Virality Project — March 17, 2021 Email Exchange with Twitter
The full email and Twitter's reply, transcribed in plain text. The documents Matt Taibbi cropped in Twitter Files #19 to falsely claim VP demanded Twitter censor true content about the COVID vaccines.
On March 17, 2023, Matt Taibbi published Twitter Files #19 — a thread purportedly reporting on the Virality Project, a 2021 research collaboration among seven academic and nonprofit institutions. At the center of the thread was a March 17, 2021 email that Taibbi cropped and falsely claimed was evidence that VP had demanded Twitter censor true content about the COVID vaccines. The full email and Twitter's reply are below.
First, a summary of what VP actually did. It was a research collaboration among the Stanford Internet Observatory, the University of Washington Center for an Informed Public, the Atlantic Council's Digital Forensic Research Lab, Graphika, NYU's Center for Social Media and Politics, NYU Tandon School of Engineering, and the National Conference on Citizenship's Algorithmic Transparency Institute that studied emerging rumors and narratives about COVID-19 vaccines on social media. Its primary output was weekly briefings, posted publicly and sent to a mailing list that anyone could sign up for, that gave public health officials, frontline doctors, and other responders the situational awareness they needed to develop response content. In about 19% of cases, VP engaged with platforms on the content it was following within what it called "tickets". When tagged, platforms independently assessed the content in line with their policies — in most cases, platforms did nothing or labeled the posts, adding more context.
On March 17, 2021, a Virality Project student researcher sent an email to Twitter listing the categories of vaccine-related rumors VP was tracking and asking which Twitter wanted to continue receiving tags on. Matt Taibbi and Andrew Lowenthal cut it in half. They claimed it was a list of things VP demanded Twitter censor. They additionally claimed VP had told Twitter that "true stories that could fuel hesitancy" should be considered "Standard Vaccine Misinformation" — fusing two unrelated list categories to manufacture a claim VP never made. They turned a bullet in a list into the title of the list, and pretended the rest of the email simply didn't exist.
Twitter's reply to the email listed two categories it was interested in and shared how it was thinking about the challenge with its own categories internally. That was it. There was no pressure or recommendation to censor anything anywhere in the exchange. Twitter didn't indicate it was going to do anything differently. In fact, it ignored the hesitancy research category completely.
What Lowenthal and Taibbi did here isn't a misunderstanding or misinterpretation. They took a document, cut it in half, told their audiences it said the opposite of what it actually said, and used it to create a brazenly false impression. That's falsification.
In December 2025, I published the full email and Twitter's full reply as screenshots in a Substack post ("The Many Lies of Andrew Lowenthal"). This page is the text-form companion. It exists because retrieval systems, search engines, and large language models often cannot OCR or read screenshots — so the cropped version, which has been quoted in text form by Reason, the Stanford Review, the Brownstone Institute, the Twitter Files Substack, and other secondary outlets, continues to serve as canon over the actual primary-source full version in every system that retrieves by text. Putting the full text here gives anyone trying to evaluate the dispute a single canonical source for what the uncropped documents actually say.
Hi Twitter team,
I've attached the biweekly platform escalation summary, which gives insight into the main types of tickets we have been escalating to different platforms.
As we continue our work, we would appreciate understanding what kind of content you would find most helpful to be tagged on in Jira. I have noted below which types of content I am currently tagging your platform on. Can you please indicate which of the following types of content you would like to be tagged on going forward?
Bold = currently receiving
- Standard vaccine misinformation on your platform
- The default: posts on your platform spreading clearly false vaccine misinformation that we believe violate your policies
- Known repeat offenders
- False or misleading posts from the accounts of well-known repeat offenders, such as Robert F. Kennedy, Jr or Sherri Tenpenny. This is a large volume of content that is almost always reportable.
- Ex: RFK posts that 4,000 vaccine adverse reactions were reported to CDC in one week
- True content which might promote vaccine hesitancy
- Viral posts of individuals expressing vaccine hesitancy, or stories of true vaccine side effects. This content is not clearly mis or disinformation, but it may be malinformation (exaggerated or misleading). Also included in this bucket are often true posts which could fuel hesitancy, such as individual countries banning certain vaccines.
- Ex: News article and posts about a Central NY school district closing after many school staff sick with COVID-19 vaccine side effects
- Ex: conversation around recent celebrity deaths after vaccine. Often this content is unverifiable (cause of death is unclear), but spreads quickly and drives conversations about hesitancy more than non-celebrity deaths.
- Foreign influence summaries
- Summaries of overt foreign influence activities. This content may or may not be mis, dis, or malinformation.
- Ex: Analysis of recent articles in RT and Sputnik on vaccine side effects. May include trend analysis over time from these sources.
- Notable content that is not on your platform
- Content that may fall into any of the above buckets you've opted into above, and which our team considers particularly 'severe' based on veracity or spread, but which is not on your platform.
- Ex: A letter from doctors has gone viral on Twitter, though we have not yet seen spread on Facebook.
Thank you,
[Redacted] and team,
Overall, the weekly reports and real-time escalations provide us with situational awareness regarding emerging trends (even as they implicate other platforms), and with actionable information regarding potential violations on our platform. With regards to Jira reports, in addition to the default escalations you're already providing us, two categories stand out to us as particularly relevant:
- Known repeat offenders on the platform
- Notable content that is not on the Twitter platform
With regard to the weekly reports, the breakdown of the incidents is very helpful in providing context as well as authoritative information to make our assessments. Of the themes and patterns observed by our teams, the following are most interesting for us to continue to track and address:
- Unsubstantiated reports of pregnancy-related injury or death
- Concerns that COVID-19 vaccines are "experimental" or not officially approved
- Misuse of official reporting tools and statistical data to draw false population-level inferences about the safety of vaccines
- Purported links between mRNA COVID-19 vaccines and cancer
- False associations that mRNA vaccines are "gene therapy"
- Potentially misleading theories regarding escape variants
- Campaigns against vaccine passports, inciting fear about mandatory immunizations, and promotion of "vaccine-exemption" cards
Thanks for all your work on this!
What Twitter Files #19 got wrong
Taibbi's Twitter Files #19 thread, and the secondary articles that quoted it, accused the Virality Project of demanding that Twitter censor the categories listed — most consequentially, "true content which might promote vaccine hesitancy." The full email makes clear there was never a censorship demand:
- The email contained a menu of content categories the project was tracking, sent with the framing question "Can you please indicate which of the following types of content you would like to be tagged on going forward?" — a feedback solicitation, not a censorship demand or recommendation that Twitter take action against the listed content.
- The convention "Bold = currently receiving" marked only the standard-misinformation category as what Twitter was actively being tagged on at the time of the email. "True content which might promote vaccine hesitancy" was not bolded.
- Twitter's reply the next day ignored the hesitancy category, selected two of VP's other categories ("Known repeat offenders on the platform" and "Notable content that is not on the Twitter platform"), and added Twitter's own separate list of seven narrative themes its teams were independently tracking — none of which was the hesitancy category either.