Yelp, the popular Web site that lets consumers review everything from bistros to body shops to yoga studios, quietly began running an experiment in San Francisco over the past week. The pages for a small fraction of the city's restaurants on the site now bear a new consumer alert.

"Following a recent inspection," the pop-up box says, "this facility received a food safety rating that is in the bottom 5% locally, and is categorized by inspectors as 'poor.'"

The notice is modeled off warnings that Yelp virtually slaps on businesses it suspects of soliciting fraudulent reviews. But this one, targeting all of the restaurants in that bottom 5 percent according to San Francisco health inspectors, could have a more far-reaching effect.

Harvard Business School's Michael Luca is tracking what will happen next. With Yelp's cooperation, he's looking for two effects: change in how consumers behave, and change in how restaurants do. Will diners shy away from these places, even when their poor health grades clash with tasty reviews? Will the restaurants themselves be shamed into upping their scores? The latter result would be much more significant.

"Without restaurant behavior change," Luca says, "it’ll be very hard to have a big effect on food-borne illness."

That Yelp would even want to affect food-borne illness isn't intuitive. The decade-old company has built a business selling online ads against reviews generated by its users. It won't make any money by preventing food poisoning or even partnering with city regulators to detect it. And yet this experiment — along with another project Yelp is running in Boston — has the review site wading further into the world of public policy, behaving itself in a way that could achieve some of the ends of regulators.

"Yelp’s job is to predict in an online way the experience consumers can expect will happen in the offline world," says Luther Lowe, Yelp's head of public policy. "To the extent that we can augment the consumer opinions and ratings that our users rely on with government data that they’re creating with their tax dollars — that’s a great win-win."

Arguably, this online test will change what happens in the offline world. That's the premise for even running such an experiment. Yelp's power to do that will no doubt rankle the restaurant industry. Luca acknowledges that demand for the poor performers may well go down. But what if, instead of nudging them to clean up, these notices push them out of business?

The National Restaurant Association says it supports consumer transparency, but it points out that Yelp itself is largely unregulated. The lobbying group, which was also skeptical when Yelp first rolled out health inspection scores on restaurant pages on a limited basis two years ago, is worried in particular about timing. What if a dinged restaurant promptly improves, but it takes weeks or months for its public score to?

Yelp's efforts raise fascinating questions about what happens to obscure government information when it becomes a lot more public.

"It’s our strong belief that this is something consumers have a right to see," Lowe says of the health scores. And this information, Yelp points out, is already public. Where Yelp becomes a player in public health is when it shows you that information, packaged in a red-framed box, right at the moment when consumers are making a decision.

"Whether or not Yelp works on this specific initiative, Yelp at some level is intertwined in policy issues," says Luca, who has been conducting research on the company's data for several years. "It’s always been clear to me that there's a deeper link between Yelp, consumer protection and policy than people might immediately think."

In some of Luca's previous research, he's found that the language contained in Yelp reviews — words like "dirty," "moldy," "pee" — can be analyzed to automatically identify food-safety threats. The sheer fact that consumers are writing things on the site that can be predictive of health violations implies that Yelp could be useful to policymakers, Luca says. Those policymakers, in turn, he adds, can either turn a blind eye to the site's intelligence or embrace it.

In Boston, Yelp ran a prediction tournament for algorithms that could forecast health inspection scores using the site's ratings and reviews. Using the winning algorithm, Luca says, Boston could catch the same number of health violations with 40 percent fewer inspections, simply by better targeting city resources at what appear to be dirty-kitchen hot spots. Boston is now considering ways to use such a model.

"There’s no doubt in my mind that within three years," Luca says, "a bunch of cities will be using things like Yelp text to predict where they should be sending their inspectors."

Yelp wants the site to be a two-way platform for food-safety information: Cities could use it to push out their inspection results to consumers (as in San Francisco), while also pulling in cues from consumer reviews to conduct more efficient inspections in the future (as in Boston).

Ideally, Luca says, none of this means that any restaurants have to go out of business. "The mark of success in this for me," he says, "would be if businesses stop getting poor scores."

There's actually some precedent for that hope. In the late 1990s, Los Angeles County began to require restaurants to post their health grades in the doorway, an old-school form of transparency other cities now follow, too. After that mandate went into effect, the C's and D's quickly became A's and B's. Research also later found one other effect: Hospitalizations for food-borne illnesses around Los Angeles dropped.