How LLMs are reshaping CTV planning, activation and measurement

As large language models (LLMs) improve, ad leaders across the industry — from holding company giants such as WPP to independents like Monks — are embracing this technology to manage the growing complexity of CTV media planning and buying.

Generative AI is emerging as a way to cut through fragmented inventory and enormous amounts of campaign data, helping teams surface opportunities that are difficult to identify through manual analysis alone.

Experts who spoke to The Current said that when used correctly, LLMs don’t replace the media planner or buyer. Instead, they “compress time, surface nonobvious insights and help humans make better decisions faster,” explained Tom Burchill, group director of programmatic at Monks.

“[An] LLM is not running the campaign,” agreed Harry Tong, director of solutions engineering at PubMatic. “It is facilitating a conversation with systems that already know how to execute at scale.”

As the industry looks ahead to 2026, here’s how experts say marketers can apply LLMs across planning, activation and measurement in CTV.

Planning

The planning process often begins with a natural language brief. A planner might select an LLM such as OpenAI’s ChatGPT or Anthropic’s Claude and input a prompt like: “I want to reach sports enthusiasts in the U.S. on premium CTV with a focus on live events,” Tong told The Current.

At PubMatic, the LLM translates that request through an “AI connectivity layer” into its agentic system. “Inventory is then searched, matched and returned based on real supply, real constraints and real forecasting,” Tong added.

For agencies like Monks, LLMs function as a “strategic accelerator” during this early phase. “Our teams use them to cobble together fragmented inputs — audience research, historical performance, publisher mix, device usage, genre consumption and macro viewership trends.”

This is particularly useful in CTV, where inventory is fragmented across platforms and supply paths. “An LLM can help identify pockets of opportunity — such as underutilized streaming environments, audience overlaps across platforms or emerging content categories — much earlier in the process than a human working manually.”

Activation

During activation, the value of LLMs becomes more tactical. This is where Monks sees its largest productivity gains.

“Our teams regularly feed large-scale reporting outputs — domain-level delivery, device performance, creative rotation, audience segment exposure, frequency curves — into LLMs to identify statistically meaningful patterns that would be difficult for a human to spot quickly,” Burchill said.

“In CTV, where campaigns can generate hundreds of thousands or even millions of rows of log-level or aggregated data, this matters,” he added.

Still, execution requires caution. LLMs can be “overly reactive or confident” when interpreting short-term signals. “The winning model, in our experience, is AI for pattern detection and human judgment for decision-making,” Burchill noted.

Misha Williams, COO at GWI, added: “In our study of marketing and advertising professionals, we found that 4 in 10 are already using AI for audience insights, and a similar number expect AI agents trained on real consumer data to produce higher quality work. The difference is that human insight grounds AI outputs in how people actually behave, not how we assume they behave.”

Measurement and optimization

Measurement is another area where LLMs can unlock differentiated strategic value — especially for CTV, which often plays an upper-funnel role.

“[CTV’s] impact is rarely isolated in a single report,” Burchill said. “LLMs help us connect dots across channels, understanding how CTV exposure influences downstream performance in paid search, social or even offline marketing efforts.”

“The ability to reason across datasets and apply learnings from CTV to the rest of the media mix is where we see long-term value emerging.”

On the optimization side, LLM-driven insights can either trigger “automatic” adjustments or be surfaced to buyers for approval, according to Tong. “The end result is a much simpler workflow without giving up transparency, control, platform choice or performance.”

Increased efficiency

Across all stages, the biggest overall benefit of LLMs in CTV is “workflow efficiency,” Tong said.

“Tasks that historically required multiple platforms, different user interfaces and a lot of manual handoffs, such as inventory discovery, deal setup and campaign configuration, can now be handled conversationally in minutes,” he added.

PubMatic data shows that LLMs reduced campaign setup time by 87% and sped up issue resolution by 70%. “Things that used to take days now take minutes. Those gains will only compound as this technology matures,” he added.

Teams that are strong AI operators are “meaningfully faster,” Burchill echoed.

“We consistently see reduced time to insight, quicker optimization cycles and more structured post-campaign analysis. In some cases, that translates to weeks shaved off learning agendas or faster identification of underperforming supply paths.”

Newer models, such as GPT-5.1, are also improving the quality of insights. “The integration of third-party research sources like GWI, combined with stronger reasoning and conversational memory, makes the tool feel more collaborative and less transactional,” Burchill said.

“For planners, that means fewer back-and-forth prompts and more cohesive strategic outputs, particularly when pressure-testing audience definitions or contextual assumptions in CTV environments.”

That said, LLMs are not infallible. “[They] often rely on lagging or incomplete source data and produce overly agreeable outputs that sound right but aren’t actionable,” Burchill said. As a result, Monks says it’s “mission-critical” to train teams not just on how to use LLMs, but how to challenge them.

Another limitation is that LLMs cannot transact, optimize or measure media on their own. “Without plugging into robust infrastructure, high-quality supply and proven decisioning logic, it is simply a conversational layer,” Tong explained.

What lies ahead

Looking ahead to 2026, Tong believes agentic AI will unlock concrete operational efficiencies across the CTV ad buying ecosystem.

“We are already seeing live agentic campaigns meet or exceed KPIs across premium CTV inventory. Campaigns will increasingly be built around outcomes rather than legacy workflows. Buyers will be able to move directly from a natural language media plan into activation. Optimization and troubleshooting will become much more autonomous, with buyers choosing how much decision-making they want to delegate to agents,” he said.

This is not about removing humans from the loop, he emphasized. “It is about removing friction from the system so people can focus on strategy, creativity and outcomes. LLMs become tools that help humans make better and faster decisions.”

CTV content itself may even become more dynamic with agentic AI. We could see elements of longer-form content generated or adapted in near real time, responding to cultural moments, live events, or shifts in user behavior, according to Gina Whelehan, group director of strategy and partnerships at Butler/Till.

“This opens up possibilities for relevance and engagement, while also raising the bar for human oversight. As machines take on more of that real-time adaptation, the role of humans becomes even more critical in setting guardrails, deciding when to intervene, and ensuring cultural context and brand integrity are preserved.”