When Better Job Information Changes Search, Not Employment
This series examines a cluster-randomized field experiment in Delhi that gave job seekers information about digital search and followed them for up to a year. Its central question is narrower, and more revealing, than whether a job platform “works”: when people learn more about available work, do they search more effectively, and does that change whether they find employment? The answer helps readers distinguish a change in search behavior from a change in the number of people employed, and judge what information can reasonably be expected to accomplish.
The result is a useful challenge to a familiar assumption. More information did change behavior, but not in one common direction. Men reduced their overall job search, particularly when they began with more optimistic wage beliefs. Women did not reduce how often they searched; instead, they shifted away from some offline routes and toward digital ones. Personalized information was later associated with higher earnings and more salaried work for women in some months, but the intervention did not produce a statistically significant increase in employment for the sample as a whole.
That combination is not a contradiction once “search” and “employment” are separated. It is a clue about where labor-market information enters the decision process.
A job search is a chain, not a single outcome
A job seeker does not move directly from receiving information to holding a job. They begin with beliefs about which jobs exist, what those jobs pay, and whether someone like them is likely to obtain one. They also face a cost of searching: time, money, transport, social exposure, and the effort of sorting through opportunities. Search then takes a form—asking relatives, visiting an institution, using a digital platform, or contacting an employer. Only after those steps does a possible match encounter an employer's response, a contract, and the worker's decision to accept.
Each link can move without the next one moving. A person may search less because new information lowers the expected payoff of spending another day looking. Another may search the same amount but replace an unreliable social network with an online channel. A third may apply to more vacancies yet receive no suitable offer. Employment is therefore an extensive-margin outcome: whether someone is working at all. Earnings, hours, and occupation describe other margins, including the quality or structure of work among people who are employed.
The Delhi experiment is valuable because it observes several of these links instead of treating an application as a job. It records beliefs and expectations in surveys, search modes and intensity, platform activity, monthly employment, earnings, and occupational category. The chain also makes the title's apparent paradox intelligible: information can change the inputs to a search decision even when it cannot guarantee the final match.
The diagram is not a claim that every person passes through these boxes in the same way. It is a map of the possible points at which an intervention can have an effect—or stop having one. The study's importance lies partly in showing that the path differs by starting position.
What the experiment actually tested
The experiment covered 3,391 current or prospective job seekers living in 130 polling-station clusters across six Delhi districts. Eligible households included someone aged 18 to 40 who was currently looking for work or planned to work in the near future, had at least a high-school education, and had access to a smartphone. Baseline interviews took place between May and August 2023. The intervention ran from September 2023 to February 2024, followed by surveys at roughly six and twelve months.
Clusters were assigned to one of three groups. The control group received neither the information session nor the phone messages. Both treatment groups received an in-person explanation of digital job-search platforms, assistance registering on a nationwide blue- and grey-collar matching platform, and guidance on creating a profile, searching for vacancies, and finding skills resources. The difference came afterward. The Non-Personalized group received six general messages. The Personalized group received six messages adapted to baseline characteristics such as gender, whether the respondent was currently or prospectively seeking work, and whether they aspired to jobs listed on the platform.
This design identifies the effect of being assigned to a bundled outreach package. It does not isolate the effect of the app from the in-person session, or the effect of one message from the other five. Nor does the comparison between personalized and general messages show what personalization would do in the absence of the common assistance. Those distinctions matter because a platform can make opportunities visible, but visibility is not the same as relevance; relevance is not the same as an employer response; and a response is not the same as an accepted job.
The researchers used both survey and administrative data. The platform recorded registrations, applications, calls to human-resources staff, and callbacks, but not the final offer, the contract offered, or whether an offer was accepted. That missing tail of the process is not a minor technical detail. It marks the boundary between changing search behavior and changing employment.
The starting point was not an empty information set
The people in the study were not simply waiting for someone to tell them that jobs existed. They already had aspirations and expectations. About 69% reported that they were in, or moving toward, their ideal career path. Salaried work was the leading aspiration for both genders, selected by 80% of women and 67% of men. Women were more likely to prefer public-sector salaried work, while men were more likely to prefer self-employment.
