Skip to main content

Amazon’s Twitter ambassadors are blurring the line between fact and fiction

Amazon’s Twitter ambassadors are blurring the line between fact and fiction


How is this dark-arts PR campaign supposed to build trust in the company?

Share this story

Illustration by Alex Castro / The Verge

One of the big lessons of Amazon’s HQ2 debacle, at least from my perspective, was that the company can be weirdly tone-deaf. A purported nationwide search to find a home for a “second headquarters” looked, as soon as the company announced it had settled on New York and Washington, DC, as if it had been a ruse. New York pushed back, Amazon dropped its plans, and the entire episode has faded into memory.

But the public-relations apparatus responsible for managing Amazon’s relationship with our democracy has continued to undertake information operations designed to burnish its image. And this week, the world took notice.

Amazon has long faced criticism over the working conditions in its fulfillment centers, where workers strive to remain marginally more useful than the package-sorting robot coworkers that will someday replace them. In response to this — and amid growing fears at the company that its workers would unionize — the company last year created a series of Twitter accounts for its “fulfillment center ambassadors.”

The ambassadors tweeted about the plentiful bathroom breaks offered them by their employer with a false cheer common to hostages. In exchange for their testimony, the workers reportedly received one additional paid day off, and a $50 gift card. The tech world spent a day or two chuckling over this particular real-life Black Mirror episode, then moved on to the next.

And then, sometime this year, the accounts were handed over to ... other people. People of different ages and genders. People with different profile photos. Ambassadors who had once been hard-working grandmothers named Michelle were now college-aged workers named Rafael. As Jonah Engel Bromwich writes in the New York Times:

The accounts have provoked suspicion. In January, it appeared that the accounts had changed hands; one that had belonged to a “Leo” had changed its display name and handle to Ciera. A “Rick” had become a “James,” and a “Michelle” had transformed into a “Sarah.” (Critics of the account occasionally call them the “Borg,” a reference to an alien race in Star Trek who operate as a collective hive mind.)

Twitter users took notice, tweeting to ask the ambassadors whether they are actually robots. The ambassadors denied being robots, but the case still felt inclusive. What is the appropriate Voight-Kampff test for corporate propaganda accounts?

Bellingcat’s Aric Toler launched an investigation. He has found “53 Ambassador accounts so far, including 29 American accounts, five Spanish, seven German, four British, four French, two Polish, and three Italian ‘ambassadors.’” Unlike most working Joes on Twitter, they uniformly tweet from Sprinklr, a paid enterprise marketing service. Toler concludes:

It’s hard to imagine how and why Amazon decided that such volunteer brand ambassadors would be a good idea — especially considering they almost all write the exact in the same manner and use the same hashtags and similar photos in their tweets.

While there may be different faces behind these accounts, it is hard to tell them apart, and their activities all seem to be thematically orchestrated from a corporate office. In reviewing ambassador accounts, only a few English-language participants stood out as having any personality and not using near-perfect capitalization and punctuation.

Amazon, for its part, testified to the legitimacy of its ambassadors. It told BuzzFeed:

“These accounts are run by FC employees who understand what it’s actually like to work in our FCs,” a spokesperson told BuzzFeed News by email. When shown Desiree’s account and asked whether she’s a worker who’s writing and posting her own tweets, the spokesperson said, “That’s correct. Desiree is the one writing and sending the tweets.”

There’s a long history of companies mounting astroturf campaigns. But as Rani Molla points out in Recode, the use of Twitter to advance this sort of message feels relatively new:

Of course, corporate-sponsored anti-union propaganda is not new. Famously, then B-movie star and Screen Actors Guild union president Ronald Reagan worked as a “traveling ambassador” for General Electric, visiting plants across the country to extol a free-market system.

But compared with older anti-union stunts, “The fulfillment center tweets are more interesting because it plays on something new: the perceived authenticity of Twitter versus older kinds of bottom-up media,” Louis Hyman, a Cornell professor and author of Temp: How American Work, American Business, and the American Dream Became Temporary, told Recode. “No one took it as authentic as [Amazon warehouse employees posting on] Twitter. That’s what’s at stake here. Those tweets can help control who to believe.”

If you only followed this story on Twitter, you could be forgiven for believing that Amazon had mounted a straight-up disinformation campaign. It seems more likely that Amazon is recruiting real workers to offer spin on the company’s behalf — but the move feels phony at best, and exploitative to boot. I like how researcher Jonathan Albright put it to the Times:

Albright, the director of the Digital Forensics Initiative at the Tow Center for Digital Journalism, said that the messages the accounts were spreading did not rise to the level of disinformation. But he said the practice could be deceptive in theory and had the potential to involve components of disinformation. He said that he preferred to refer to the campaign by what it was, calling it “dark art PR.”

