Restaurants given free search advertising on Yelp received more page views and customer inquiries in a large field experiment, but the gains differed across businesses, researchers found.
Independent restaurants and those with stronger existing ratings tended to see larger effects than otherwise comparable businesses. The findings give small-business owners a reason to assess their online reputation alongside the price of an advertising campaign.
The Marketing Science study, published in 2023, examined 18,294 restaurants. Researchers Weijia Dai, Hyunjin Kim and Michael Luca randomly assigned 7,209 of them to receive the platform’s standard advertising package free for three months.
The experiment measured activity on one platform during a particular period. It did not establish that every restaurant would make a profit by buying ads at current prices.
What customers did after seeing ads
Yelp’s advertising placed participating restaurants in paid search positions. The researchers tracked the resulting activity at the business level, including visits to restaurant pages and actions that could indicate an intention to visit.
In the researchers’ working-paper account, page views rose 19% on average. Other measured actions, including calls, requests for map directions and visits to restaurant websites, rose between 7% and 14%. The number of customer reviews increased by 5%.
Those figures refer to different outcomes. A request for directions is a stronger signal of interest than an ad impression, but it is not proof of a completed meal or a profitable sale. The percentage increases are relative to the comparison group, not percentage-point changes.
The paper reports that the measured effects disappeared after the free advertising stopped. A short campaign may therefore generate activity while it runs without leaving the same level of additional traffic afterward.
Why results differed between restaurants
Independent and local-chain restaurants gained more than national chains with similar measured characteristics. One possible explanation is that a familiar national brand is already easier for customers to recognize, so another search placement provides less new information.
Restaurants with higher ratings, more reviews and more organic traffic before the experiment also showed larger gains. Organic traffic means visits that arrive without the paid placement being studied.
Advertising may draw attention to a listing, but the customer still sees the information on that listing. Businesses assessing a campaign should consider whether their opening hours, menu, contact details and customer feedback give a potential visitor enough information to make a decision.
The study does not prove that changing a rating will produce a particular advertising return. A business’s rating and other characteristics can reflect many differences that the experiment did not independently change.
How to assess the cost of a campaign
The participating restaurants received advertising free during the test. A restaurant paying for a campaign must also weigh the cost against the additional customers and contribution from each sale, after food, labor and other incremental expenses.
A useful measurement plan starts before the campaign. Record the number of calls, direction requests, bookings and orders the business normally receives, then compare performance during a defined test period. Where possible, use a comparison group or holdout, a similar audience that does not see the ads, to account for changes that would have happened anyway.
Managers should avoid treating every click as a new customer. Some people who respond to an advertisement may already have planned to visit. Likewise, customers may see an ad on one device and order through another channel, making a single metric incomplete.
The research team worked with Yelp, which provided data and engineering support. The authors disclosed that they had consulted for technology companies including Yelp; the paper says their compensation and ability to publish were not dependent on the results.
Our earlier coverage of a retailer’s product-page experiment shows a related measurement issue: an action at one stage of a customer journey is not automatically a completed purchase.