Not all profiles on care platforms represent active platform users. Some users may create an account out of curiosity and never return, while others may use the platform to find a longer-term job or a child carer and later become inactive. We therefore estimated the number of users who are actively looking for platform work or a platform worker by considering a user to be active if (1) they logged into the platform within the last 31 d and (2) did not sign up for the platform within the last 31 d. In our data analysis, we refer to a user fulfilling these requirements as “active”. We set the threshold at 31 d because the largest platform (Babysitting24) also uses a threshold of 1 month to determine (in)activity. On platforms for which we did not have access to the membership length, we considered active users to be those who were last online within the last 31 d, so we only use the first condition. We refer to a user fulfilling this requirement as “recently active”. The differences in metadata (e.g. how and in what granularity the date when the user was last online or joined the platform is shown) between platforms do not allow us to compare activity levels between platforms but can give us insight into how active users are on these platforms more generally.

We acknowledge that this approach has limitations. First, “active” does not imply that this user necessarily found a worker or any work via the platform (let alone for full-time), but just that they tried to. Second, it likely undercounts users who found longer-term placements and have not recently visited the platform and may instead overcount those who recently registered on a platform but never found work or a worker. Third, it is also limited from a feminist perspective in that it undercounts (predominantly female) workers who take longer leave, for example for maternity leave or due to childcare responsibilities (cf. James, 2024a). This is particularly unfortunate as it invisibilises aspects that are particular to life courses of women.

In addition, previous studies on delivery, ride hailing and also crowdwork platforms have disclosed a widespread practice of several workers sharing one user account or users subletting their accounts to other workers (Tironi and Albornoz, 2022). On delivery and ride hailing platforms, this can be advantageous for workers whose own accounts have been deactivated by the platforms or who do not have work permits and are therefore unable to sign up on platforms that check certificates. On crowdwork platforms, sharing an account with high ratings might provide better access to work. Such practices might lead to undercounting the effective number of workers. As the marketplace platforms we analysed for this study do not check certificates and hardly any have ratings, there is no equivalent need for account sharing. Furthermore, childcare platforms typically require workers to share photographs and personal information about themselves to create trust for customers, which creates substantial hurdles for account sharing. For this reason, we assume that the number of workers' user accounts roughly equals the number of workers looking for jobs via the platforms we analysed.

Childcare work has traditionally been dominated by women. The advent of platforms has raised the question whether platformisation will lead to more men entering this labour market segment. To test this hypothesis, we sought gender data for users. However, none of the platforms we scraped showed users' self-reported gender in their profiles. The only platform that would have indirectly given us access to self-reported gender is Babysitting24, as it is possible to filter users by gender in the search function (using only a female/male binary gender selection). Because this would have required running twice the number of searches, which would have significantly increased the scraping time, we therefore opted to scrape the profiles without filtering and infer gender the same way as for the other platforms: by using name-based gender inference from an application programming interface (API) (Gender API, 2026).

The API returns one of three labels: female, male or unknown. It does not include a non-binary category, which is sensible given that names typically do not indicate non-binary identities. Non-binary individuals often use gender-neutral names, but these names are also widely used by cisgender individuals, making them unsuitable for reliable classification. We cleaned the name fields before submitting them to the API. For some platforms, for example, names included a first initial of the last name (e.g. “Maria S.”). The API returns a confidence score (“accuracy”) for each prediction, which is defined as a confidence measure that they calculate based on their data pool of name and gender pairings. A score of 100 indicates maximum confidence. We chose a minimum accuracy of 80. Gender inferences with a lower accuracy were re-labelled as “unknown”.

Lastly, we manually reviewed a subset of the gender labels to correct obvious misclassifications. This was especially necessary in cases where the input name was a company rather than a person's name. In those cases, the API sometimes returned an incorrect gender rather than “unknown”. In those instances, we reassigned the gender label to “unknown”.

