Artificial intelligence has changed the mechanics of political advertising faster than the market has changed its standards for governing it. Campaigns now move across connected TV, online video, display, native, and audio through automated pipes built for speed, not for ambiguity. That matters because political advertising is one of the few categories where monetization, disclosure, trust, and regulations collide at the same moment. In the 2025–2026 cycle, that pressure is only intensifying: AdImpact projects U.S. political ad spending at $10.8 billion, including $2.5 billion in connected TV and streaming, making this the most expensive midterm cycle on record.
Punit Shah, Director of Product Marketing, technical leader at PubMatic and Senior IEEE Member with over 15 years of experience driving AI-powered platforms, product strategy, and go-to-market execution across the technology and media ecosystem has built his recent work around resolving that tension. His work on AI-driven political advertising classification, including the design and commercialization of frameworks such as Creative Category Manager, represents a body of contribution to how the open internet governs sensitive advertising categories—work that has since influenced publisher infrastructure, compliance standards, and transparency tooling across the programmatic ecosystem. His focus sits at a part of the industry most people never see directly: the infrastructure layer that determines whether political advertising can scale across the open internet without collapsing into opacity, manual bottlenecks, or policy inconsistency. This work addressed a problem that has become central to modern democratic media systems: how to support political messaging at auction speed while preserving transparency, publisher control, and auditability.
“Political advertising does not become safer by making it slower,” Shah says. “It becomes safer when transparency and policy enforcement are natively built into the system that makes the decision.”
Manual Review Breaks at RTB Scale
The hardest part of political advertising today is not reach. It is identification. A political creative is rarely defined by a single text string or disclosure line. It can signal intent through spoken language, visual context, sponsor disclaimers, candidate references, issue framing, or combinations of all four. In automated advertising environments, that complexity becomes operationally serious because every delay introduces commercial friction, and every miss introduces compliance risk. Manual review may still work for isolated campaigns or static formats. It does not work in live bidding infrastructure, processing decision-making at machine speed.
That is the problem Shah’s work was built to solve. He led the design and commercialization of a multi-pass, multi-modal framework that combined automatic speech recognition for audio transcription, OCR for disclaimer extraction, video frame sampling, facial and contextual analysis, NLP-based political intent detection, candidate and issue tagging, and supervised classification for policy enforcement. The significance of that design is not just technical. It changes when governance happens. Instead of treating classification as a cleanup step after delivery, it makes classification part of the auction path itself. That is the difference between reactive moderation and operational control.
“The critical question is not whether a platform can review political content eventually,” Shah says. “The question is whether it can understand enough, fast enough, to make a responsible decision before the transaction happens.”
Real-Time Classification Enables Policy Enforcement
The industry conversation around political advertising often gets trapped in the wrong debate. It is usually framed as a choice between openness and control, scale and safety, and monetization and accountability. In practice, the real divide is between systems that can govern intelligently and systems that can only govern bluntly. When platforms lack reliable classification infrastructure, the fallback is broad restriction, fragmented policy handling, or labor-intensive exception management. None of those models hold up well once inventory expands across streaming, programmatic video, audio, and publisher-specific rules.
That broader market gap is now well documented. A 2026 study on political advertising across streaming services and connected TV found that political ads in these environments have expanded quickly while regulatory visibility and disclosure consistency have lagged behind, leaving major blind spots in how these ads are identified and governed. That is precisely where Shah’s work sits. His framework was designed not only to identify political content but to support publisher-level policy controls, explainable outputs for auditability, and transparency tooling that could stand up to federal and state election requirements.
This is what made the system commercially viable as well as compliant. Shah’s role extended beyond architecture into AI-to-market translation, regulatory alignment, publisher enablement, and global go-to-market execution. He worked across engineering, data science, legal, compliance, product, and commercial teams to build a framework publishers could actually adopt. More than 250 independent publishers and 420 domains were able to open political advertising inventory through this infrastructure while maintaining clearer rules for how sensitive demand would be classified and handled.
“Responsible scale starts with publisher control,” Shah says. “If the system cannot explain what it saw, why it classified the creative, and how policy was applied, then it is not ready for political advertising.”
Infrastructure Enables Transparency and Publisher Control
The strongest dimension of Shah’s work is that it treated political advertising as an infrastructure problem rather than a policy memo. That distinction matters. Political advertising does not break systems because the category is controversial. It breaks systems because the category forces platforms to make fast decisions under conflicting requirements. They have to support monetization, enforce rules, detect disclosures, respect publisher preferences, respond to jurisdictional variation, and preserve records that can withstand scrutiny later. Most legacy ad operations were not designed for that mix.
Shah’s framework addressed that by embedding AI classification layers directly into live RTB auction pipelines that process trillions of bid requests annually. It was built for millisecond-level performance while still producing outputs usable for transparency reporting and review. The system automated work that had historically required manual classification. reduced approval latency for campaigns, standardized policy enforcement across all 50 U.S. states, and created a structure that could be extended into future European regulatory contexts. That combination is what gives the project industry-first significance. It was not merely a detection tool. It was a governance architecture designed to function inside high-scale programmatic infrastructure.
The results were measurable. During the 2024 U.S. federal election cycle, the framework processed political advertising across CTV, online video, display, native, and audio at scale, demonstrating that AI-driven classification infrastructure could support high-volume monetization without sacrificing compliance (99% classification accuracy) or transparency. It also reduced risk exposure for a NASDAQ-listed company operating in one of the most scrutinized advertising categories in the market.
“The future of political advertising will not be decided by who has the most inventory,” Shah says. “It will be decided by who can make high-speed decisions with enough transparency to earn trust from publishers, advertisers, regulators, and voters.”
Trust Will Be Decided by Infrastructure
That trust question is becoming more urgent because the category itself is changing. The challenge is no longer limited to ordinary disclosure enforcement or sponsor labeling. Synthetic media has introduced a second layer of pressure, where platforms may need to distinguish not only political content from non-political content, but credible political communication from manipulated or misleading media artifacts. Reuters reported on March 28, 2026, that AI-generated deepfakes are already influencing the U.S. midterm environment, while regulation remains fragmented and largely state-driven.
That environment raises the stakes for systems like the one Shah helped lead. The market does not need more vague commitments to trust and safety. It needs operational systems that can classify, document, and govern political content in ways that are fast enough for programmatic markets and rigorous enough for democratic scrutiny. This is where Shah’s work has broader importance beyond one company or one cycle. It points toward a model in which political advertising is not treated as an exception handled through manual escalation but as a governed category with infrastructure built specifically for its demands.
His contribution is best understood in that context. He did not simply help commercialize a feature for a regulated ad segment. He helped build a framework for how the open internet can carry political messaging without abandoning transparency or publisher autonomy. That is a more consequential problem, and it is one the industry will keep confronting well beyond 2026.
“Political communication at scale only works when systems are designed to preserve both access and accountability,” Shah says. “That balance is not a policy slogan. It is an engineering and product discipline.”
