The Research Firm That Wasn’t

For a few days in August, a polling company that didn’t exist produced political facts that did.

Starting on August 9, according to reporting by the Los Angeles Times, a previously unknown research firm called Median Strategies published surveys in three states, including one showing Los Angeles Mayor Karen Bass leading a reelection rival by nearly 12 percentage points. The Bass campaign shared the result as evidence that it was “gaining momentum.” A news outlet reported it. Trading followed on prediction markets.

Then the company turned out not to be a company at all.

Median was created by Rahil Prakash, a 21-year-old college graduate. Prakash told The Guardian that he alone invented the organization and its polls and used AI to build its website. After reporters began asking questions, he withdrew the surveys and shut down the site. An unsigned statement written in the institutional voice called Median a “short-term social experiment” into how false polling could spread without independent verification.

The company was one guy. The research was fake. The responses it produced were real. The Associated Press found small movements and unusually concentrated prediction-market trading after Bass shared the Los Angeles poll. There is no evidence that Prakash traded on the information, and the episode did not determine an election. Its importance lies elsewhere.

Median didn’t merely fabricate a claim. It fabricated the institutional machinery that was supposed to make the claim credible. A name, website, methodology and set of quantitative findings carried its numbers into campaign communications, media coverage and financial markets before anyone established what stood behind them.

The polls were false. The authority they briefly acquired was borrowed from everyone who repeated it.

Different eras, same hoax

There is nothing new about manufacturing the appearance of authority. In 1770, Wolfgang von Kempelen unveiled the Mechanical Turk, an automaton that appeared to play—and defeat—human chess opponents. Hidden inside its elaborate cabinet was a human chess master. Audiences attributed intelligence to the visible machinery rather than the concealed person producing the moves.

The New York Sun’s Great Moon Hoax of 1835 counterfeited a different kind of authority. A series of articles claimed that the eminent astronomer Sir John Herschel had discovered lunar forests, bison and humanoids with bat-like wings. The stories borrowed scientific credentials, technical language and the authority of the Edinburgh Journal of Science, which had folded years earlier. The moon creatures were imaginary. The authority of science and scholarly publication was not.

AI has changed the economics of hoaxing. 

A century later, Orson Welles borrowed the conventions of breaking radio news to dramatize a Martian invasion. His 1938 War of the Worlds broadcast billed itself as fiction, but some late-arriving listeners mistook it for an emergency. Newspapers then exaggerated scattered confusion into a story of nationwide panic, helping create a second layer of amplification.

The Mechanical Turk counterfeited machine intelligence. The Moon Hoax counterfeited scientific provenance. The War of the Worlds borrowed the immediacy of broadcast news. Median counterfeited a research organization. 

These hoaxes share a basic approach to manufacturing credibility. Every era develops standards for what credible information looks like. The hoax succeeds by copying these signals, which audiences then use as shortcuts for credibility.

This creates an authority gap: the distance between the appearance of credibility and the proof that should support it. 

Show your work

AI has changed the economics of hoaxing. Earlier hoaxers usually needed scarce, costly infrastructure: an elaborate physical apparatus, a newspaper or a broadcast network. The Median hoax needed one person, AI and a website.

A single person can now create a company identity, website, methodology, dataset, report, executive commentary and apparent archive in hours. The organization can have researchers who never conducted research, customers who never bought anything and leaders who don’t exist.

In “Proof Beats Positioning,” I argued that markets increasingly discount polished claims unless they are backed by proprietary data, customer outcomes, operating insight or original research.

That argument still holds. But the Median hoax exposes its next complication. A chart can be fabricated. A methodology page can describe research that never occurred. Customer quotes can be invented, along with the customers themselves. Synthetic accounts can create the appearance of professional consensus.

Proof beats positioning only when the market can inspect the system that produced the proof.

The answer is not to abandon proof. It is to make the production of proof auditable.

Provenance is the first requirement: Who produced the evidence? But provenance alone is insufficient. A known company can commission a biased survey. A real executive can make an unsupported claim. An authentic document can contain obsolete information.

Method establishes how evidence was produced. Underlying data shows what supports it. Judgment explains what it means. Accountability identifies who will defend the interpretation. Calibration shows how that judgment has performed over time.

Who produced this? How? What supports the conclusion? Who interpreted it? What uncertainties remain? Who is accountable if it proves wrong? Has the organization earned reliance on its judgment?

Proof beats positioning only when the market can inspect the system that produced the proof.

Why companies should care

Most companies will never fabricate an election poll or announce the discovery of lunar bat people. They nevertheless operate in markets increasingly shaped by synthetic signals.

Buyers, investors and executives rely on research, reviews, media coverage, analyst commentary and social recommendations. AI systems ingest the same material to answer consequential questions: Which companies lead? What trends matter? What criteria should buyers apply?

Repetition can resemble confirmation. A weak claim appears on a company website, is restated in an AI-generated article and repeated on social media. An AI assistant later encounters several apparently independent references and presents the claim as consensus.

A falsehood doesn’t have to fool everyone. It only has to spread among people and institutions that assume someone else already checked it.

A falsehood doesn’t have to fool everyone. It only has to spread among people and institutions that assume someone else already checked it.

The same dynamic operates in markets. A less capable competitor can copy the visible markers of category leadership—a research franchise, benchmark, executive platform and confident account of where the market is heading.  Buyers can’t distinguish real expertise from a convincing imitation unless a company shows the proof behind its claims.

What organizations should do

Showing your work doesn’t mean dumping your data. It means making the chain between evidence and judgment visible enough to test.

Establish the right to know. Show why the organization is in a privileged position to make the claim. The proof could be proprietary data, repeated customer exposure, technical expertise or operating experience. 

Share the recipe. It’s always important to explain the methodology behind your analysis. For research, that may include sampling, field dates, questions, exclusions and partners. For customer evidence, connect claims to identifiable outcomes and conditions. For operating data, explain what was measured, over what period and with what limitations.

Who made the call? Evidence requires interpretation. Identify who reached the conclusion, explain what qualifies them to do so and make their reasoning available for scrutiny.

Preserve a canonical record. Give important claims an authoritative home. Maintain sources, dates, authorship, version history and revision status. Distinguish current findings from superseded material. This matters for human readers and AI systems that may otherwise treat an obsolete claim as authoritative.

Invite verification and disagreement. Independent partners, outside experts and transparent methods make evidence more credible. So does a willingness to publish inconvenient findings, acknowledge uncertainty and correct mistakes.

Build a record over time. A single report is easier to imitate than a continuing body of work. Recurring research, updated benchmarks and public corrections allow the market to judge whether an organization’s interpretations survive contact with reality.

Taken together, these practices constitute the infrastructure of authority. They make an organization harder to imitate because competitors can’t reproduce the experience, evidence, relationships and accumulated judgment beneath the visible work.

False fronts

A report, article or website is the street frontage of an argument. Behind genuine authority lies an intellectual city: experience, evidence, methods, disagreements, editorial decisions and accumulated judgment.

Hoaxes build the frontage without the city. That’s an ancient pattern, but AI has  changed its economics. The visible signs of proof can now be generated almost as  easily as positioning.

That doesn’t make proof less important. Instead, it makes the system behind the proof more valuable. In a market crowded with convincing façades, proof separates expertise from imitation.

How we use AI

Walled City uses AI as part of its editorial process—the same human-led approach we use to help clients build market authority at scale. Our essays draw on original reporting, data, research, experience, sustained revision, and human verification. People remain responsible for the ideas, judgment, accuracy, and final work.