Online search powered by the Voice of the People

Online reviews and information about products, businesses, and services strongly influence the purchasing behavior of customers. According to Klarna’s Global Shopping Pulse Report, 83% of shoppers research products online before visiting a store, and 72% occasionally use their smartphones for in-store research. A meaningful share of what they encounter can be false or manipulated. Industry estimates place the share of online reviews that are fake or inauthentic at around 30%, and on major platforms as many as 43% of reviews have been flagged as suspicious. In 2021, an EU-wide sweep of online consumer reviews found that at least 55% of online businesses violated the Unfair Commercial Practices Directive, which required truthful information to be provided to customers to allow them to make an informed decision.

The concern is that the problem is getting harder, not easier. Generative AI has sharply reduced the cost of producing convincing synthetic reviews at scale. Large online platforms are not ignoring the problem — Amazon has reported spending hundreds of millions of dollars and assigning thousands of staff to it in a single year, blocking or removing hundreds of millions of fake reviews; Google reports removals on a similar scale. Regulators have begun to move as well. In December 2025, the U.S. Federal Trade Commission issued its first warning letters against fake reviews under the Consumer Review Rule, which carries civil penalties for each violation.

Look closely at those numbers and a pattern appears. A large share of the effort is being spent on detection — on classifiers that try to tell a fake review from a real one after both have been written. That creates a difficult arms race because generative tools can lower the cost of producing convincing synthetic content while detection systems must continually adapt. The problem is not that filters are useless. It is that filtering alone does not establish whether a review is anchored to a real commercial event.

The problem of fake and manipulated information is unlikely to be solved by detection alone. Three structural limitations matter:

• Reviews are usually confined to the platform that collected them. A typical customer buys from many vendors, but feedback remains scattered across separate merchants and service providers. That makes it difficult to evaluate a reviewer’s history across participating markets rather than only within a single storefront.
• Many reviews are weakly linked to the event they describe. A review may be written against a listing without carrying a reusable record of what was bought, from which merchant, or when. Some platforms verify purchases inside their own systems, but that verification generally does not travel with the review across platforms.
• One of the strongest behavioral signals is largely fragmented. A customer transaction record (CTR) says what a person actually did with their money. Repeatedly buying the same product can indicate preference; switching away can indicate the opposite. These signals are not written reviews, but they reflect completed purchases. Today there is no widely adopted cross-merchant infrastructure that lets customers retain item-level transaction records pseudonymously and reuse those signals under their control.

How search engines build their index

Search engines build their index through a multi-step process involving crawling, processing, and organizing web content. Automated crawlers systematically browse the web to discover new or updated pages. Once a page is crawled, the search engine analyzes its content to understand its meaning, structure, and context, and adds the processed information to its index. Ranking algorithms then consider factors such as content quality, user engagement, backlinks, and freshness to determine which indexed pages are most relevant and authoritative for a particular query. When a user searches, the engine retrieves and ranks the relevant entries and presents them as search results. The index therefore provides an organized representation of what has been published on the web; it does not ordinarily establish whether the commercial events described in that content actually occurred. That limitation can be addressed by incorporating records of completed transactions alongside what people say about their experiences.

Building a search index from online reviews and Customer Transaction Records

Online reviews provide an explicit account of customer experience, while Customer Transaction Records provide evidence of the underlying commercial activity. In the blogs Token-based Loyalty Program and Customer Owned Transaction Records, we described how a transaction record can be filed against a token rather than a name. The record establishes what was bought, from whom, and when, without establishing who bought it. A review solicited against that record therefore has a referent: a specific item, from a specific merchant, on a specific date.

It would be easy to draw the wrong conclusion from that and say: let only verified purchasers write reviews. Retail and travel platforms already do it, inside their own walls. It is not what our system does, and it should not be. It shuts out the person who was given the product, borrowed it, or bought it somewhere that does not participate — and a purchase is not, on its own, evidence that a reviewer has anything worth reading. The platform therefore accepts unsolicited reviews with no transaction behind them.

What the transaction record supplies is not credibility. It supplies a referent and an occasion. Because the PCX Platform knows which item was bought, it can put a contextually relevant feedback form in front of the customer — built from that item’s own attributes — rather than a blank box and five stars. An unsolicited review is handled the same way by a different route: the reviewer identifies the subject, by phone number in the case of a business, and the platform builds the form from what its directory says that business does, so a hair salon is reviewed on the services a hair salon provides. Credibility is a separate problem, and it is handled separately: by the standing a reviewer earns inside the review system, by the balance of endorsements and rejections a review attracts, and by the promotion of a demonstrated expert — the Voice of the Expert — above a first-time contributor.

Imagine a customer named Bob who purchases goods and services from multiple vendors over a three-year period. To the platform, Bob is represented by a pseudonymous identifier rather than by his name. If the same product is bought repeatedly in the presence of alternatives, the record can reveal a preference that a written review may not. If one restaurant is consistently abandoned in favor of another serving similar food, the record may reveal a behavioral choice. These are signals nobody had to write, and they are grounded in completed transactions.

Read one at a time, these signals can be noisy. Read across a sufficiently large market and reported at cohort level, they can provide a strong behavioral account of what people value. The two kinds of signal answer different questions: explicit reviews record what customers say, while implicit signals — the purchases themselves — record what they did.

An index is not yet a search engine. But a search engine built on verified transaction signals could rank on evidence that conventional web search does not normally possess. It could ask what people actually bought, what they bought again, and what they quietly walked away from — making rankings harder to manipulate through content, backlinks, advertising, or unverified reviews alone.

The wisdom of the crowd

The wisdom of the crowd is the idea that, under the right conditions, aggregated judgments from large groups can outperform individual judgments in some problems involving estimation, decision-making, and prediction.

James Surowiecki popularized this idea in his 2004 book, The Wisdom of Crowds, which examines how large groups have made superior decisions in pop culture, psychology, biology, behavioral economics, and other fields. A market’s transaction record, read pseudonymously and reported at cohort level, is the Voice of the People (VOP). It is a crowd that never has to be polled, because it has already spoken with its money. A search engine that draws its rankings from that crowd is what we mean by online search powered by the Voice of the People.

The Voice of the People describes what verified transaction records could one day support. It does not describe anything that exists today. This is a long-term ambition, not a product on a roadmap. ValiDeck’s near-term work is infrastructure: the protocol, the platform, and the transaction records themselves. But it is worth being clear about where that infrastructure leads. A search index grounded in what people actually bought, and bought again, is not a better filter for a broken system. It is a different system — one in which the loudest voice does not win, and the truest one does.