Twitter Called Elon Musk’s Bluff About Bots. It Could Backfire.

Well, one good theatrical turn deserves another, which is just what has happened to Elon Musk.

The world’s richest person has spent the past several weeks complaining that Twitter is undercounting the amount of automated spam bots on its site and that it won’t give him access to the data he needs to make an independent assessment, which he needs before he’ll finish closing on his acquisition of the company. Twitter estimates bots are less than 5% of active accounts, a number recorded in many years of SEC filings. Since Musk could access the figure from the very beginning, his sudden fixation on it has seemed like an act, a means to manufacture a pretext for renegotiating his $44 billion offer for Twitter amid a wide drop in tech stock prices.

On Monday, Musk’s lawyer sent Twitter a tersely worded letter saying it had better turn over the data. Or else. Or else? Or else Musk and his attorneys will consider their refusal a breach of the merger agreement and call off the deal, a dubious argument that probably wouldn’t hold up in court—but one they could use to make things unpleasant for Twitter.

On Wednesday, we learned about Twitter’s own Musk-type move. The business is reportedly preparing to give him access to its so-called “firehose” API, a stream of every tweet sent. (Every. Tweet.) That works out to something like 500 million micro-blog missives per day. A Twitter spokesperson wouldn’t comment about how exactly it’ll share this trove with Musk, saying only: “Twitter has and will continue to cooperatively share information with Mr. Musk to consummate the transaction in accordance with the terms of the merger agreement. We believe this agreement is in the best interest of all shareholders. We intend to close the transaction and enforce the merger agreement at the agreed price and terms.”

To be clear, it is entirely overkill for Twitter to provide Musk with such broad access, a move meant to say to him, You want data? Here’s all the data we have. We’re not hiding a thing. Enjoy! “He doesn’t need access to everything. You don’t want access to everything,” says Goran Muric, a computer scientist at USC’s Institute for Information Sciences who has worked with similar Twitter APIs. In truth, Musk probably only needs the “decahose” API Twitter makes available to some researchers, which is 10% of all tweets. The difference between the results gleaned from the decahose and the firehose equates to the difference between “a poll and a census,” says Muric. Polls work with smaller, easier-to-use sample sizes. “And you can have multiple polls all the time and pretty much accurate results” dovetailing with what a more expansive census would produce, Muric says.

Sure, it’s amusing to see Twitter pull a Musk on Musk—to try, ostensibly, to call his bluff. But the decision is unlikely to provide a meaningful end to the dispute between the billionaire and the company he may possibly buy. And not just because it will take Musk considerable time to do an analysis of bots on Twitter, a task requiring a team of researchers who will need to laboriously construct software to review the tweets. Most to the point, while Musk can use his access to the firehose API to come up with an estimate of bot activity on Twitter, it seems almost inevitable that his figure won’t match Twitter’s.

To start, his definition of what constitutes a bot account could quite easily differ from Twitter’s. Musk can define a bot however he wants. There’s no universally accepted definition, even among the field’s top researchers. “If you put two people in the room and ask about the definition of anything, they would have a different opinion—and especially about the definition of what is a bot on Twitter,” says Muric. “So maybe if somebody tweets more than 1,000 tweets in a day, he’s a bot right? But maybe somebody else will say, It should be if they tweet more than 50 times.” The best tool for identifying bots, Botometer, which comes from an Indiana University team, offers only a rough gauge of bot activity, giving a probability score for whether an account is a bot, never a certainty. With so much wiggle room, Musk can probably use the firehose API to assemble whatever conclusions best fit his goal, which, again, seems like it is to find some reason to force Twitter to accept a reduced price.

Further, it’s unclear whether even a good-faith effort on Musk’s part could fully replicate Twitter’s process, doubling check its internal estimate. To be precise, Twitter has said bots account for less than 5% of its “monetizable daily active users,” a figure of Twitter’s own creation. (More often, an app will report monthly active users, people who log on at least once in the last 30 days.) So it’s uncertain whether even a well-intentioned Musk could use the firehose API to calculate the same number of monetizable daily active users.

Another matter: What happens if Musk detects some other problem at Twitter from going through the firehose API? Brandon Silverman, founder of the social media data tool CrowdTangle, purchased by Facebook, pointed out on Twitter that the platform may wind up making the situation worse for itself.

Really, it wouldn’t matter whether any additional problem was real or fake. All theater, after all, revolves around make-believe.