Is the internet really dead?

0:00 Because of my job, I read a lot of comments on the internet,

0:03 and some of them can be a little strange.

0:05 Alexistedceleti1989.

0:10 These can't be written by real people, right?

0:13 One subscriber joined nine hours ago.

0:18 Oh.

0:20 Oh!

0:22 What the- (quirky music) I've been thinking about this a lot,

0:26 because I see headline after headline about how bots are everywhere.

0:30 They're in YouTube comments, in Reddit threads,

0:32 in Instagram DMs, and all over X.

0:36 Some people go as far as to say that we've passed a major threshold,

0:39 and that the majority of what we see

0:41 on the internet isn't human anymore, it's bots.

0:45 The vast majority of the internet now is just bots.

0:48 Is just bots.

0:49 How many of these people are fake?

0:50 More often than not, the people you interact with aren't people at all.

0:54 I thought they got rid of the bots, though.

0:56 No.

0:57 No, the bots are back?

0:58 By mostly bots.

0:59 And there could be some truth to it.

1:00 A 2025 report claims that bots now make

1:03 up more than half of all internet traffic.

1:06 Does that mean that the threads, and feeds,

1:08 and comments that I read every day look like this?

1:12 Figuring out the answer matters.

1:14 We are collectively spending enormous amounts of time on social media,

1:19 so I set out to figure something out:

1:21 Am I reading, and engaging with, and reacting to real people on the internet?

1:27 And if I'm not, how would I know?

1:30 (quirky music) Okay, let's back up for a second.

1:38 I hear about bots in headlines all the time,

1:42 bots interfering with politics, bots duping us into scams,

1:47 bots amplifying internet outrage cycles, but I wanted to understand the basics,

1:54 like what is a bot and how does it work?

1:58 (quirky chiming music) If you look up bot farm on YouTube,

2:04 this video is the first result.

2:06 (operators speaking in foreign language) (quirky chiming music)-

2:12 MIN Software is a company based in Hanoi, Vietnam.

2:15 On their website, you can buy services like increased engagement on TikTok.

2:20 This video, racks of phones on a wall,

2:22 separate social media accounts all running on each phone,

2:25 it was kind of the stereotypical image I had

2:27 in my head of what's behind a bot network,

2:30 (operator speaking in foreign language)- but it turns out,

2:32 most of the time they run on servers with SIM cards,

2:35 like the ones you can see here

2:36 in these cybercrime busts by police in Latvia and Ukraine.

2:40 Bots like these might be programmed to post messages at a certain

2:43 time or to monitor something and post updates about it,

2:46 or to amplify a certain topic or account.

2:49 I learned that the majority of the time,

2:51 bots like these are not designed for high-level

2:53 propaganda or foreign interference, but for profit,

2:56 to make money by raising the prominence of something on the internet,

3:00 like buying followers on social media to make

3:02 an account look more popular than it actually is.

3:05 I've known for a long time that that is something that people can do,

3:09 but I wanted to understand how we can actually have proof when it's happened.

3:15 Before we dive deeper on bots, this is an 11-page report about someone,

3:20 or at least, what data brokers think they know about someone.

3:23 It's full of things like their address history,

3:25 their income range, criminal record, even some financial information.

3:30 And if you Google your own name right now, you'll find the same thing,

3:32 websites claiming that they have your personal data ready to sell to anybody.

3:37 Our personal info is a product, but you shouldn't have to go completely off

3:41 the grid to have some degree of privacy.

3:44 That's why this video is sponsored by Aura.

3:46 They scan a massive database of data broker sites looking for your info,

3:49 and then automatically send opt-out requests on your behalf,

3:52 and then keep following up to make sure it stays off.

3:55 We call this stuff personal information for a reason,

3:58 and it feels empowering and reassuring to be

4:02 able to regain some degree of control over it.

4:04 There are other tools that claim to do

4:05 this, but none offer the complete protection that Aura does,

4:08 from data broker removals to identity theft

4:11 alerts and credit monitoring, all in one place.

4:14 If you wanna see what's out there about you and start cleaning it up,

4:17 go to aura.com/christophe to try Aura for free for 14 days.

4:21 There's probably more about you online than you realise,

4:24 and Aura can help you take control of it.