Those aspirations were not matched by realistic wage expectations. The researchers compared expected salaries for occupations advertised on the partner platform with average earnings in comparable urban occupations from India's 2022–23 Periodic Labour Force Survey. Expected salaries exceeded the benchmark on average for both salaried and casual work. The gap was larger for men: INR 7,878 versus INR 6,714 across the listed occupations, and INR 8,608 versus INR 7,489 for salaried jobs.
These figures do not show that the respondents were foolish or that ambition was a mistake. The comparison is between a person's expectation and an average occupational benchmark, not between a person's forecast and the exact job they would eventually receive. But it does establish a relevant starting condition: many respondents were making search decisions with an imperfect picture of the likely payoff, and men's picture was more upward-biased than women's.
That distinction changes the interpretation of later behavior. If information corrects an overly high estimate of the wage or probability of a desirable job, the rational response may be to search less, wait, acquire skills, or accept a satisfactory existing position. If information makes an opportunity visible to someone who previously lacked a usable route into the market, the response may instead be to search through a new channel. “More informed” does not prescribe one behavioral direction.
Information is also a question of access
The baseline differences were not only about expectations. Women in the sample were much less likely to be employed—about 25%, compared with 50% for men—and among those who had searched, women were less likely to have applied for a job: 69.4% versus 74.3%. Men relied more on family, friends, and neighbors for information and referrals. Women relied relatively more on educational institutions and digital platforms. Among respondents who had ever applied, offline modes were still more common than online ones.
These patterns are consistent with a gendered information friction. A social network can be a powerful source of leads, but access to a network does not ensure access to vacancies that fit a person's preferences, safety needs, schedule, or desired occupation. If women are less able to draw on traditional networks, their lower use of those networks may reflect constrained access rather than weak interest in work. The study also notes that women were more likely than men to express interest in registering for the partner platform and to register once offered the opportunity.
This is why the digital component has two possible meanings. It can be a cheaper or more convenient way to repeat the same search. It can also be a substitute for information that was previously available mainly through personal contacts. In the second case, a woman can search more online while searching no more overall: the channel changes, the cost structure changes, and the opportunities encountered may change, even if the number of hours spent searching does not.
The gender contrast should not be read as a fixed difference in how men and women value work. It reflects the interaction between prior beliefs, labor-market exposure, social access, family circumstances, and the design of the information. The same message can correct an inflated expectation for one person and open a credible route for another.
What the headline result has—and has not—established
By twelve months, the intervention reduced the full sample's likelihood of reporting that they had searched for work by 3.5 percentage points. That average was driven mainly by men. Women showed a different pattern: their likelihood of using an online or digital mode rose by 4.3 percentage points at six months and 6.0 points at twelve months, while their use of offline modes declined. This is a change in the composition of search, not simply a change in its volume.
The employment result was quieter. Across the twelve-month period, the study found no statistically significant effect on current employment for the full sample. That does not mean nothing happened. Women assigned to personalized messages later earned more than women in the general-message group in several months and were more likely in some later months to be in salaried rather than casual work. Those estimates are selective and often marginally significant; they do not amount to a large, uniform employment transformation. They do suggest that information may affect who finds a better-fitting job without increasing the total number of people working.
The safest conclusion is therefore neither “digital platforms solve labor-market frictions” nor “information failed.” The intervention changed beliefs, take-up, applications, and search channels. It may have improved the quality of search for some women and reduced costly search based on unrealistic expectations for some men. But the study does not directly observe welfare, job stability, contract quality, job duration, or the final offer-and-acceptance process. Nor should results from a smartphone-connected, relatively educated sample of Delhi job seekers be treated as estimates for every labor market.
The first question, then, is not whether the participants wanted work. It is what they thought work would look like, and which routes they had to learn about it. Those starting conditions determine whether more information feels like a reason to keep searching, a reason to change how one searches, or a reason to revise the plan itself.