AI Briefing: How Perplexity plans to win over enterprise and regular users with AI search

Starting a new search business has always been a nearly impossible hill to climb, since one is invariably up against Mt. Everest — aka Google. However, Perplexity aims to bring its new momentum from generative search into the enterprise world of AI even while exploring ads further down its timeline.

Last week, Perplexity announced its entrance into the worlds of enterprise tech. Flush with new funding, the AI-driven search startup hopes to build a broader business case with a new offering called Enterprise Pro. Adding to both the free and subscription versions of its generative AI search platform, Perplexity already has a handful of noteworthy names using Pro including media agency the IPG-owned media agency Universal McCann. Other early clients across numerous industries including Zoom, HP, Stripe and the NBA’s Cleveland Cavaliers.

By improving how companies search documents — and everything else on the internet — companies are using generative AI to conduct market research, find sales prospects, analyze sales, and draft product road maps. Advertising agencies and media companies already use search to research campaigns, companies and people.

There are a few reasons Perplexity is approaching enterprise as a part of its overall user strategy. In an interview with Digiday, chief business officer Dmitry Shevelenko said the strategy is not only a good revenue generator, but also a growth channel by letting people try Perplexity at work while finding value outside of work. However, Perplexity prioritizes not being bogged down by ad-driven SEO.

“Google’s business model is not getting you answers, it’s getting you to click on as many links as possible,” Shevelenko said. “And the reason they’re able to offer it for free is because they monetize your data and your attention.”

That doesn’t mean Perplexity is avoiding ads altogether. While there’s no set timeline in place, Shevelenko said it likely won’t be for “at least one or two quarters” at the earliest. When it does add them, he said ads won’t appear above answers and won’t allow advertisers to skew search results by paying for placements. Instead, they might appear in the “related questions,” Shevelenko said, noting 40% of all Perplexity user queries have a follow-up question.

Follow-ups could be a “rich surface area for brands to get to point a consumer in their direction,” Shevelenko said, adding that sometimes most important thing for brands and users is asking the right question.

“Let’s say you’re researching a product, you get an answer about it,” Shevelenko said. “Then in the follow up questions, obviously be clear that it’s sponsored — but there will be a question about specific brands’ products and its features. And that the answer would be the exact same answer as if you would just organically popped in that question.”

Ad measurement might also look different for Perplexity. Shevelenko said he doesn’t think the company’s early ad products will be focused on traditional metrics like performance value or demand-generation. Instead, they might be focused on driving future rather than immediate actions.

“Asking somebody a question isn’t necessarily driving an immediate action,” he said. “It’s about driving consideration. And so ironically, whenever our ads businesses will first [begin], it won’t be directly competitive with [Google] AdWords because we’re not trying to drive conversions to a specific action.”

As anyone playing in the generative AI space has learned by now, getting consistently accurate answers from large language models is a serious challenge, especially with larger document sets and longer documents. And just like with traditional search, knowing which links and documents are more trustworthy and more valuable than others is a tricky task. That’s also something search giants like Google and Microsoft — as well as startups like Brave — are still trying to solve.

“The reason it’s a non-trivial challenge is it’s very hard to rank the signals on internal data versus web data,” Shevelenko said. “There are many signals. Like how often a site is visited, how often it gets refreshed, how others index it. When we rank the sources that respond to a query, there are a lot of useful signals. That’s what leads to good answer quality.”

Developing enterprise search using LLMs is a lot harder than just plugging in another platform. Kian Katanforoosh, cofounder of Workera, which helps upskill companies with AI abilities, said companies need three pillars: when developing their approach to AI: strategy, technology and an AI-trained workforce. However, he said the third leg of that stool is often where risky failures happen.

“Unknown unknowns are a huge problem,” Katanforoosh said. “Like people thinking they know but they don’t know, or people not realizing they know a lot but they know [more than they realize].”

Perplexity is just one of several companies going after the enterprise market for generative AI. Last week, Snowflake announced a new Arctic open-source AI model for enterprise customers while other startups such as Cohere and Snorkel AI also announced new features for their platforms. Snowflake’s LLMs are built using tech derived from Neeva, which the cloud data company acquired a year ago. Like Perplexity, Neeva had also sought to also compete with Google by making a better search product for everyday users that was without ads and also privacy-focused.

SEO experts say competing in search requires having a data moat that most companies don’t have, other than Google, Microsoft and Apple. However, some say Perplexity could succeed by staying focused on enterprise first.

Ethan Smith, co-founder of the SEO agency Graphite, said Perplexity could win by making a veritable application to enterprise research — kind of like ResearchGate or Google Scholar. Another way could be to partner with a company that has a lot of proprietary data — like Google, Apple or Microsoft.

“Everyone is going to have access to the same algorithms,” Smith said. “What will matter is the training data, and the training data would be outcomes. So whoever has more outcome data and a better feedback loop, they’ll win regardless of what the algorithm is.”

Prompts and products — AI news and other announcements

Earnings from social and search players signal that AI will be a long-play investment.

Adobe released its new Firefly 3 image model to power AI tools for Adobe and other platforms.

Coca-Cola debuted a new app called “Coke Sounds” to make AI-generated sounds. It also separately announced a new five-year $1.1 billion deal with Microsoft focused on generative AI and cloud computing.

Apple reportedly acquired the French AI startup Datakalab to help with on-device processing.

Hugging Face now has a journalism & AI community for learning and discussing AI and media.

New research from Reuters/Oxford looks at news and chatbot headlines.

Thousands of explicit “AI Girlfriend” ads are showing up all over Facebook and Instagram, according to a Wired investigation, which reported the controversial ads violated Meta’s policies.

The UK is increasing its scrutiny of AI startups, according to The Guardian.

A report by Fortune explores how the BBC is using AI in ways that help inform audiences.

Five AI lessons from the International Journalism Festival.