Nearly a Decade Later, the Autocomplete Origin Story: Kevin Gibbs and Google Suggest
If you head over to the search field on Google, Amazon, YouTube, Facebook, Twitter or any of the world’s other big websites and start writing, a little box will appear below. As you type each letter, the site will make its best guesses as to the words you’re trying to find.
For instance, I just put the query “beach” into Google.com from San Francisco. Starting with “B” — Bank of America? “Be” — Best Buy? “Bea” — bearfacts? (a U.C. Berkeley site, apparently).
“Beac” — beacon? Beach Blanket Babylon? beach? Yup, that’s it.
Sometimes the search engine hits on what you want right away. Other times, despite typos, it magically understands what you meant to look for. Sometimes the results are existential and funny and reveal odd inclinations. (Try “Google is …” or “I hate …”)
This particular feature is known as “typeahead” or “autocomplete.” It’s based on analysis of the enormous dataset of everything everyone else in the world has searched for, narrowed down using personalization and location filters. Basically, Google thinks you’re likely to be looking for something that lots of people, especially people like you, have looked for before.
The more common a search is, the more it’s judged to be likely that you’re doing it, too. Except, of course, if it’s for porn. For that, you’ll have to click through to see the actual results. Google’s not going to presume it for you.
This little feature originated on Google nearly nine years ago, sprung from the mind of a junior software engineer. He wanted to call it “Google Complete,” but then-exec Marissa Mayer stepped in and suggested the more accessible “Google Suggest,” which stuck.
Google Suggest launched in December 2004. It’s a geeky thing, and the human race would survive just fine without it — but I’d argue that it’s one of the more influential online user interface changes of the past decade. So, over the past couple of weeks, I set out to find out a bit more about how it came to be.
Autocomplete, Saffer said in an interview this week, is part of the microinteraction canon, along with things like Twitter’s “pull to refresh,” which has become part of so many smartphone apps.
“What started as a signature moment for Google became the de facto standard,” Saffer explained.
Saffer described two signs that a microinteraction has gotten it right: 1) You wonder why things that don’t have it are broken, and 2) You only notice it when it messes up.
A Side Project Is Born
Google Suggest was built by Kevin Gibbs, a recent Stanford grad who joined Google just a couple of months before it went public. Having spent a few years at IBM, Gibbs was drawn by the lure of big projects and “20 percent time,” as well as a new program that would shuttle workers from San Francisco to Mountain View headquarters (today, ubiquitous Google Buses are the bane of anti-gentrification San Franciscans, but at the time there was just one trip a day from the Glen Park transit station).
When he started at Google, Gibbs’s role was to work on the systems infrastructure that helped run Google’s data centers, so there wasn’t much of his day job he could do from a laptop on the shuttle, with its crappy Edge Internet connection, Gibbs said in a recent interview.
So, with that extra offline time, rather than taking a nap, he thought it would be fun to work on something that combined some of the hot new geeky developer stuff of the time:
- Big Data. Because he worked at Google, Gibbs could play with billions of Web documents and use thousands of computers to process them. This was before Amazon Web Services came out, and before that term meant much of anything to anybody.
- High-speed Internet. By 2004, more than half of American Internet users had access to broadband at home or work. That made it possible to build something that was much more complex and still potentially accessible to a big audience.
Gibbs said he doesn’t spend a lot of time thinking about or taking credit for Google Suggest these days — he’s now at Quip, the new mobile productivity startup he co-founded with former Facebook CTO Bret Taylor.
“I don’t feel when I look at a search box that it’s something I did,” Gibbs said of Google Suggest. “It feels like this is just how the world’s supposed to work. I don’t feel any personal attachment to it unless I stop to think about it.”
The technological timing was right, Gibbs said. “I’m sure it would have happened if I hadn’t done it. I think it’s one of those history of invention things — where there was one guy who developed it in Germany and one guy in Russia, and it turns out they were doing it in the same year. I haven’t found my guy, but I think it was just an idea that was just so ripe to have happened.”
“That’s Cool, What If You Did It for Search?”
Back in 2004, the first thing Gibbs built on his shuttle trips was a URL predictor. So, as you started typing a URL into a browser, it would autocomplete the remaining options by analyzing Google’s immense corpus of Web content. Kind of like how, in Outlook, when you started typing an email, the names of the people in your address book would pop up — but on a much larger scale.