And if it were happening on Facebook, incidentally, it seems like the sort of thing the social network would call “coordinated inauthentic behavior” — which would be grounds for kicking all those ambassadors off the platform.

Thanks to various state-sponsored actors, we already have enough problems sorting out fact from fiction on social networks. To see Amazon mounting its own dark-arts PR campaign on Twitter — turning employees into paid flacks, without disclosing that they are being compensated as such — feels like a grim new development in our information sphere. At the same time, as with HQ2, the ruse fell apart under the lightest scrutiny. Perhaps Amazon’s public-relations campaigns would go better if it tried waging them out in the open.


The algorithms that detect hate speech online are biased against black people

Important story from Shirin Ghaffary on new research:

Two new studies show that AI trained to identify hate speech may actually end up amplifying racial bias. In one study, researchers found that leading AI models for processing hate speech were one-and-a-half times more likely to flag tweets as offensive or hateful when they were written by African Americans, and 2.2 times more likely to flag tweets written in African American English (which is commonly spoken by black people in the US). Another study found similar widespread evidence of racial bias against black speech in five widely used academic data sets for studying hate speech that totaled around 155,800 Twitter posts.

This is in large part because what is considered offensive depends on social context. Terms that are slurs when used in some settings — like the “n-word” or “queer” — may not be in others. But algorithms — and content moderators who grade the test data that teaches these algorithms how to do their job — don’t usually know the context of the comments they’re reviewing.

Teens exposed to highly charged political ads on Facebook and Instagram

Rowland Manthorpe reports that the UK’s conservative party is going after teenagers on Facebook:

Sky News has seen 208 political ads shown to 13 to 17-year-olds on Facebook and Facebook-owned Instagram, where advertisers can target campaigns according to age.

The majority of the ads came from the Conservatives, which showed 102 ads to teenagers, mostly featuring Boris Johnson.

Facebook isn’t ready for 2020

A new study has worrisome implications for the US presidential election. Trevor Davis, Matthew Hindman, and Steven Livingston write:

What we found were bizarre patterns favoring the far-right, anti-immigrant party Alternative fur Deutschland. AfD is a medium-sized party in Germany that polled between 10 and 15 percent in the months leading up to the E.U. elections.

On Facebook, though, AfD dominated to an astonishing degree. AfD pages received 86 percent of total shares and 75 percent of all comments — four times the comments, and about six times the shares, of all other political parties combined. It is likely no other political party has ever dominated Facebook during a free election as thoroughly as AfD.

No, Ilhan Omar has not been arrested 23 times

And speaking of bad stuff getting traction on Facebook, from Daniel Funke:

A viral photo on Facebook claims that Ilhan Omar has been arrested 23 times and is the daughter of a Somali terrorist.

She was arrested once and the charges were dropped. Minnesota court records show that Omar has had 24 traffic violations in the past 10 years, not arrests. We could not find any evidence that either of Omar’s parents were involved in a terrorist organization in Somalia.

You can now report a suspicious Instagram post and expect a certified U.S. fact-checker to verify it

This seems good. From Cristina Tardáguila:

Facebook announced today it is expanding its Third Party Fact-Checking Program (3PFC) to the photo- and video-sharing social network it bought seven years ago. The technical rollout starts today in the United States and should take two weeks to reach all international users.

To report suspicious content, users will only have to click on the three dots in the upper right corner of each Instagram post, choose “it’s inappropriate” and then “false information.” Then the posts will be reviewed by IFCN members, who are already working with 3PFC in more than 30 countries.

Google employees ‘refuse to be complicit’ in border agency cloud contract

Colin Lecher has your internal Google activism of the day:

Google employees are demanding that the company not bid on a cloud computing contract with US Customs and Border Protection in the latest act of protest inside the tech industry.

In a petition circulated today inside Google and on Medium, a group of employees said immigration officials are “perpetrating a system of abuse and malign neglect” at the border. The employees point to the Trump administration’s family separation policy and the recent deaths of children in immigration officials’ custody. “These abuses are illegal under international human rights law, and immoral by any standard,” the petition reads. In the hours after it was released, hundreds of employees added their signatures to the petition.

Top Takes: Suspected Russian Intelligence Operation

The Digital Forensic Research Lab reports on the activity of a Russian information operation discovered on Facebook earlier this year:

The operation was strongly reminiscent of the Soviet-era “Operation Infektion” that accused the United States of creating the AIDS virus. That operation planted the fake story in distant media before amplifying it through Soviet channels: it ultimately spread through genuine news media around the world and was often reported as fact. The latest operation — which the DFRLab has dubbed “Secondary Infektion” — used a similar technique by planting false stories on the far reaches of the internet before amplifying them with Facebook accounts run from Russia.

