Skip to main content

Slack introduces a new search feature powered by artificial intelligence

Slack introduces a new search feature powered by artificial intelligence


Can it really help you find information faster?

Share this story


Slack is rolling out an improved search experience today bolstered by sophisticated machine learning, the company said. The feature, which was built by the company’s year-old search learning and intelligence group, is designed to help users find relevant channels and subject matter experts more quickly than traditional search. It represents a move by the company to extend its lead in team collaboration software as stiff new challenges emerge from giants including Microsoft, Facebook, and Google.

The new Slack search, which will be available only on paid accounts, attempts to highlight the most relevant people and channels for your query. Search for “hiring process,” for example, and Slack will try to show you the person who discusses that phrase the most, and in what channel. “It’s really about tapping into that collective knowledge from your company,” says Noah Weiss, who runs Slack’s search efforts, said that.

“It’s really about tapping into that collective knowledge from your company.”

Part of Slack’s pitch from the beginning has been that it will let the average user figure out what’s going on inside their own company more easily. The “S” in Slack stands for search — the company’s name is an acronym for “searchable log of all conversation and knowledge.” But until recently, Slack’s actual search features have been rudimentary. Last year it added a “top results” feature to searches that attempts to show you more relevant results. But I’ve never found it to be much use — if I query “Uber” in the Verge Slack, for example, the most recent result is from April 7th, even though we write about the company many times each week.

It’s for that reason that we should approach Slack’s claims of building “AI search” with skepticism. Weiss, who used to work on the Knowledge Graph at Google, says his team’s work is just beginning. But he shed some light on how Slack tries to guess at what makes a message relevant.

If a person frequently responds to questions about a certain word or phrase, for example, that suggests expertise. If a person’s messages about a keyword receive a lot of replies, or reaction emojis, or trigger new threads, that also suggests relevance. The speed with which that person replies could also be taken into account, Weiss said. Add in the channel where those conversations are taking place and Slack has a decent chance of pointing you in the right direction, he said.

In early tests, the machine learning-powered results led to clicks 30 percent of the time — much higher than the average for a standard set of search results presented in reverse chronological order, he said. It’s a good signal that Slack is onto something here, even if it doesn’t work all of the time. The Verge wasn’t able to test it ahead of time, so I can’t yet vouch for it.

Still, the problem Slack is targeting here is real. The company points to a recent study by market research firm IDC that said modern office workers spend 16 percent of their time looking for information about their own companies, and find it barely half the time. If Slack can offer a better way for employees to answer questions about their employers, it will represent a significant step toward achieving its product vision. And with Microsoft, Facebook, and Google circling it, the company likely can’t afford not to.