Autocomplete Completion
Autocomplete completion is the term for the feature that turns a few typed characters into a finished suggestion — for instance in the search bar, in a messaging field, or in a code editor. Modern variants do this by estimating which continuation is most likely to follow what has been typed so far.
You type “Weather Ber” into a search field, and below it already appears “Weather Berlin tomorrow.” That’s exactly what autocomplete completion is: a program guesses what you meant to write and offers it to you to accept. You can accept the suggestion, usually with the Tab or Enter key, or simply ignore it and keep typing. So the suggestion is never a decision, only ever an offer. The feature exists in very simple forms, such as a list of previous entries. And it exists in very elaborate forms that complete entire sentences or several lines of program code.
Why a few saved keystrokes are worth so much money
Autocomplete seems like a small thing, but it has enormous economic leverage. For search engines, the suggestion list steers a large share of queries. Whoever appears at the top gets attention without having to buy advertising. That’s why the order of suggestions is a fiercely contested field and regularly keeps courts busy as well.
In software development the effect is even more directly measurable. Programming tools like GitHub Copilot complete lines of code, sometimes entire functions. The manufacturer advertises that a noticeable share of newly written code comes from such suggestions. For companies this is a productivity promise, for providers it’s a subscription business with monthly fees per user.
But there is also a downside. Anyone who constantly accepts suggestions checks less carefully. In programming, this can let faulty or insecure code slip into a project. In search fields, suggestions can reinforce prejudices, because they simply reflect what many people have typed before.
From dictionary to probability calculation
The simplest construction is a list. The program stores all previous entries and looks for those that begin with the typed letters. For this, a tree-like data structure is used, in which shared word beginnings are stored only once. This is fast and economical, but it can’t suggest anything it hasn’t seen before.
The modern variant works with a language model. This is a program that has learned from huge amounts of text which words typically follow one another. It breaks your input down into small building blocks, so-called tokens, and calculates a probability for each possible next building block. The most probable one gets appended, then the calculation starts over from the beginning. This is how a suggestion emerges, word by word.
A common misconception is that the model looks something up somewhere. It doesn’t. It only estimates what fits well — which is why suggestions sometimes sound convincing and are nevertheless wrong. Good systems combine both: exact matches from a database for names and facts, the model for the fluent rest.
From smartphone keyboard suggestions to news about Copilot
You encounter autocomplete every day in everyday life, usually without thinking about it. Your smartphone’s keyboard suggests words. The email application completes addresses and entire polite phrases. The browser completes web addresses, the online form your street after two letters.
In business news, the term mainly comes up in two contexts. First, with programming tools: companies like Microsoft, Google, and smaller providers like Cursor compete for developers, and autocomplete is the core product there. Second, in data protection, since every completion requires that someone processes your input — often on a foreign server.
It’s important to distinguish this from a chatbot. A chatbot answers a question, autocomplete continues your own text. Technically, the same kind of model often lies behind both today, but the usage differs significantly. Autocomplete stays in the background and is successful when you barely notice it.