Googling for Abortion: How Google Mediates Access to Abortion Services— and Why It Matters Now More Than Ever
Access to abortion in the United States is now hanging by a thread. But how do most people find abortion services? By searching for them online. In our study published earlier this year in the Journal of Quantitative Description: Social Media with Tatiana Gracyk, director of the Bioethics Program at the Cleveland State University and Ronald E. Robertson, a postdoctoral fellow at the Stanford Internet Observatory, we showed that the locations returned to Google’s search results reflect current inequalities in abortion access. If these inequalities are to drastically change, the information (and lack of thereof) returned to people living in states that may outlaw this medical service could further impact already stressful, life-changing decisions.
To understand what abortion search results look like from a diverse set of places in the US, we conducted searches from 467 counties that varied in terms of population size, economic status, proportion of urban v. rural living, and vote leaning in the previous Presidential election. We then searched Google using 10 different queries (search terms) that a woman looking for abortion services may use, such as “abortion clinic” and “pregnancy termination”. We were especially interested in the 3 location results one would often see within their results, and whether those locations were abortion clinics (ACs) or crisis pregnancy centers (CPCs) — places where women are often dissuaded from getting an abortion.
We found that the search results Google returned varied drastically between states. Washington DC, New Jersey, and Connecticut returned the most AC results, while Montana, North Dakota, and Alaska returned the fewest. The majority of results returned by the search engine were ACs (79.4%), followed by other clinics (9.2%), and CPCs (6.9%). (Below, black dots align with the right Y-axis and indicate the number of known ACs in the state.)
Plotting the average number of ACs and CPCs returned for a query in a state showed that the areas on the coasts returned larger numbers of AC results and fewer CPC results than inland. This is not a surprising finding, as the number of AC results returned for our searches correlated with the number of ACs in the state that the search was conducted in (obtained from ANSIRH database) at Spearman p = .62. This means that people in states with fewer local abortion resources may be shown fewer relevant abortion location results, making it less likely that such services are found. Note that this does not need to be the case, as the search engine could show abortion clinics that are further away, or even in neighboring states, if necessary.
But what else could explain this variation in search engine performance? We built a regression model to see if urbanization, population size, wealth, and political affiliation could be associated with the number of returned ACs (or CPCs). We also checked the various Targeted Regulation of Abortion Providers (TRAP) laws designed to put undue burden on the abortion clinics beyond what is necessary to ensure patient safety. We found that more AC (and fewer CPC) results were returned in places that have more ACs (as shown with correlation earlier), but also in areas that tend to vote Democratic, and those that have higher incomes. Both kinds of results were likely to be returned for those in rural areas. Some TRAP laws also have an effect, especially those governing the standards for facilities and service providers, after taking into account the above demographics. Although it is difficult to determine the direction of causation, the fact that, even after taking into consideration the number of ACs and other variables, the negative effect of income remains strong reflects the fact that those living in poverty may be more strongly affected by antiabortion regulations.
Other signals around the results may further affect the experience of the user. AC results usually get lower ratings as businesses on Google Maps (median of 3.7 stars, compared to 4.3 for CPCs) and more reviews (median of 22 compared to 9 for CPCs), possibly pointing to adversarial review posting. Further, when a crisis pregnancy center result was returned, it was the closest result to the location from which the query was sent 76% of the time (this is likely related to the fact that, in our estimate, there are roughly 4,000 CPCs and only 800 ACs in the United States).
However, there are encouraging signs that Google is directing its users to the closest possible abortion clinic. When we compared the address of the closest AC result Google returned to the addresses in a list of known ACs: 90% of the results were within 5 miles of the closest known location. Further, when the new BERT update (a neural network-based technique for natural language processing (NLP)) was introduced around October 2019, the proportion of ACs being returned increased, especially for locations with fewer known abortion resources — an interesting possible side effect of improvements in AI and language understanding. Finally, this study looks only at the location results, and more information may be available in the 10 “organic” links returned by the engine, as well as sponsored content and other kinds of content like news and videos, which sometimes appear at the top of the page.
Search and recommendation engines — including those built into the social networking sites — provide a window into both the web and our physical world, and may thus substantially constrain the scope of our actions. The unchecked crowdsourcing of business information by the platform may present another hurdle to its users, as we found several instances of locations we identified as CPCs to be inaccurately classified as “Abortion Clinic.” Yet, other hurdles are presented by the websites of the businesses themselves. During detailed labeling, we found that more than half of the CPC pages had detailed descriptions of abortion (on average 773 words long, possibly a sign of keyword stuffing that raises the page in search results ranking), and among those, only 63% had a disclaimer on that page noting that they do not perform or refer for abortions.
If Roe v. Wade is indeed overturned, the access to abortion services across the US will become even more unequal, possibly impacting those most vulnerable. Some states will become the sole providers of the service to neighbors spanning hundreds of miles. Funds helping women arrange for travel and to cover expenses associated with abortion services will become ever more important. In order to find these abortion providers, Google and other search engines must surface the relevant information, even if it is further away geographically, in order to allow those living in states whose governments would deprive them of this critical medical service. If the issues we have identified continue to persist in the search tools people use every day, this may further hinder access to whatever resources advocates are fighting to keep available.
Ideally, a collaboration between the tech giant’s teams and public health specialists such as the Guttmacher Institute would help identify quality concerns and address them. For instance, Instagram has teamed up with the National Eating Disorders Association (NEDA) to provide information about eating disorders, and now directs those searching for hashtags like #anorexia to help resources. During the height of the COVID-19 pandemic, Facebook and Apple provided valuable mobility data to monitor the lockdown situation around the world. It is time to acknowledge that abortion is a vital public health resource, and to provide the quality controls necessary to ensure that the hard work of abortion providers and their advocates does not remain trapped behind a geographical search engine bubble.