The operation’s goal appears to have been to divide, discredit, and distract Western countries. Some of its stories were calculated to inflame tensions between NATO allies, especially Germany and the United States, as well as Britain and the United States. Others appeared designed to stoke racial, religious, or political hatred, especially in Northern Ireland. Few posts gained traction, but one anti-immigrant story penetrated the German far right and continues to circulate online. It appears likely that the Russian operation fabricated the entire story, including its spurious “evidence.” This was a particularly disturbing case of weaponized hatred stemming from a foreign operation.

Bernie Sanders joined Twitch to reach people where they are

Bijan Stephen explores Sanders’ adventures in streaming:

So far, Sanders’ channel has mostly broadcast town halls and rallies; the candidate himself rarely shows up on streams. Even so, 70,000 people followed Sanders’ account on Twitch in the first 24 hours they were on the platform, and 20,000 people tuned in for the first Democratic debate on their channel, according to his communication staff.

No one’s exactly sure who came up with the idea, Miller-Lewis says. But internally, they’d been kicking around the idea for a live-streaming show, and Twitch was a natural place for that to live. It helped that Sanders thought it important that he had a show again. Sanders had one on public access cable back in the ’80s. In the ’90s, he was making videos in DC to send back to his constituents in Vermont. “I think this next stage with Twitch and live streaming is sort of the natural extension of his interest and focus on finding new ways to communicate with people and bring them into the political process,” Miller-Lewis says.

Less than Half of Google Searches Now Result in a Click

I can’t get over these stats from Rand Fishkin, which speak to various ongoing Google antitrust issues:

June (as shown at the top of this post) is when zero-click searches in browsers passed 50%, but the pie chart above shows that even before that, Google was sending a huge portion of search clicks to their own properties (~6% of queries and ~12% of clicks). Those properties include YouTube, Maps, Android, Google’s blog, subdomains of, and a dozen or so others (full list here).

Maybe Google’s websites are ranking exclusively because they’re the best result, but if Congress is asking questions about whether a monopoly is potentially abusing its market dominance in one field to unfairly compete in another, I’ve got something else they’ll want to see. It’s a chart of where searches happened on major web properties in Q2, and as you can see, there’s no competition

Do Tech Companies Really Need to Snoop Into Private Conversations to Improve Their A.I.?

April Glaser reports on ways that tech companies could use humans to improve voice transcriptions while reducing privacy concerns:

But contractors aren’t the only way to train a machine to get better at understanding us. There are other less privacy-invasive ways of improving A.I.’s ability to comprehend human language than human review. “Instead of sending out the exact piece for human transcription, you could create a way that has the same kind of noise or other acoustic features and have a human transcribe that so that you’re not divulging anything private of your users,” said Micha Breakstone, an expert in natural language processing and co-founder of Chorus, which builds A.I. for understanding conversations for sales teams.*

Breakstone means that a copy of a recording could be made to imitate the same sounds, inflections, and words so that the actual, potentially identifying recording isn’t being reviewed by strangers. It’s also possible to shift the voice or gender of the person talking in the recording to further protect their identity. Still, “at the end of the day, the answer is you don’t need to send those materials out to humans, but it’s much, much easier if you do,” Breakstone said.

Hong Kong protests: Police defend use of ‘disguised’ officers

Speaking of phonies, the BBC reports that Hong Kong is creating decoy protesters:

Hong Kong police have admitted deploying officers disguised as anti-government protesters during mass unrest that rocked the city on Sunday.

Some officers disguised themselves as “different characters”, a spokesman said, adding that the “decoy operation” had targeted “extreme violent rioters”.


How an Online Mob Created a Playbook for a Culture War

Killer package from the Times on the five-year anniversary of Gamergate.

Cloudflare warns investors that sites like 8chan are a risk to its business

The internet service company warned that its business could be materially affected by the culture war in which it has become an unwilling participant:

Internet services company Cloudflare is going public, and it describes the controversy over banning hateful websites — like 8chan and the Daily Stormer — as a potential risk factor. The company’s S-1 filing reports that objectionable sites could cause “significant adverse political, business, and reputational consequences.”

After ad denied, gay creators are fighting for systemic change at YouTube

Chris Knight and Celso Dulay are suing YouTube after their ad was rejected — they say for discriminatory reasons. Julia Alexander reports:

This “egregious” event is what led Knight and Dulay, who operate a YouTube news channel called GNews! that’s dedicated to covering LGBTQ issues, to launch a lawsuit against the company for discriminatory practices against LBGTQ+ people, who are “considered a protected group by California law,” Knight told The VergeThe Verge obtained an audio recording of the conversation between Dulay and two members of Google’s AdWords team, which confirmed Dulay and Knight’s statements. A YouTube spokesperson declined to comment on the recording.

