20 Years of Online Reviews: From Amazon 1995 to AI-Era 2026

The internet in 1995 was a peculiar place. Mosaic had recently given way to Netscape. Most American households were still on dial-up. And Jeff Bezos, running a fledgling online bookstore from a garage in Bellevue, Washington, made a decision that puzzled much of his own staff: he would let customers post reviews of the products he was trying to sell.

The first Amazon customer review appeared on the site in 1995. It was for a Douglas Hofstadter book, "Fluid Concepts and Creative Analogies," and it was not universally enthusiastic. Publishers were alarmed. Why would a retailer broadcast negative opinions about their products? The conventional wisdom of 1995 said you controlled the message. You did not hand the megaphone to the crowd.

Bezos understood something the publishers didn't. Reviews were not marketing — they were infrastructure. Trust, once built, compounds. A buyer who knows the negative reviews are real trusts the positive ones far more than any ad copy. Amazon's reviews were not a feature. They were the product.

The first dedicated review platforms

By 1999, the idea had propagated. Three pioneering platforms launched within months of each other: RateItAll, Deja.com, and Epinions. Together, they generated over one million reviews in their first year. Epinions, founded in June 1999, was architecturally interesting — it paid reviewers a tiny cut of the affiliate revenue their reviews generated, creating an early "web of trust" that weighted reviews by the reputation of the reviewer.

TripAdvisor arrived in February 2000, initially aggregating professional travel content from newspapers and guidebooks. The 'add your own review' button came later — almost as an afterthought — but it would redefine the platform entirely. These were the years that proved the concept: real people, real opinions, real impact on purchasing decisions.

In 2001, something important happened on the periphery: Yellow Pages and CitySearch added review capabilities to their business directories. For the first time, you could not only find a plumber but read what his last three customers thought of him. Local service businesses would never be the same.

But 2004 is the year local reviews truly arrived. Jeremy Stoppelman caught a bad flu in San Francisco and couldn't find any reviews for local doctors. He and Russel Simmons — both former PayPal engineers — founded Yelp that same year to solve exactly that problem. Within five years, Yelp had collected over 4.5 million reviews. By 2007, the site was receiving 4 million unique visitors per month.

Yelp was different from Amazon reviews in a crucial way: it was about *places*, not products. The reviews had stakes beyond a book purchase. A two-star Yelp review could close a restaurant. A five-star surge could make one. Businesses that had never thought much about customer feedback suddenly found their entire livelihood indexed and visible to every smartphone-owning passerby.

The economics of a five-star rating

The Harvard Business School professor Michael Luca published research showing that a one-star increase in a Yelp rating leads to a 5–9% increase in restaurant revenue. That number, replicated across dozens of later studies, crystallized something for business owners: the review score was not a vanity metric. It was a direct lever on revenue.

Google Maps launched in February 2005. Reviews on Maps came in 2007 — quietly, almost as a footnote. But the combination of satellite imagery, walking directions, business listings, and user reviews created something unprecedented: a real-time, crowd-sourced map of human judgment about every business on earth.

In April 2010, Google launched Google Places, which absorbed the Local Business Center and put reviews front-and-center in local search results. A business without Google reviews was now essentially invisible to anyone searching for it. The 'local pack' — those three businesses Google highlights above the organic results — was built almost entirely on review signals.

Meanwhile, Facebook launched business Pages in 2007, and by 2009 was nudging its 350 million users to rate businesses they'd visited. The review was no longer something you sought out. It was something that found you — in your feed, on your map, in your search results.

Why TripAdvisor became impossible to ignore

TripAdvisor had by 2010 accumulated over 40 million reviews and 20 million unique monthly visitors. When it launched the Certificate of Excellence in 2011 — awarded only to the top 10% of listed businesses — hoteliers and restaurateurs suddenly had a new benchmark: not just 'are our reviews good' but 'are they in the top decile.' The gamification of reputation had begun.

The iPhone arrived in 2007. Android followed. But it was between 2011 and 2013 that smartphone penetration crossed a tipping point in developed markets — and it changed the review ecosystem permanently.

Before the smartphone, you researched a restaurant at home and then went to dinner. After the smartphone, you researched it standing on the pavement outside. You left a review on the way home on the bus. The friction between experience and review collapsed to almost nothing. This was revolutionary in two directions: it dramatically increased review volume (more reviews, fresher reviews) and it dramatically increased the vulnerability of businesses to real-time reputation shifts.

