Disinformation and Social Media correspondent Marianna Spring set up a series of social media accounts to investigate what voters are being recommended online at a turbulent time for US politics. After over two months of running the profiles, this is what she found.
I open up Britney's Instagram and click on an account that's been recommended in her feed. I'm greeted by a meme falsely declaring that President Joe Biden never really won the 2020 election, and several others targeting named female politicians with misogynistic comments and abusive language.
Britney is one of five profiles I've created to track what US voters can be recommended and exposed to online ahead of the midterm elections. While social media sites say they are committed to tackling disinformation and hate on their platforms ahead of the poll, for my undercover voters, misleading and violent posts appear to have only increased in recent weeks.
My five voters were created to represent views from across the US political spectrum, based on data gathered by the Pew Research Centre. I gave each of them a profile on Instagram, Facebook, TikTok, YouTube and Twitter, with names and computer-generated photos.
They are:
For all of them, I followed additional accounts when they were suggested, as well as keeping up their original interests.
While these profiles can't offer an exhaustive insight into what every US voter could be seeing - and they don't have friends or followers - they do give us a snapshot of what voters across the political spectrum are being exposed to.
The Institute for Strategic Dialogue (ISD), which tracks extremism and disinformation worldwide, says that election conspiracy claims and abusive language aimed at politicians online have intensified in the run-up to the midterms.
As I've checked into each of my undercover voters' accounts over the past two months, it's Britney's that have been most frequently exposed to violent and misleading content.
When setting up her account, I liked pages and accounts that supported Mr Trump, opposed mandatory vaccination and questioned the motives of billionaires. These topics appear to have been a gateway to more extreme content, when compared with the other undercover voters.
Larry was exposed to posts featuring misleading claims about the 2020 election and abusive language directed at politicians, while Emma was also recommended several pages using abusive language aimed at supporters of Donald Trump and Supreme Court Justices, but nothing on the scale of what Britney was recommended.
When I logged onto Instagram in particular, I found she was recommended more and more accounts that made false claims about fraudulent voting and denied that President Biden won the election.
They regularly featured hashtags like #Trumpwon in their profile descriptions and on their posts, and shared memes about the riots on the Capitol claiming that "January 6th wasn't an insurrection".
Britney's profile was also shown TikTok videos from accounts promoting conspiracy theories that the election was rigged, also repeating the phrase "Trump Won".
Last week, Instagram showed her conspiracy theories contradicting the police account of what happened when Paul Pelosi, the husband of US House of Representatives Speaker Nancy Pelosi, was violently attacked in their home. Posts also made light of the violence.
Finally, pages recommended to Britney's Instagram account also featured posts talking about female politicians in abusive and misogynistic terms.
The most frequent targets were Nancy Pelosi, Kamala Harris, Alexandria Ocasio-Cortez, Michelle Obama and Hillary Clinton. There were comments about them performing sex acts, and slurs about their appearances, alongside criticism of their politics. While posts often spoke about Joe Biden and Donald Trump in hateful terms, they were not subject to the same sexualised language.
This language was unique to Britney's accounts. On Facebook, Emma was recommended pages that promoted naming and shaming racists and Trump supporters, but the rhetoric had not escalated in the same way.
Gabriela was increasingly recommended right-leaning content on social media, often about inflation and the cost of living crisis, but it stopped short of the more extreme disinformation and hate that Britney encountered.
Larry and Michael were targeted more frequently by official campaign adverts and content from the political parties themselves, especially on sites like YouTube. For Larry, the messaging was focused on crime, inflation and immigration. For Michael, the adverts and posts talked about abortion, climate change and education.
As well as noting the uptick in hate ahead of the midterms, the Institute for Strategic Dialogue says disinformation online has focused on conspiracies about the Biden administration, as well as the issue of voting in specific US states.
"That, when combined with social media product features and failed policies, is giving permission to more hate and vitriol," said Jiore Craig, the ISD's Head of Elections and Digital Integrity.
In particular, the ISD found election conspiracy theories in short-form video formats across several platforms. Social media companies have "inconsistent and insufficient moderation policies" to deal with this kind of content, its research concludes.
The research also expresses concerns that the major social media companies are not taking election denials seriously, although they have made various commitments to tackling disinformation and hate ahead of the midterms.
In August, Meta - which owns Facebook and Instagram - said it would devote "hundreds of people across more than 40 teams" to ensure the security and safety of the midterms, and apply "learnings" from the past election.
Meta told the BBC that it has robust measures in place to combat misinformation, including partnerships with 10 fact-checking organisations in the US.
It also said it has "clear policies about what is and isn't allowed on our platforms" and "continually reviews content" to see if it violates policies.
TikTok also announced its own "commitment to election integrity" ahead of the vote, saying it had partnered with fact-checking organisations. It told the BBC it took its "responsibility to protect the integrity of our platform and elections with utmost seriousness".
Populist Right Britney, Progressive Left Emma and even Apolitical Gabriela have been recommended increasingly polarised content on Twitter over the past two months, particularly on immigration, crime and climate change.
Britney's account was also shown a conspiracy theory shared by Elon Musk about the attack on Mr Pelosi - Mr Musk later deleted his tweet.
There are concerns that lay-offs and potential changes to moderation policies on Twitter, which Mr Musk now owns, could also lead to an increase in online disinformation just before voters head to the polls.
Several of the Instagram accounts that Britney has been recommended promoting claims that Trump really won the 2020 election have also praised Mr Musk for buying Twitter and referenced their plans to return to the site.
Twitter says that while it has lost employees, their "core moderation capabilities remain in place".
Biden and Trump to make final pitches on eve of midterms
Ukraine is reason to act fast on climate - PM
US confirms 'communications' with Kremlin
What did people ditch tweets for? Take our timed quiz…
'We'd never seen this much water' - flood survivors
In the name of atheism: The case of Mubarak Bala. Video
The Indian-American 'helping' Elon Musk run Twitter
Why an old train could point to a clean energy future
How US voters get shown hate online
Looking for clues in video of forgotten massacre. Video
The agony of not knowing, as Mariupol mass burial sites grow
Twitter users jump to Mastodon - but what is it?
The young US men choosing vasectomies
Istanbul's beloved 'brothel dessert'
The world's greatest unpaid debt?
© 2022 BBC. The BBC is not responsible for the content of external sites. Read about our approach to external linking.