During the call, Dulay spoke with both a customer service representative and their manager about running the ad. The manager told Dulay that because the video contained “any content about sexuality or anything like that,” it would “actually violate the policies of AdWords under ‘shocking content.’” When Dulay pressed the manager on whether that meant “content about gay people” in general, he was told yes.

Google Employee Writes Memo About ‘The Burden of Being Black at Google’

Lorenzo Franceschi-Bicchierai posts a letter from a recently departed Googler who experienced racism:

The memo, obtained by Motherboard, is titled “The Weight of Silence,” and argues that Google is lacking in diversity, and that some of its employees make racist or at least insensitive comments about minorities.

“Over the last 5 years I’ve heard co-workers spew hateful words about immigrants, boast unabashedly about gentrifying neighborhoods, mockingly imitate people who speak different languages, reject candidates of color without evidence because of ‘fit’ and so much more,” the employee wrote. “So, just as I’m doing with this essay, I ultimately resolved to break my silence. And though I eventually grew more comfortable using challenging moments to educate my co-workers, I never stopped feeling the burden of being black at Google. And the more insensitive comments weighed on me, the less safe I felt here—and the less capable I was of being my best self at work, or myself at all.”

TikTok Users Are Inventing Wild Theories to Explain Its Mysterious Algorithm

Caroline Haskins reports on TikTok users desperate to hack its opaque recommendation algorithms, who add useless hashtags to their posts as a form of prayer:

Probably half of the videos I see on TikTok include one of the following hashtags: #fyp, #foryou, or #foryoupage.

The hashtaggers’ theory is that if they use these tags in their captions, their posts are more likely to surface on more people’s For You pages. The For You page is TikTok’s recommendation feed, which is personalized to each user based on how that user interacts with videos on TikTok, according to the company.

There’s absolutely no proof that using these hashtags does anything, but it seems like they do.

TikTok’s best comedy duo is a loud man and his duck

Megan Farokhmanesh has your TikTok account of the day:

Balaskovitz adopted Jerry as a duckling about five years ago from a local farm. “He’s like a dog,” says Balaskovitz. Duck pastimes include car rides, eating McDonald’s french fries, napping in his own bedroom, and waiting at the front door for Balaskovitz to come outside. Balaskovitz will post videos featuring Jerry and a menagerie of pets — including birds, dogs, plunger-stealing ferrets, and sunglass-clad guinea pigs — but not all members of the animal brood are stars, and few garner the same sort of love and attention as Jerry. It’s normal nowadays to see animals in the spotlight. It’s rarer for said pet star to be a shower-loving, pool-hopping duck with an indifference to being yelled at.


Facebook movie ads will now include ticket and showtime details

Sure, why not.

Instagram revamps Boomerang, creates Layout for Stories and more

Jane Manchun Wong’s preview of new features coming to Instagram includes an emulator for new Boomerang modes. Unbelievably cool.

YouTube is testing a members-only videos feature

This seems like a logical and creator-friendly move:

After introducing its Memberships program in June 2018 to sell subscriptions for individual YouTube channels, YouTube has begun testing an option for channels to upload members-only videos, according to a document published to the platform’s support forum. A YouTube spokesperson did not provide a comment by press time.


Casey Newton on dismantling the platforms and taking Facebook’s cash

Hey that’s me! I talked to Columbia Journalism Review about platform stuff, including whether publishers should take Facebook’s money for its new news effort:

“I do think publishers should take the money here,” he said. “Social network ‘carriage fees’ have been a pet issue of mine for a while. Just as cable companies pay for access to high-quality channels, so, too, should social networks pay for access to high-quality journalism.” That kind of deal is a win-win-win, Newton said. “Publishers get money for journalism; readers get news they can trust; and Facebook gets a higher-quality news environment that can bolster our democracy while making the whole site more attractive for readers and advertisers.” As for those who are afraid that the money might vanish, that has always been a risk, said Newton. “Will the money disappear at some point? Probably. But that’s true of so many sources of income that journalists rely on already: ad revenue from Google AMP clicks; inscrutable deals with OTT providers; kindly billionaires; and so on. A publisher’s job in 2019 is to get wherever the getting’s good, and use it to fund the maximum amount of journalism.”

And finally ...

Dwayne Johnson, Dany Garcia Team With Powderkeg for Quibi Comedy Series

We don’t cover a lot of movie news around here, but I was struck by the plot of the show:

Titled “Last Resort,” the series centers on a Polynesian family-run resort in Hawaii that’s suddenly thrown into a whirlwind when a tech billionaire puts in a bid to buy the land. Johnson and Dany Garcia will executive produce via Seven Bucks Productions along with Paul Feig and Laura Fischer under their Powderkeg banner.

Hmmm … now where have I heard that before?

Talk to me

Send me tips, comments, questions, and plot points for the Rock’s Zuckerberg show: