Nicole Froio Prism
As AI chatbots routinely share misinformation about abortion, advocates are working to make their websites more visible to abortion-seekers who use AI as a search engine
An estimated 2.4 million Americans turn to the internet each year in search of abortion care, according to Re/Imagination Lab, an organization that helps people find reproductive health information and care online. The accuracy of the information they encounter is increasingly shaped by artificial intelligence.
Leaders of abortion access organizations told Prism that over the last 12 months, there has been an uptick in AI chatbot summaries directing people seeking information about abortion to their organization’s websites. Previously, such organizations relied on search engine optimization (SEO) to improve their website’s visibility and ranking on search engines such as Google or DuckDuckGo. Until recently, they were not worried about how large language models (LLMs) might scrape and summarize online information about their work.
But a lot has changed in just a few short years.
This rapid shift in the information landscape has pushed abortion access organizations to reformulate their approach to online discoverability. At the same time, movement leadership, researchers, and organizers told Prism that they are also deeply concerned about AI’s proliferation of false or misleading information and politically driven information summaries. An anti-abortion political landscape that coalesces with a new age of information access has also created new concerns about privacy, they said.
“A real disruption”
For the last 25 years, search engines have shaped internet discoverability and information access, directing users to websites that are relevant to their queries. People in need of abortion care have historically relied on this imperfect model, which ranked responses according to ads and SEO strategies. But in recent years, Big Tech companies pushing generative AI have shifted the status quo by routing people to their chatbots, which take information scraped from the web and present it conversationally.
A University of Pennsylvania study published in June tested OpenAI’s ChatGPT-4o, Google’s Gemini 2.0 Flash, and Perplexity’s Sonar and found just 78% accuracy in the chatbots’ responses to a range of scenarios “involving different types of abortion access and restrictions.” The study, conducted by researchers Ro Encarnación, Christen Hammock Jones, and Danaé Metaxa, also revealed that in states with restrictive abortion policies, the rates of accuracy fell to 44%, Metaxa told said.
“I think [78% accuracy] is not high enough in a high-stake setting like this one,” Metaxa told Prism.
While Metaxa wasn’t able to give a definitive reason for the large drop in accuracy in abortion-restrictive states, they speculated that the tested chatbots draw from a wide array of conflicting and unclear information. Further complicating the information landscape is shifting news coverage and new legislation, but it’s impossible to know for sure without more information, they said.
“We would know if these LLM providers would give us access to the data that they’re using to generate these responses,” said Encarnación. “You can sort of understand where the data is coming from based on the citations that they’re pulling from.”
The researchers also noticed an alarming and sudden drop in web traffic to the websites of organizations focused on abortion care access. Metaxa said the business model of the internet has historically relied on people directly visiting websites. In the case of these organizations, site visits often translated to donations.
The increase in information access being mediated through AI could result in a real disruption to the business model that lets [abortion] resources stay afloat.Danaé Metaxa, University of Pennsylvania researcher
“There’s a real understudied risk that the increase in information access being mediated through AI could result in a real disruption to the business model that lets these resources stay afloat,” Metaxa said.
Rebecca Nell, founder and director of I Need An A, a website that helps people find abortion providers, and Amy Merrill, the founder of Plan C, an organization that provides abortion pills in the mail, told Prism that the online search landscape has radically changed over the last 12 months.
Nell and Merrill have been working together to understand the effects of AI on the abortion information landscape. Both organizations have seen increased traffic from ChatGPT and Google Search’s AI overview.
“[Plan C experienced] a drop-off last spring from Google search results changing and AI summaries being implemented, but higher search impressions overall this past year due to Plan C showing up in search summaries,” Merrill explained in an email. “We’ve also seen a similar recent increase in ChatGPT referral traffic.”
According to Nell, AI summary traffic to I Need An A increased by 2,000% in May. “People are using these AI tools to ask abortion questions,” Nell said.
The Canadian abortion pills provider Women on Web has also noticed a shift in traffic. Executive Director Venny Ala-Siurua told Prism that misinformation, censorship, and lack of search engine discoverability have been an issue for the organization since the pandemic. AI is the latest frontier of the same problem, she said.
“We’ve done a lot of work on our website just to optimize it, so we are able to stay visible knowing that all this misinformation is proliferating, and it’s very well-resourced,” Ala-Siurua said. “[Now] we are seeing a lot of people using ChatGPT and finding us and commenting how the support they received was very useful and compassionate even. But we know that’s just what we see. It’s a completely biased sample.”