Google Maps for mobile, reintroduced as a standalone app in December 2012, was downloaded over ten million times in its first two days. By 2013, it was the world's most popular smartphone app — used by over 54% of global smartphone owners. The review prompt, tucked at the end of a Maps navigation session, was reaching hundreds of millions of people in moments of maximum relevance.

Facebook Reviews and the social trust layer

In 2013, Facebook introduced dedicated review features for business Pages, adding a new dimension to the review ecosystem: social proof with identity attached. A Facebook review came from a real person with a real name and a profile photo — unlike anonymous Yelp reviews, which platforms like Yelp Elite were already being accused of curating unfairly.

The mid-2010s saw a proliferation of vertical review platforms: Healthgrades and Vitals for doctors; Houzz for home improvement; G2 and Capterra for B2B software; Glassdoor for employers. The review had colonized every domain of professional judgment.

The problem with a system built on trust is that it becomes a target the moment the stakes are high enough. By 2016, the market for fake reviews was a small but thriving underground industry. Services on Craigslist, later on dark-web forums, offered five-star Google and Yelp reviews for $10 apiece. Amazon, which had accumulated hundreds of millions of reviews, was discovered to have entire product categories — especially electronics and dietary supplements — where review manipulation was near-universal.

The FTC had been circling this space for years. It filed formal complaints against companies for fake-review conduct in 2019 and 2020. In New York, the state Attorney General ran 'Operation Clean Turf' — an undercover sting that created a fake frozen yogurt shop in Brooklyn to catch review manipulation services in the act. Nineteen companies were charged and $350,000 in fines collected. It was a signal, even if the magnitude was small compared to the scale of the problem.

Yelp, for its part, admitted that approximately 25% of the reviews submitted to its platform are never published — caught by its fraud detection algorithms before they can influence business rankings. Google was quietly building ML classifiers to detect sudden review spikes, reviewer pattern anomalies, and coordinated inauthentic behavior. The platforms were fighting the fake review problem — just not winning.

The paradox of platform incentives

There was an uncomfortable structural tension at the core of the fake review problem. Review platforms needed reviews to be trusted in order to have value. But they also monetized the businesses whose rankings depended on those reviews. The result was a perverse equilibrium: platforms were moderately aggressive against fake reviews — enough to maintain surface-level credibility — but not so aggressive that they alienated the business customers who paid for advertising and premium placement.

ChatGPT launched on November 30, 2022. Within weeks, it had become obvious to anyone in the review industry what was coming. A service that could generate a convincing 5-star restaurant review in four seconds — one that passed spelling, grammar, and sentiment analysis with flying colors — had just been handed to anyone with an internet connection.

By mid-2023, the scale of AI-generated review fraud had expanded dramatically. Google responded with its most aggressive enforcement action to date: 170 million fake reviews removed from Maps and Search in 2023 alone — a 45% increase over the previous year. The detection mechanism was a new machine-learning algorithm that analyzed longer-term behavioral signals: did a reviewer leave identical reviews on multiple businesses? Did a business receive a sudden spike of five-star reviews the week after launching an ad campaign? These patterns had always been suspicious; now they were flagged automatically at scale.

The FTC finalized its Trade Regulation Rule on fake reviews and testimonials in August 2024, enabling civil penalties up to $50,000 per violation for companies that knowingly purchase, deploy, or incentivize fake reviews. The rule was the most significant legal intervention in the review space in decades — arriving just as the technology to generate fake reviews had become trivially cheap.

The detection arms race of 2024–2026

The detection problem is, in a technical sense, unsolvable — or at least unsolvable with current methods. Researchers studying AI detection note that it is inherently adversarial: as detection improves, generation techniques evolve to evade it. Studies show that consumers rate AI-generated reviews as significantly less trustworthy and useful when they know a review was AI-written — but without disclosure, the gap in perceived authenticity is narrow.

The most serious platforms have responded with identity verification requirements, behavioral biometrics, and purchase verification gates — reviews only accepted when matched to a confirmed transaction. Google requires a Maps account with activity history. Amazon's Verified Purchase badge has become a minimum credibility threshold. The arms race continues. Neither side is winning decisively.