We are aware that this name-based classification limits the explanatory power of our analysis in that it reproduces a cisnormative understanding of gender. It does not tell us about how our users would self-identify and thus does not allow us to break out of a binary gender classification. This is a limitation that we share with many quantitative studies that rely on name-based gender inference or self-identification data based on only binary options. Non-binary gender identities thus remain uncounted and invisible.

To figure out the extent to which platform-based childcare is an urban phenomenon, we mapped users to locations and checked for correlation between population density and users per capita. For our analysis, we used the population density data from the Swiss Federal Statistical Office (SFSO, 2023). We had access to location information from the users on all platforms, but there is no standard format for this information. Given the format of our population density data, which provides the population density for every municipality name, we tried to count the number of users for every municipality.

While we could identify each user's municipality or village, we sometimes also had users' postal code, canton name or country name. This additional information helped resolve issues when mapping the municipality name to our population density data. Such issues can occur for multiple reasons. (1) Due to the multilingual nature of Switzerland, the same municipality can have multiple names. Even on the same platform, the same municipality might thus be spelled in different ways. On Babysitting24, for example, we find the city of Geneva listed as Genève, Genf, Ginevra and Geneva. We therefore standardised municipality names based on official city names given by the Swiss Federal Statistical Office (2026) (listed alongside German, French, Italian, Romansh and English translations). We added unmatched names that we came across that we could easily match ourselves, such as the Portuguese Zurique, to the list. (2) Some city names contain the prefix “St.”, which is sometimes spelled out as “Sankt” (in German) or “Saint” (in French). These variations are not included in the SFSO's list, so we had to normalise these names ourselves. However, these kinds of names are sometimes not displayed properly (for example, on the platform MisGrosi, we found instances of municipality names like “St. Moritz” being displayed as “St.” only) or are scraped incorrectly (only scraping the “St” instead of “St. Gallen”). In cases like these, we relied on the postal code to find the correct municipality name. (3) Sometimes we only had the village name, which is at times different from the municipality name. To map our data to the population density data, however, we needed the municipality name. (4) Some municipalities share names and can only be differentiated by the addition of their canton or municipality. (5) Sometimes, the combination of the village name and postal code matches multiple municipality names.

We standardised the location of users as follows. First, we matched the municipality or village name to the translation dataset to get the official name where possible. For the location names that we could not match to our population density data, we checked if the name matched a village name in the official directory of towns and cities (Swisstopo, 2024) and then got the corresponding municipality name. If that did not work, we used the postal code to get the municipality name if it was available to us. If this also did not work, in the special cases of Babysitting24 and Seniorservice24, we ran our location data through Nominatim API (Nominatim contributors, 2026) to remove users based outside of Switzerland. The full location data that we derived from the Nominatim API included the municipality name for users based in Switzerland, giving us better chances of matching users' location data to the population density data. Due to these issues, our standardisation of location names is necessarily an imperfect one. However, the process worked well enough for our aim of understanding correlation between users per capita and population density: we were able to standardise all but 285 of the total 88 269 user locations. Only in 0.32 % of cases were we unable to map the user to a location that appears in our population density data.

Our analysis of location data has a key limitation: users might not register at the address where they are actually located, but in areas with a high population density where they expect to find most job opportunities to increase the visibility of their profiles. We did indeed find some indication of this worker strategy when analysing user locations at the neighbourhood level in the city of Zurich. This analysis showed a disproportionate share of workers registering exactly in the city centre. Our data do not allow us to assess the extent of this phenomenon, but this means that we probably undercount workers who live in areas with a low population density and overcount those who live in areas with a high population density. This is relevant for our results presented in Sect. 4.3.

Table 1 provides an overview of the number of accounts we found on each platform from April till July 2025. These numbers exclude duplicates (i.e. if the same profile gets shown in multiple searches). Note that on some platforms, users can set their profiles to a private mode, where they can browse other profiles but cannot be found by other users. Such profiles can also not be seen by our scraper and are thus automatically excluded from our analysis.

