Story one
Any venture can say it uses AI. Any founder can post about their struggles. Both claims cost almost nothing to make, so classical signalling theory says they should convey nothing. Audiences act on them anyway.
28,291 Product Hunt launches. One mark per launch; the bright lobe is the segment able to verify the claim. These fields are generated, not plotted. Each draws one mark per observation in the study it belongs to and its structure mirrors the finding, but the coordinates are not measured values.
The idea
Cost was never purely a property of the claim. To an audience that cannot verify, an AI claim is cheap talk; to an expert who can inspect the product, the same words invite scrutiny a weak product will fail. And once a platform makes third-party behaviour observable, substitution between signals moves out of the sender's portfolio and into the receiver's information environment. Both halves say the same thing: the receiver, not the sender, sets what a signal is worth.
In the world right now
Coverage by journalists, regulators and research bodies — none of it produced by this research programme. It is included because it is the world the argument speaks to. Every item was checked against the live source.
A $1.5 billion AI app-builder collapsed after former staff said outsourced human developers, not AI, did most of the work.
The identical claim sustained a unicorn valuation among generalist investors while engineers who could inspect the output saw through it. Two audiences, two costs.
US prosecutors charged nate's founder with raising more than $40 million by claiming AI automation in fact performed by overseas workers.
Enforcement makes an unverifiable claim costly after the fact. That is the mechanism turning cheap talk into a genuine signal.
The FTC barred Workado from claiming AI-detection accuracy without competent and reliable evidence to substantiate it.
A substantiation requirement is a verification cost legislated onto claims the market could not verify on its own.
A securities litigator argues AI-washing suits and regulatory scrutiny are accelerating, echoing the dot-com and greenwashing enforcement waves.
The audience is shifting from credulous to expert. Plaintiffs' lawyers and regulators are now a receiver class for whom AI claims carry real risk.
Researchers identified roughly six million suspected fake GitHub stars used to inflate repository credibility, often to distribute malware.
When the cheapest popularity signal is trivially faked, discerning receivers move to harder third-party evidence.
Curl's maintainer ended the project's bug bounty after AI-generated, poorly researched vulnerability reports overwhelmed security triage.
Once producing a plausible contribution costs nothing, the signal collapses and the receiver must change the incentive rather than read the claim.
Kubernetes retired the widely deployed Ingress NGINX controller after maintainer burnout, with one or two unpaid volunteers sustaining critical infrastructure.
A well-documented, widely adopted project can still lack the contributor activity sophisticated adopters should be reading as the real signal.
Global trust in news fell to 37 percent, its lowest since 2015, while concern about false information online rose to 62 percent.
The information environment in which founders and audiences both form judgements is measurably degrading.
The mechanism
The evidence
Peer-reviewed and accepted papers only. Work still under review is summarised below rather than listed, because a manuscript in review is not yet evidence of anything.
This is the newest strand. The published work above establishes how firms judge what they cannot directly observe; the platform-scale studies of AI claims, vulnerability signalling and open-source governance that carry the argument furthest are under review. Details follow publication.
The second move
So what
Establish which audience will price your claim before you make it. Regulators and litigators are steadily converting the cheap version into the costly one, and durable credibility comes from evidence a sender cannot manufacture.
Honestly
Audience sophistication is inferred from behaviour rather than measured directly: a commenter who engages technically is treated as expert, which is reasonable but not verified. The platform findings also rest on what is publicly visible, and much of the real evidence a sophisticated adopter uses may sit in private channels. The work is concentrated in English-language, largely Western communities, and whether audience-imposed cost behaves the same way where verification norms differ is open.