Kalshi and Polymarket moved more money in July 2026 than legal U.S. sportsbooks took in bets that month, according to a Pew Research Center analysis cited by CBS Sports. That is a big number for a category most PR professionals still file under “fintech curiosity” rather than “client risk.” My firm just finished The Prediction Markets AI Visibility Index 2026, a study that should move it into the second category, because what we found has nothing to do with trading volume and everything to do with how AI systems handle a question regulators have not finished answering.
We tested how ChatGPT, Claude, Perplexity, Gemini, and Google AI Overviews describe Kalshi and Polymarket’s legal status. Every engine hedges the same way: is this a financial exchange or a form of gambling. None of them picks a side. I want to walk through why that hedge is correct, why it will not stay correct without active maintenance, and what it means for anyone doing communications work in a regulated or contested category.
The hedge holds up because the primary sources disagree with each other, not because the AI is confused. The Commodity Futures Trading Commission’s own press materials describe Kalshi as a designated contract market, a federal financial exchange. Washington’s attorney general describes the same company’s contracts as an “illegal gambling operation” in an August 2026 court filing. Two regulators, two characterizations, both from primary sources with real authority. An AI system that resolved that disagreement in either direction would be wrong. The mixed answer is the accurate one.
Here is where it gets directly relevant to client work. We ran a second test, splitting Claude’s answers into two conditions: one informed by a full year of current research, and one relying only on training knowledge, with no live search. On questions about which states currently restrict or allow Kalshi to operate, the no-search version did not just miss a few details. It missed the record almost entirely, because Washington’s final order against Kalshi, New York’s lawsuit, and Minnesota’s injunction all happened after that training data was collected.
The AI was not reasoning poorly. It was answering honestly from a point in time that had already passed. I have said for two years that Generative Engine Optimization is not primarily about getting mentioned. It is about staying current inside a system that only knows what it has been shown. This study is the clearest evidence I have seen of what that actually costs a client. A press release from six months ago is not old news to an AI system with no live search access. It may be the only version of events that system has, and it will repeat that version with the same confidence it would use for something published this morning.
The trust and safety findings are the ones I would bring to a client meeting first, because they show the downside is not just staleness, it is false reassurance. A cold, no-search answer to “is Kalshi under investigation” did not just fail to mention the CFTC’s marketing-practices inquiry into Polymarket, opened in June 2026. It could answer no outright, missing entirely that the CFTC used its own emergency authority in August 2026 to keep Kalshi operating in New York over a state lawsuit. That is not an incomplete answer sitting a few weeks behind the news. It is a clean bill of health delivered with the same tone as a correct one, for a company under active regulatory scrutiny. Any client operating in a regulated space should hear that sentence twice.
There is a second finding I think the PR industry specifically needs to sit with, because it changes what earned media work is actually accomplishing in a contested category. We reviewed 36 sources covering this space and found the framing split does not track commercial incentive the way it usually does. Financial press and the CFTC’s own releases default to exchange and derivatives language. Sports and gambling trade press default to bet and wager language, describing the identical products. The split tracks which regulator is speaking, federal versus state, not which outlet has something to sell. That means a press release issued during an active legal dispute is not just informing reporters. It is training material for whatever AI system answers the next person who asks about the client, and the specific vocabulary in that release, not just its substance, shapes how the client gets described months later.
I would also flag the engine-level differences, because they change where an agency should actually spend effort. Google AI Overviews and Perplexity scored highest on legal currency in our modeling, because both pull from live search by default. ChatGPT scored lowest, because it does not always browse without an explicit prompt cue. That is not a permanent ranking, these products change constantly, but it tells you where a stale answer is most likely to surface right now, and it tells you the fix is not the same everywhere. You cannot optimize your way around an engine that simply is not looking. The only lever that works there is making sure the underlying record, the filing, the release, the executive statement, is accurate and dated whenever a system finally does look.
None of this is unique to prediction markets. Any client under active litigation, in any regulated industry, faces the identical structural gap between what has actually happened and what a fixed-knowledge AI system can know without being told to check again. I would encourage every agency running GEO work for a regulated client to ask a version of the question we asked here: if I strip live search away from the AI systems my client cares about, what does the answer look like, and is it still accurate. If the answer changes meaningfully, that gap is not a technical curiosity. It is exposure, and dated, sourced, actively maintained content is the only thing that closes it.
I would add one practical note for agencies building this into client work. The gap does not close itself, and it does not close once. A press release published today becomes exactly the kind of stale answer we found in this study the moment the facts move again, which in a fast-regulated category can be weeks, not years. Treat the maintenance of a client’s AI-facing record the way you would treat a monitoring service, not a one-time deliverable. The clients who will be described accurately a year from now are the ones whose agencies checked back.
Agility PR Solutions is a recognized leader in enterprise-ready AI-optimized PR tools and resources, including AI-powered media monitoring, media database, and media intelligence and measurement.