The user numbers reveal a highly concentrated market, with one platform (Babysitting24) having 10 times more profiles than any other. While two more platforms (Babysits.ch and Care.com) have sizeable numbers of more than 1000 workers, all other platforms have much lower worker numbers. The oligopolising (Costhek Abílio, 2023) tendencies of platforms, in which one tends to dominate the market, have been observed across platform markets. They result from network effects that incentivise users to choose the biggest platform in the hope of finding the best offers (Srnicek, 2017).

Using our definition of (in)activity from Sect. 3.3.1, between 14 % and 37 % of workers and between 8 % and 38 % of customers are active per platform (see Fig. 1).

We excluded MisGrosi in these numbers even though the platform has a feature that is meant to indicate whether a platform worker is active or not. Using this feature, we found 29 of the 64 registered platform workers to be active, so just below 50 %. However, as we do not know how this activity status is determined, we do not consider it further in our analysis.

The worker-to-customer ratios across the analysed care platforms reveal stark imbalances that are indicative of a surplus workforce. As Fig. 2 shows, workers outnumber customers in all but one case by a factor of 2.18 to 11.01. When considering only active users, workers still outnumber customers by a factor of 5.28 to 9.43.

Here, we excluded GreatAuPair as the number of active users is so small that the ratio would not be robust.

A key reason for the surplus of workers compared to customers is that platforms actively strive to attract a workforce much larger than their customer base because this gives potential customers more choices and flexibility. This oversupply weakens the negotiation position of workers and potentially leaves many of them unable to find jobs. Thereby, they mostly operate on standby, looking for work and ready to work if they find a job (Keller, 2023). The price for the abundance of choices and flexibility for customers is thus borne by the workers (Keller, 2022).

A notable exception to this imbalance is MisGrosi, a platform whose name literally translates to “my grandma”. It thus aims to help parents find a substitute grandma for their kids. Here, customers outnumber workers by a factor of 22.5 (0.04 workers per customer). The scarcity of “grandparent workers” compared to the large number of parents seeking childcare on this platform can be explained in several ways. One explanation is that the target demographic for the “grandparent” role may be less tech-savvy than the potential customers. However, we also hypothesise that parents might also expect “grandparents” to offer care for free or at much lower rates, as the concept of the platform focuses on the familial construct of grandmothers, suggesting the workers might not need the income from this work and would do the work out of love.

Overall, our data mirror the broader pattern of an oversupply of workers: the rhetoric of flexibility that is common in the platform economy often disguises this structural imbalance. As long as there is this reserve of workers, customers' needs are prioritised while the risks of irregular hours and lack of available jobs are placed onto workers (Wiesböck et al., 2023).

To assess the amount of care work that is actually mediated through platforms, we analysed review information as a proxy for mediated gigs wherever such data are available. This is the case for Babysitting24 (workers), Seniorservice24 (workers), Care.com (workers) and Babysits.ch (workers and customers). We find that, across all platforms, reviews are extremely sparse: considering all scraped worker profiles, the median number of reviews is zero on every platform. Mean values are very low, ranging from 0.01 reviews per worker on Seniorservice24 to 0.14 on Care.com.

For two platforms, we also have access to the number of bookings. Again, we find a median of 0 and a very low mean of 0.10 (Babysits) to 0.14 (Care.com) bookings per worker profile. This could mean that the platforms mediate very little work. However, this would be at odds with the high user numbers we found. Rather, we assume that workers and customers use the platforms to get in touch with each other but often do not leave reviews and also do not register their transaction as a formal booking but arrange it off-platform. This can be advantageous as it might allow them to save on platform commissions (Gruszka et al., 2025). For the purpose of this paper, however, this means that our data do not give us any estimate as to how many transactions occur. Neither ratings nor bookings provide us with a reliable assessment of the amount of care work that is actually mediated through these platforms. From a feminist perspective, this is relevant because it means that much of the feminised care work remains invisible also in our age of apparent data abundance – not just in labour market statistics but also in our scraping data – even when the work is being paid and mediated through platforms.