Women on Web is working to make the information on its website more scrapable for AI so its visibility increases. “We’re trying to get ahead of the curve, but it’s admittedly moving so fast that I don’t know if we are going to be able to do that,” she said. “But that’s the only way we can survive. The service depends on a free and open internet and staying visible.”
Before the rise of AI in search, SEO hadn’t been great for abortion-seekers either, said Shantha Susman, executive director of Re/Imagination Lab. According to Susman, wading through 10 or 20 links when seeking abortion information is not “an ideal experience for someone who has never thought about abortion before.” The shift toward AI means that when people search “how do I get an abortion?” on Google, it’s imperative that “the answer they get is a good one,” Susman said, adding that 64% of U.S. girls ages 13 to 17 have reported using an AI chatbot.
In May, TechCrunch reported that instead of returning a simple list of links, Google Search would instead drop users into an “AI-powered interactive experience.” At Google’s annual developer conference, the tech giant announced its chatbot AI agent Gemini was being integrated into its search platform and responding to queries with an AI-generated summary and conversational follow-up questions. The shift means that web searches will increasingly be performed by AI agents, and internet users will be demotivated from clicking on external links found below the AI summary, according to TechCrunch.
During tests and other research Re/Imagination Lab has conducted in partnership with Brown University, Susman said the accuracy of AI summaries has not been high.
“Preliminary research for our internal use shows that answers often mention crisis pregnancy centers as options,” she said, referring to anti-choice nonprofit organizations that falsely present themselves as clinics where people can access abortion care. “They do not tell somebody in Texas that they have the ability to order pills online or travel for care.”
The executive director said that AI summaries also routinely include outdated information, mentioning clinics that no longer offer abortion care, for example, or that have moved out of state. “This puts people looking for abortions at a supreme disadvantage,” Susman told Prism.
The University of Pennsylvania researchers also found that crisis pregnancy centers (CPCs), which are notorious for spreading abortion misinformation, were cited as sources in AI-generated summaries.
“This is something we are really concerned about,” Encarnación said. “What’s the prioritization process for choosing what gets surfaced to users? Because we know this kind of information could get pulled in.”
In August 2022, Google announced that it would clearly label CPCs so as to not mislead people searching for abortion information. However, since shifting its search engine system to AI, the company has made no similar assurances. Other AI chatbot providers such as OpenAI, Anthropic, and Perplexity have not made public statements regarding their approach to abortion information. Prism reached out to these tech companies for comment, but received no response.
“People who have to get abortions have to work extra hard to get through the misinformation, the bad answers, to click through multiple different websites to call through a variety of different clinics or funds, or try to make sense of online pill providers,” Susman said. “What’s at stake is overburdening people who are already trying to make a time-sensitive step toward whatever their option is in a highly regulated landscape where their access is much more limited than it should be.”
AI is here to stay. Now what?
In the face of growing AI use among pregnant people seeking care, abortion access organizations are attempting to communicate with AI companies to improve the information their chatbots share.
Nell and Merrill told Prism that they have offered to help tech companies prioritize important, accurate information. Though they declined to reveal which companies, Merrill said they’re “in talks” with the top AI chatbot providers in the industry.
“We know that AI tools are synthesizing these answers from things that exist on high-authority websites,” Merrill explained. “What we’re doing is understanding what websites those are across the movement, what websites are being pulled in as citations and references for answers regularly that are trusted and evidence-based, and then working with those sites to figure out ways that language can be more nuanced and more complete so the answers can be synthesized from the best possible on-page content available.”
Other organizations are trying a different approach to get a sense of how AI prioritizes information. Women on Web is interviewing people who sought abortion access information using AI. “We want to get firsthand information on what they asked, what happened, [and] how they got to us,” Ala-Siurua said.
Tech companies’ lack of transparency is one issue. Another problem is the public’s general lack of AI literacy.
Many people assume that when they read something from an AI answer engine, it’s giving them a consensus. But when you’re asking about abortion, there’s so much misinformation flying around.Garin Marschall, co-founder of PatientForward
“Many people assume that when they read something from an AI answer engine, it’s giving them a consensus,” said Garin Marschall, co-founder of PatientForward, an organization focused on abortion care in later pregnancy. “But when you’re asking about abortion, there’s so much misinformation flying around.”
