Yelp might not be the first company that comes to mind when someone mentions artificial intelligence, but Chief Product Officer Craig Saldanha said AI is already transforming the Yelp experience.
In fact, most of the company’s recent announcements center on AI, whether that’s adding new AI-powered summaries or launching an AI assistant to connect consumers with service providers. So I spoke to Saldanha (who joined Yelp after nearly a decade at Amazon) to learn more about Yelp’s AI strategy.
We also discussed what advantages Yelp brings to the AI race, how Yelp can add AI without threatening the authenticity of the user reviews on the platform, and how it’s competing with new avenues for local discovery like TikTok.
This interview has been edited for length and clarity.
Going back through all the recent news from Yelp, it’s all AI, AI, AI. Can you say more about how you look at AI and the role it plays at Yelp?
Just to set the table, our stated mission hasn’t changed. Our goal is to connect consumers with great local businesses, and that hasn’t changed over time.
We’ve been investing in AI for more than 10 years now. But over the last couple of years, the advances in generative AI and other LLMs has really allowed us to take advantage of a couple of things. The first is, the real differentiator of Yelp is the hundreds of millions of reviews that we have. LLMs essentially allow us to parse all of that data in a way and at a speed that we’ve never had before. It allows us to present information to consumers in a way that feels both precise, as well as personal — you can now find that needle in the haystack.
We recognize that users come to Yelp to connect with either other users or pros, and they come because of the authenticity of our content, because they know it’s from actual human beings. We’ll never take that away. So we use AI, essentially, to remove all of the friction to facilitate those types of connections.
We think about the consumer as having three phases when they come to Yelp. The first is, they come with a very strong search intent, they know they want to find a plumber, they know they want to find a good place for lunch, etc. So the first step is essentially defining that intent. The second step is, once we’ve helped them define that intent, and they know exactly what they’re looking for, we present them with a lot of different results, and they need to pick either a single business or like a couple of businesses that they want to connect with. Then the third step is actually making that connection. We’ve invested heavily in AI in each of those steps.
The first step, refining search intent when a consumer comes to Yelp. [If you’re doing a simple search like] “I’m looking for a Mediterranean restaurant,” we have a pretty sophisticated model that first understands what you’re looking for, and then essentially decides not only what restaurants to show you, but the order in which to show you those restaurants.
What’s really cool now is the advent of LLMs means you can search for even more specific things, and it will understand what you’re looking for. As an example, we live in suburban Seattle, and my wife is always on the hunt for these very specialized spices for different kinds of cuisines. In the past, let’s say she’s making Indian food, I would look for “Indian grocery store,” and we essentially do a match for those words and return the results. Now, I can search for a very specific Indian spice, and the LLM will understand that it’s a spice, that it’s found in an Indian store. Even better than that, it is able to go through all of the reviews that we have, understand when other consumers are referring to those spices — so it could be a different spice, but it understands now that those grocery stores actually carry these types of spices.
Then when it shows me the results, it will not only order them in a way that is a better match for me, but it will highlight the specific snippets of consumer reviews. That’s super powerful, it genuinely feels very, very personal.
In the past, say, if you were looking for tacos, we would show you restaurants that had tacos, not a big deal. Now, we are able to look at every photo that consumers have submitted for every single restaurant, pull out tacos from those specific restaurants and show them right in search.
I think the piece that I’m most excited about is that we’re taking [these capabilities] off of Yelp as well. So we’ve recently announced what we’re calling our Yelp Fusion API. [This interview was conducted prior to a recent controversy among indie developers over paid access to Yelp’s API.]
Now, someone on a third party, let’s say a travel website, can essentially ask a question, “Where can I find a Sunday brunch that’s open after 11, and kid friendly?” And through our API, we can return with the same level of personalization off of Yelp. I think that just expands the number of consumers we can help simultaneously.
For Yelp to differentiate in AI, you don’t need to have the most incredible AI team or create breakthrough core technologies, it’s more about this unique data set. Is that right?
I think it’s both. Our core value proposition is content. Our consumers are just awesome, they write such deep reviews that are so nuanced. And that’s what keeps folks coming back.
For finding snippets and stuff like that, we can use a lot of off-the-shelf models, because the core problem we’re trying to solve is simply natural language processing.
I think the place where our technology shines is in areas like Yelp Assistant. In 2016, Yelp introduced “request a quote,” and that allowed consumers to quickly get a variety of quotes from a variety of service providers. We’ve expanded that over time, we added Yelp Guaranteed, all of that has helped to reduce the friction and drive quicker and deeper connections.
Then last year, we updated our whole back-end AI model to use neural networks; that really helped drive precise matches. So then the next problem to solve was, what if you don’t know which [type of pro] you’re looking for? If you see a wet spot in your wall and you don’t know if your roof is leaking, your gutter’s leaking, or if you have a broken pipe.
We felt like the next step of this was: Just tell us what your problem is, we’ll help you narrow down, we’ll help you find the pro.
And I think that’s where we really push the technology, because general models will give you general responses. What we have, and what we’ve built up over time, is a very deep understanding of what pros do, and what types of jobs they don’t actually do, too.