Across care platforms, we observe pronounced age differences between types of care work, most notably between childcare and senior care. Among the scraped platforms, Babysitting24, Seniorservice24, Babysits, Care.com (childcare) and GreatAuPair provide age information for workers, with very high coverage rates (between 94 % and 100 % of profiles). In the following, we present the data only for the first four of these platforms, as GreatAuPair does not have a sufficient number of worker profiles to reliably calculate age distribution. Our results show that the childcare platforms consistently attract young workers, while our senior care platform is characterised by significantly older workers. These differences probably reflect the distinct life stages, skill expectations and time commitments associated with different forms of care work. Table 2 provides an overview of key age statistics across platforms, while Fig. 3 visualises the full age distributions. Together, they highlight substantial variation between the two care types.

Across childcare-oriented platforms, the age distribution is strongly right-skewed, indicating that childcare platform work is predominantly taken up by younger individuals. On Babysitting24, workers have a median age of 25 and a mean age of 28.8 (standard deviation 11.8). Similarly, Babysits shows an even younger profile, with a median age of 21 and a mean age of 24.5, while Care.com's childcare workers are somewhat older on average (median 29, mean 34.0).

Many workers on childcare platforms are under 25 years old. This is particularly pronounced on Babysits, where 66.5 % of workers are younger than 25, followed by Babysitting24 with 48.2 %. Even on Care.com, which skews older, 33.1 % of childcare workers are under 25. These figures indicate that platformised childcare work is largely sought out by adolescents and young adults.

The lower bound of the observed age distributions corresponds closely to the platforms' minimum age requirements, which are typically set at 15 or 18. However, manual inspection of worker profiles reveals that some users appear to be younger than the officially permitted age, suggesting that age restrictions are occasionally bypassed by entering the minimum allowed age while disclosing a lower age in the profile description. As a result, the observed age distributions are effectively left-truncated at the platform minimum, likely inflating the number of workers reported at exactly 15 or 18 years of age.

In contrast, Seniorservice24, which focuses on senior care, attracts a significantly older workforce. Workers on this platform have a median age of 43 and a mean age of 42.9 (standard deviation 13.5), with only 8.3 % under 25, highlighting a stark contrast to childcare platforms. This age profile reflects how care labour markets are segmented by age and life stage, assigning different forms of care work to distinct worker demographics.

Care work is widely documented as a highly gendered form of labour, with both childcare and senior care predominantly performed by women (Schwiter and Steiner, 2020). In this section, we explore the extent to which platform-mediated care work mirrors this pattern. As described in Sect. 3.3.2, we infer gender from users' names and thus are limited to a binary gender categorisation.

Across all platforms for which gender could be inferred, we found that care work is overwhelmingly feminised. On the worker side, women constitute the vast majority of profiles across all platforms, ranging from 78.8 % on Seniorservice24 to 88.3 % on Care.com and Tagesmutter Verein Zürich, respectively. Childcare platforms in particular exhibit consistently high shares of female workers, while male workers account for only a small minority of profiles, typically between 3 % and 10 %.

This gender inequality is also visible on the demand side of the care platforms. For platforms where customers' names are available, which is only Babysits.ch and GreatAuPair,

Due to the small number of job profiles on GreatAuPair, we do not consider these data for our interpretation.

Taken together, these findings indicate that platformisation does not neutralise the gender inequality in care work. Instead, care platforms mirror highly gendered divisions of labour on both the supply and demand side. While this gender inequality is so normalised in society that it would probably be hard to build a platform where these inequalities do not occur, some platforms subtly reinforce this gender division. MisGrosi provides a particularly illustrative example: here, the platform name strongly suggests that the target group on the workers' side is older women – which presumably shapes who creates a profile on the platform. Therefore, gendered participation is not only driven by societal expectations but also reinforced through platforms' cues that signal what an appropriate care provider's gender would be. Other examples are the pictures of female care workers on the entry pages of Babysitting24, Care.com or Seniorservice24. This highlights how platforms actively reinforce gendered divisions of labour through naming, the pictures they use in their advertising and how they frame the platform.