Google came out in stages, and the first public version went live in September 1998 on a server at Stanford University. The company was incorporated on September 4, 1998, but the search engine itself had been running on Stanford's network since at least 1996 under the name BackRub. So the honest answer depends on what you mean by "Google": a research project (1996), a public search engine (1998), or an actual for-profit company (1998–1999). This explainer separates those three dates with primary-source precision.

The Stanford mechanism: how BackRub actually worked

Google did not begin as a search engine in the modern sense. It began as a ranking algorithm, and the algorithm is the reason the dates matter. Larry Page and Sergey Brin built BackRub at Stanford around 1996 while working on a digital libraries project. The core insight was that the web's link structure is itself a signal of importance.

The mechanism was PageRank, described in their 1998 paper "The Anatomy of a Large-Scale Hypertextual Web Search Engine." The idea: a page is important if important pages link to it, and importance propagates along links. Each link is treated as a vote, but votes are weighted by the voter's own PageRank. That definition is recursive, so the system solves for a stable score across the whole graph rather than scoring pages one at a time.

Two design choices made this practical. First, PageRank is computed offline, in advance, not per query. Second, the index stored the full text of pages compressed, so a query could be matched against a compact structure and ranked quickly. A plain-language version: Google pre-sorted the web by link popularity so a search could return results almost instantly instead of re-scanning everything.

This is why the "when" is fuzzy. BackRub was a working system on Stanford servers years before any commercial launch, and the 1998 paper describes an architecture that already existed. The launch date, not the invention date, is what most timelines report.

The quantified timeline: five dates that all deserve a place

Here is the reconciliation, in one place, with the distinction that each row actually establishes.

Date Event What it establishes What it does not establish
~1996 BackRub runs on Stanford servers Working prototype exists Public availability
Sept 15, 1997 google.com domain registered Name exists Live public engine
Sept 4, 1998 Google Inc. incorporated Legal company exists Product launch
Sept 1998 Public search engine indexed about 25 million pages Public launch Widespread adoption
1998–1999 Office at 232 Santa Margarita Ave, Menlo Park Company operating Market dominance
1999 First outside funding round Commercial scaling Product maturity
2004 Initial public offering Public-company status Search quality peak

The key reconciliation: the incorporation date (September 4, 1998) is the date Google's own history uses as its founding, but the public engine began serving queries in September 1998 as well. Those two events are different in kind, not merely in timing. A company can be legally formed one day and launch a product the next; only one of those is the "release."

A counterintuitive detail worth stating plainly: the 1998 index of roughly 25 million pages was not obviously the largest index at the time. Competing engines had indexed more pages in some cases. Google's edge was ranking quality, not raw coverage. That distinction matters because "came out" often implies "arrived with a complete product," and Google arrived with a better sort than a bigger pile.

Edge cases and the dates that confuse people

Several specific dates get conflated, and each has a distinct meaning.

The domain google.com was registered on September 15, 1997, not 1998. A registered domain is a name reservation, not a live service. The name "Google" itself is a misspelling of "googol," the number 10^100, chosen to signal the scale of the index.

The rewrite of the early search engine, described in the 1998 paper, was a Stanford research artifact before it was a commercial product. The paper names Sergey Brin and Lawrence Page and describes a system tuned for a campus-scale web. That is the closest thing to a "release note" for the original engine.

The company address in Menlo Park, 232 Santa Margarita Ave, was a garage adjacent to the Stanford campus, according to Google's own corporate history. This is often cited as "where Google started," which is accurate for the company, not for the software.

A practical caveat: archived web pages from 1998 can be incomplete. If you are verifying a launch claim, prefer a primary source (the 1998 paper, the domain registry record, or the incorporation filing) over a secondary timeline. Secondary timelines routinely list 1998 without specifying which event they mean.

Misconceptions that survive to this day

Misconception one: "Google launched in 1998 fully formed." The 1998 engine was a research-derived system with a modest index, and the behaviors that made it dominant accumulated over years. The launch was a starting line.

Misconception two: "Google started as a search engine." It started as a ranking method applied to a crawl. The search interface came with the packaging, not before it.

Misconception three: "BackRub was just a prototype name." It was the system that held PageRank, and PageRank is the reason Google's early results sorted differently from competitors. The name changed; the mechanism did not.

Misconception four: "The founding date is the same as the launch date." They are close in time but distinct in kind, and conflating them is the most common error in 900-word explainers. A company can incorporate before it ships, or after.

A professional caveat: search engines are evaluated by retrieval quality at a point in time, and early Google's advantage was measurable but not infinite. Without access to the original logs, any claim about how much better it was than specific competitors on a specific date is not verifiable, and should be stated as a qualitative claim.

Practical actions for verifying the date yourself

If you need a defensible answer, follow a source hierarchy rather than a single date.

Start with the 1998 conference paper by Brin and Page for what the system did and when. Check the domain registration record for google.com for the name date. Use the state incorporation filing for the company date. Treat 1998 as the public-launch year and state the month, not just the year. Re-check the precise day only if the claim depends on it, because some sources round to "September 1998." Distinguish "public engine" from "commercial company" in any table or citation.

The threshold for a confident answer: if you can name the event (incorporation, domain, launch, clustering of index pages), you can attach the right date. If you cannot, "September 1998" is the honest floor.

Limitations and the "ahead of the engine" problem

The deepest limitation is that the modern web has no complete archive of 1998. The Wayback Machine began crawling later, and any reconstruction of what Google indexed on launch day is an inference from the paper plus stored crawls, not a logged snapshot. That means edge-case questions — "what was the first query," "which competitor it beat first" — are largely unanswerable with certainty.

A second limit is definitional. "Came out" is not a technical term, so any timeline is only as good as its stated definition. Google did not come out once; it came out in layers, and each layer has its own date. That is not hedging; it is the actual structure of the facts.

There is also a professional caveat about single-number answers. Reducing a multi-year development to one date is convenient, but it hides the mechanism that made the launch worth remembering. The reason September 1998 keeps appearing in rankings is not the date itself, it is that the event on that date shipped a new way to sort the web.

Reconcile the thresholds explicitly: the founding date is September 4, 1998, the public-launch month is September 1998, and the domain predates both by about a year. Use the date that matches the question, and say which one you are using.

This is general educational information about a technology timeline, not a legal or financial opinion. For matters that depend on the exact incorporation record or contract terms, consult a qualified local professional.