand suitability filters to align messaging with safe and supportive contexts, or avoiding made-for-advertising (MFA) content to reduce wasted impressions, ContextGPT gives advertisers the tools to define what “right fit” means for their brand. 

The result is greater confidence that every dollar is spent in environments that strengthen brand equity.

Continuous Learning and Quicker Campaign Turnaround Times

Campaign planning is rarely one-and-done. Goals shift, creative evolves, and strategies are refined over time. With ContextGPT, you don’t have to start from scratch every time you log in to create your next contextual campaign. 

Our AI remembers your inputs across sessions, allowing for an iterative planning process that mirrors how strategists actually work. If yesterday you outlined a target audience and today you want to adjust messaging or exclusions, ContextGPT picks up right where you left off. 

This continuity translates directly into quicker turnaround times. Teams can move from brainstorming to activation faster, test new strategies with confidence, and react in real time to market shifts. 

So, instead of campaign planning feeling like a reset button every week, it becomes a fluid conversation, one that saves time while improving results. 

Looking Ahead

When I think back to that quick conversation with Mohsen at Top Golf, I’m still amazed at how one idea set us on the path to transforming ContextGPT. It’s a reminder that breakthroughs often start small, with a question, a spark, or a challenge to the way things have always been done. 

At the same time, the pace of AI continues to accelerate. LLMs have opened new doors, but it’s our team’s work, building a proprietary relevancy engine around those advances, that makes this technology usable, fast, and precise for advertisers. That blend of external progress and internal innovation is what sets ContextGPT apart. 

We believe ContextGPT is the most powerful contextual targeting solution available today, but in many ways, we’re just getting started. This launch builds on Cognitiv’s history of industry firsts, and we’re excited to keep pushing the boundaries of what’s possible in contextual advertising. 

Request a demo and see how the new ContextGPT can transform your campaigns. 

‍