Garin is also the co-director of Abortion Policy API, which provides developers up-to-date information on state abortion policies that can be integrated into online abortion resources from Planned Parenthood, Charley the Chatbot, Abortion Finder, and I Need An A, among others. According to Marschall, Abortion Policy API’s dataset is manually updated as experts in the abortion access field note policy changes in a group Slack channel. Once a consensus is reached, the dataset is published within 15 minutes, automatically updating the websites that use it as a source of information.
Because of how quickly the U.S. abortion landscape changes, this approach ensures the most up-to-date and accurate information. However, the application programming interface isn’t readable or scrapable by AI, which the organization soon plans to change. Still, Marschall worries that AI chatbots will “hallucinate” wrong information based on existing abortion misinformation online. It’s a serious problem, given that half of U.S. adults now use AI chatbots, according to Pew Research Center.
“The key is to figure out how we can make sure that information is accurate, up to date, and relevant to their question and context,” Marschall said.
Abortion later in pregnancy is just one example of a subject riddled with misinformation online. Many people in the U.S. are totally unaware that abortion after 21 weeks is possible.
“We need to start making sure all of that information is dumped online in a very plain, readable way, so we can make sure it’s being fed into the AI because sometimes [that’s] the first and only call people make,” Marschall said.
As part of these efforts, top players in the abortion healthcare field started an AI working group last year that includes leadership from Plan C, I Need An A, PatientForward, Power to Decide (formerly Abortion Finder), Planned Parenthood, the National Abortion Federation, Charley, the Abortion Care Network, and Re/Imagination Lab.
As part of one experiment, the working group put together a small, stripped-down resource online and tested whether it changed AI-generated answers. What they realized is that it would be far more useful to create an abortion resource website that is built for machines, not humans.
Susman said the experiment gave the working group insight into how AI labs are potentially training LLMs, which the group could use to create “a unified source of truth” that has “all of the information that an AI engine needs to answer a question related to abortion.”
The website the working group is building will be “stripped down,” “high authority,” and in a format that is “as easy as possible for an AI machine to scrape and to make sense of,” Susman said. This means the site will feature no graphics or videos.
“It has all of the databases of which providers are available, what the laws are state-by-state, where the CPCs are, and thus, how to avoid those fake clinics and caution people against them in their answers,” Susman explained. “And then a protocol for how machines should use this information to answer the questions that people have.”
The website is still being built and has no launch date yet.
Protecting privacy
Abortion advocates who spoke to Prism agreed that there are security risks associated with using AI tools to seek abortion care, though many see the use of AI as an extension of already existing privacy issues online.
“There are obviously steps that people could take to protect themselves and protect their privacy,” Marschall said, citing the use of incognito tabs and virtual private networks (VPNs) that encrypt a user’s internet traffic and route it through a secure, remote server.
He acknowledged that most people don’t take these precautions, which is why abortion access groups are also urging tech companies not to cooperate with law enforcement unless they “absolutely have to—and not hold data that they don’t need to hold,” Marschall said.
Data security and privacy issues are another major knowledge gap for most people online, but Nell said that ultimately it’s up to Big Tech to protect users’ data.
Now that Merrill and others are in conversation with tech companies, she said abortion access groups are also thinking through recommendations for making their AI policies and practices more transparent and “privacy-forward” to safeguard people’s data.
Prism contacted OpenAI, Anthropic, and Google for comment regarding their data policies in relation to abortion criminalization. None responded.
“It’s really, really important that we do more education for folks around their digital footprint and digital privacy,” Nell said. “At the end of the day, we as consumers need to hold these companies more accountable.”
Nicole Froio is a writer and researcher currently based in Rio de Janeiro in Brazil. She has a doctorate in Women’s Studies from the University of York. She writes about gender in pop culture, social movements,
Prism investigates systems of power through the experiences of communities most affected by injustice, connecting their stories to the forces shaping their lives. Explore its independent reporting.
Editor’s Note: At a moment when the once vaunted model of responsible journalism is overwhelmingly the play thing of self-serving billionaires and their corporate scribes, alternatives of integrity are desperately needed, and ScheerPost is one of them. Please support our independent journalism by contributing to our online donation platform, Network for Good, or send a check to our new PO Box. We can’t thank you enough, and promise to keep bringing you this kind of vital news.
You can also make a donation to our PayPal or subscribe to our Patreon.