You also mentioned the importance of protecting the authenticity of user reviews. As you imagine AI, including generative AI content, becoming a more central part of the Yelp experience, how do you protect that authenticity?
First, just to say upfront, using Gen AI to write reviews is a violation of our policies. We work very hard to keep those types of reviews out. We have been investing in pretty sophisticated solutions for a long time to validate the authenticity of reviews, and whether it’s bots, or solicited reviews, this was something that we were thinking about from day zero. And so we are prepared for it, we’ve deployed a bunch of solutions, all types of technology. It’s a constant game of keeping ahead of what bad actors might use; we will continue to draw a hard line.
I imagine that one of the incentives for writing a thoughtful review is that I’m hoping somebody will actually read it, not that it’s just going to be fed into an AI model that spits out a summary. How do you make sure there’s still an incentive for users to write good reviews?
Overall, I think Gen AI will be very helpful for both the quantity and the quality of the reviews. The more connections you get between consumers and businesses, the more shots you have at writing reviews.
On the review writing piece, there are a couple of things that are very helpful. First is, we are now using AI — and specifically Gen AI — to give you gentle nudges and prompts to help you remember what made your experience special. So as you’re typing, if you talk about the ambience, it will give you a little tag that says, “You’ve checked off the ambience, now you can talk about the service, you can talk about the food, etc.” We’ve rolled this out for restaurants, we’re rolling this out for other categories. That really helps with the depth and the quality of the reviews.
The second piece is photos. Now your photo surfaces into places which are new. We have a brand-new home feed, which is very visual, it’s very photo- and very video-heavy. And we talked about [photos in search].
Then to answer your specific question: We put our reviews front and center. So instead of telling you what the answer is, we have gotten to the source faster. We’re taking you to the reviewer and to the review. We’re making it easier for you to find the exact user who had the same experience.
So my hypothesis is that it’s actually an even bigger motivation [now]. Because in the past, if you’re at a restaurant that has 200 reviews, and you’re the 200th, [you might think,] “Can I really add value?” But now, knowing that I can say, “They brought my 18-month-old a highchair and they gave her something to color with,” that’s new information. If somebody with an 18-month-old is looking for it, they’ll find my specific review.
And now we actually close the loop. So if you write a review, we will actually send you feedback and say, “Since you wrote that review, this business has got 200 more views” or “seven people found it helpful,” etc.
So we’ve been talking about how AI has already changed the Yelp experience. Is there anything you can say about what you’d like to see happen with AI and Yelp in the future?
We have pictures and we have video and we have descriptions, and we’re using AI to stitch all of those together and give you that whole 360 experience of what it’s like to actually be there. I’m very, very excited about that because that’s not a single person’s point of view, but it’s all user-generated content. We’re not artificially generating anything, so it feels authentic.
On the business side, it’s not Gen AI, but we have a ton of AI, and a really big team focused on matching. Pros and businesses have told us we have high intent consumers, and they want those high intent leads. So we spend a lot of time just focusing on how do we get a better match? How do we match the right pro with the right consumer at the right time?
The second piece [for businesses] is: we announced smart budgets. We found that a lot of new businesses, they’re really good at what they do, but they don’t know how to run a business, it’s day one for them. So we have this AI tool that takes a bunch of information about where they’re located, what competitors are spending, what’s the size of their business, what the number of leads we think they would need to grow, and every business gets its own recommendation for how much money we think they should spend.
[Back on the consumer side,] AI is getting good enough that you can just show me a picture or take a video [and we can match you with the right pro or business]. We’re not there yet, but it’s quite logical to see that’s the path. And then on the pro side, there’s a lot we can do to help them qualify leads, whether it’s asking questions on their behalf, whether it’s making sure that they never miss a call by having an assistant for them, by guiding them on how users might prefer their response, whether it’s structured or unstructured.
Stepping back from AI, the local discovery landscape has changed dramatically in the last few years. I have friends who now say, “Let’s go try this dish, let’s go to this restaurant because I heard about it on TikTok.” And obviously, search is changing. So as all this is happening, what do you see as Yelp’s role and differentiator?
First, we already talked about the breadth and depth and volume of our reviews. At Yelp, you get the wisdom of the crowd, you get a collective sense of what a restaurant is, and you’re able to very quickly combine different points of view and choose which one is closest to your own. Versus with the influencer model, you could trust an individual, that’s why you follow them, but it’s a single individual.
I think the two less obvious [differences] are, one is just the breadth of categories that we have on Yelp. It’s quite easy to follow influencers for restaurants and maybe home decor and stuff like that. But as you think about plumbing and roofing and accountants and lawyers and doctors, the breadth of coverage that we have is very, very useful.
Then the last one is really the balance of the views. Most of the time on social media, people will share if they had a phenomenally good experience, or a phenomenally bad experience. There was a study done on the review distribution of various platforms, and Yelp has the most even distribution between one, two, three, four and five stars. If you really want that balanced view, as opposed to the polarizing one star or five stars, that’s where Yelp can make a difference.