A co-worker — Gibbs can’t now recall who it was — looked over and said, “That’s cool. What if you did it for search?”
So Gibbs redid the system. It clicked right away. Search leaders at Google, including Jeff Dean and Rob Pike, got wind of the feature and started promoting Gibbs’s work internally. Mayer contributed the name “Google Suggest.”
But before the product could launch, Gibbs had to personally create the blacklist of words that wouldn’t appear on Google Suggest.
Since the feature would be actively showing results before someone had finished a query, there was a huge risk of Google putting forth something that offended people — even if it was the most likely result algorithmically.
That meant many hard and strange decisions, Gibbs recalled, about terms like “hooters,” which could mean owls or the restaurant or boobs; and “lesbian sex,” which on its own is descriptive, but when followed by words like “video” is perhaps not appropriate for all eyes. (In the second case, he didn’t remove the root search, but blocked some derivations.) For many letters of the alphabet, the most commonly searched word was something related to porn.
Gibbs’s blacklist was clunky and imperfect. He worried that he would be shaping people’s behavior, since “taking something out of results implies it doesn’t exist.” There was also the creepiness factor of Google appearing to reach into people’s private thoughts and history and anticipate what they were thinking — one of the first of so many such examples over the past decade.
But in December 2004, after a short period of internal testing, Google Suggest launched as a Google Labs feature.
Gibbs wrote the brief blog post introducing it to the world:
Today we launched Google Suggest, a new Labs project that provides you with search suggestions, in real time, while you type. We’ve found that Google Suggest not only makes it easier to type in your favorite searches (let’s face it — we’re all a little lazy), but also gives you a playground to explore what others are searching about, and learn about things you haven’t dreamt of. Go ahead, give it a spin.
The project stemmed from an idea I had a few months ago, and since then I’ve been working on it in my 20% time, which is a program where Google allows their employees to devote 20% of their working hours to any project they choose. What’s really amazed me about this project is how in a matter of months, working on my own, I was able to go from a lunch table conversation to launching a new service. In my opinion, this is one of the things that really makes Google a great place; that the company’s systems, resources and, most important, people are all aligned to make it as easy as possible to take an idea and turn it into something cool.
Plus, we have Segways.
Aside from the notion of Segways being cool, the funny thing is that, after the initial launch, Google Suggest didn’t fully launch for nearly four years.
The feature languished in opt-in mode in Google Labs until it was finally made default on Google.com as well as mobile, maps and browsers in 2008.
Gibbs admitted that was partly because he lost interest in the project, and wasn’t willing to neglect his day job to add new features to it. “Once I launched it, I wasn’t really that interested in continuing to do it,” he recalled.
But, since 2008, Google has embraced Suggest wholeheartedly, to the point that it’s not even a feature anymore; it just happens with every search. Sites like Facebook soon followed suit. In 2010, Google expanded the feature to start automatically refreshing the entire search-results page, not just the search query, with Google Instant.
Podunk Towns Have Searches, Too
Reflecting back on that time now, Gibbs says he’s most proud of two aspects of Google Suggest. First is the democratic nature of the feature. Google Suggest wasn’t just about guessing that people who typed in “b” were looking for Britney Spears (as they invariably were at the time, as she was the top search term for years). It was about the long tail, too.
In fact, that was a bit of a hard sell, Gibbs recalled. “It took hundreds of machines to store everything that it might return to you. And I think a lot of people thought it should just have the most common searches.
“But what I really liked was that I could put in just the name of just my very small hometown in Central California — Porterville — and even at that time, it would show all the most popular results just for that town. It made me realize that within your community, whatever you care about, wherever you happen to be on the Internet, that world is still a big, enriched world that could otherwise be left behind.”
Second, Gibbs said, Google Suggest “respected someone’s time.” If it was already obvious to Google what someone was searching for, why not just show it?
Of course, that’s not just altruistic. Creating a website that was more responsive meant that people used it more, and thus saw more and better Google ads.
But that wasn’t Gibbs’s concern. He was back to working on data centers, and then starting Google’s cloud infrastructure tools for developers, in order to compete with Amazon Web Services. Gibbs finally left Google in July 2012, to start Quip.