Detailed · 9 events
A History of Search Engines
1990s
In under five years in the 1990s, web search passed through three methods. Yahoo! began as two Stanford students' bookmarks, sorted by hand into categories. AltaVista, a showcase for DEC's Alpha servers, indexed every word in the body of a page. In 1998 the founders of Google described the field as split between hand-kept directories and keyword engines whose results were too often poor, and ranked pages instead chiefly by the links pointing at them, alongside anchor text. When Google began supplying Yahoo!'s search results in 2000, the relationship between the directory and the ranking engine started to reverse.
Verizon Media / Pentagram (Wikimedia Commons) · Public domain (below threshold of originality; trademark applies) · Commons ↗ Jerry Yang and David Filo, doctoral students in electrical engineering at Stanford, published their personal bookmark list as 'Jerry and David's Guide to the World Wide Web' in early 1994 and renamed it 'Yahoo!' that March. The hierarchically organised, hand-curated directory became the first real map of a web that full-text search could not yet read. The company was incorporated in March 1995, went public in April 1996 at a market capitalisation of roughly US$848 million, and by 1998 was serving 95 million page views a day. Through the 2000s, Google's algorithmic ranking displaced it from the centre of the web, and Verizon's US$4.48 billion acquisition of its core business in 2017 ended Yahoo's life as an independent company.
AltaVista (Wikimedia Commons) · Public domain (below threshold of originality; trademark applies) · Commons ↗ Digital Equipment Corporation (DEC) launched AltaVista, a search engine developed jointly by its Western Research Lab and Palo Alto research arm. The multi-threaded crawler 'Scooter' paired with the 'TurboVista' back-end indexed roughly 16 million pages at launch, drew 300,000 hits on day one, and peaked at about 80 million hits per day in 1997. Unlike Yahoo!'s contemporaneous directory, AltaVista delivered full-text search over the entire page body — the first such service to do so at scale. It was also a showcase for DEC's 64-bit Alpha servers. Compaq acquired DEC in 1998, Overture bought AltaVista for US$140 million in 2003, Yahoo absorbed Overture later that year, and the service was finally shut down on 8 July 2013.
- EVT.003T1Google FoundedA History of the Internet and the WebA General History of Information TechnologyRead more →
2000s
By the end of the 2000s search had settled around one leader and one business model. In 2008 Google held roughly 65 percent of US searches, Yahoo about 20 and Microsoft about 8, and all three kept users' search histories to target advertising. That September Gabriel Weinberg, working alone from his home in Paoli, Pennsylvania, launched DuckDuckGo on the opposite premise: no logged IP addresses, no stored queries, the same results for everyone. It could not compete on size, but after the 2013 surveillance revelations its traffic tripled within a month, and the idea of an untracked engine carried into the next decade.
DuckDuckGo (Wikimedia Commons) · Public domain (below threshold of originality; trademark applies) · Commons ↗ Gabriel Weinberg launched DuckDuckGo from Paoli, Pennsylvania — a privacy-first search engine that does not log IP addresses, search histories, or personal profiles. Built solo from his home in early 2008, it had its commercial launch on 25 September 2008. Through the 2010s it established the position of 'the alternative to Google for users who object to tracking'. By the 2020s it was serving around 100 million queries a day, holding roughly 1.8% of the US search market and about 3 billion queries a month worldwide. It became a built-in choice in Apple Safari and Mozilla Firefox from 2014, and shipped its own privacy-first browser in 2018.
2010s
This timeline's 2010s turn on how a search engine reads the query. In October 2018 Google researchers released BERT, a language model pretrained to use the context on both sides of every word, and the weights went public a month later. By October 2019 Google was applying it to one in ten US English searches; by December it covered more than seventy languages, and in 2020 Google said almost every English query used it. A research paper had reached production in about a year. In the 2020s the language model inside search would start writing the answer itself.
2020s
In the 2020s search began moving from a list of links toward a written answer with its sources attached. A start-up showed the form first: Perplexity launched a week after ChatGPT, grounding an LLM's replies in live web results with footnoted citations. The incumbents followed. Microsoft built a chat interface into Bing, Google placed AI Overviews above its US results in 2024, and advice to put glue on pizza soon spread as an example of what could go wrong. OpenAI folded its own search prototype into ChatGPT. What answers on the page mean for publishers and advertising is still unsettled.
Perplexity AI / Alisperic (Wikimedia Commons) · Public domain (below threshold of originality; trademark applies) · Commons ↗ Perplexity AI — founded in San Francisco by Aravind Srinivas, Denis Yarats, Johnny Ho and Andy Konwinski — launched its answer-style search engine, combining an LLM with real-time web search. The launch came just seven days after ChatGPT's debut on 30 November 2022, and it presented, ahead of the incumbents, what would become the standard form of modern AI search: an LLM grounded in retrieved web content, with footnoted citations on every answer. The design — using web search to anchor LLM output against fabrication — was first deployed at scale by Perplexity. By early 2026 its Series E-6 round valued the company at about US$21.2 billion; monthly active users were around 45 million, and Perplexity handled 780 million queries in May 2025 alone.
- EVT.007T1New Bing and Microsoft Copilot — LLMs Enter SearchA History of Artificial IntelligenceA History of MicrosoftRead more →
- EVT.008T1AI Eats Search — Google AI Overviews and the Rise of PerplexityA History of the Internet and the WebRead more →
OpenAI (Wikimedia Commons) · Public domain (below threshold of originality; trademark applies) · Commons ↗ OpenAI unveiled SearchGPT as an explicitly 'temporary prototype' of AI search, opened through a waitlist to a small test group — and said in the same post that it planned to fold the best of it into ChatGPT, so it was never meant to ship as a standalone product. The interface paired LLM-written natural-language answers with real-time web results and in-line source links. On 31 October 2024 the feature launched as 'ChatGPT search' for Plus, Team and SearchGPT waitlist users; on 16 December 2024 it reached logged-in free users; and from 5 February 2025 chatgpt.com served it without a sign-in. It marked the point at which OpenAI entered the search market through its own user surface, opening a three-front contest against Google's incumbent dominance, Perplexity's specialist lead, and Microsoft Copilot's Bing-backed integration.