4:26 Okay, let's get back to the bots.

4:29 As I started diving into the research on social media bots,

4:32 there was one name that kept coming up

4:34 again and again,- Comments can be fabricated, fake.

4:37 The YouTube video, the view counts,

4:39 there's monetary incentives for influencers to have lots of followers.

4:43 (chiming music)- These charts show audience growth on X.

4:48 The X-axis tracks an accounts first follower,

4:51 all the way up to its most recent follower,

4:53 and the Y-axis tracks when that follower's account was created.

4:57 Here's NTV, a Turkish news channel.

5:00 The acquisition or the growth of accounts is quite organic,

5:04 that I can say easily.

5:06 Over time, NTV gained an audience of followers who'd

5:09 created their Twitter accounts across a broad time range.

5:12 When NTV gained its 3-millionth follower in 2015, for example,

5:16 its most recent followers had created their accounts in 2015,

5:19 2014, 2013, all the way back.

5:22 Some of those new followers had created their Twitter accounts as early as 2007.

5:27 The fact that this chart has a steady incline

5:29 shows that it grew its audience slowly over time.

5:32 Compare that to the distribution of account

5:34 creation dates on this politician's followers.

5:37 This is, for instance, an interesting one.

5:39 This particular account has 200,000 follower at the time of analysis,

5:43 so this is kind of like suspicious pattern.

5:46 This flat line indicates a sudden increase in followers in February,

5:50 2023, but all of this space is blank,

5:52 because those new followers didn't create their accounts over a broad,

5:56 organic range of time,

5:57 all of these new followers created their accounts at the same time.

6:01 Those are likely bots.

6:03 Onur's work focused on Turkey, where he lives.

6:06 but these charts would look similar for anyone who

6:09 purchased bot followers or had bots purchased on their behalf.

6:13 So when we're on social media,

6:15 we are sharing the platform with tonnes of these entirely automated accounts.

6:21 My next question was how would I recognise one when I see it?

6:27 I started to study lots of media literacy guides on how

6:30 to spot bots using clues like what their account looks like,

6:33 what they post, and how often they post.

6:36 Look at this account, for example, it has no bio,

6:38 it doesn't have many followers, it isn't following many accounts.

6:41 The account was created in 2019, and it was very active in 2019,

6:45 but it hasn't been active since then, and when it was active,

6:48 it posted very frequently, mostly quote tweets with hashtags.

6:52 These media literacy guides are a helpful

6:54 starting point for doing a manual check,

6:56 but Onur also showed me a tool that makes that whole process much more simple.

7:01 Boom.

7:02 Andreah83940354 gets a bot score of 4.5 out of 5, very likely a bot.

7:09 (quirky music) This is Botometer,

7:11 a bot detection algorithm that Onur developed back in 2014.

7:15 In Botometer, we were extracting lots of features, looking from content,

7:19 social network, to temporal dimensions of, like,

7:22 how frequently they are posting.

7:24 It works by analysing X accounts for features associated with bot

7:27 accounts and giving them a bot score from 0 to 5.

7:31 Okay, I'm a little disappointed I'm not a 0 out of 5, but I'll take it.

7:36 (quirky music) Using a bot detection algorithm like this to check

7:39 if an account is a bot or not is entertaining,

7:42 but by repeating this process again and again at a massive scale,

7:47 we can get an idea of how many

7:48 bots there are on an entire social media platform.

7:52 But wait, I thought we already knew

7:53 from this estimate that that number is over 50%.

7:56 This chart comes from a 2025 report

7:59 by a cybersecurity software company called Imperva,

8:02 and it generated a lot of headlines by claiming

8:04 that bots now account for half of all internet traffic.

8:08 I even covered it on this channel.

8:09 So why is this bot traffic rising?

8:11 When I first saw this, I assumed it meant that half of all posts,

8:14 content, and accounts on social media were generated by bots,

8:18 but Imperva's estimate includes all bot activity,

8:21 the bots that index websites for search engines,

8:23 the bots that commit cyber attacks,

8:25 or the bots that monitor website performance.

8:27 Social media bots are just part of all bot activity.

8:30 I wanted to know what percentage those bots were of all users on social media.

8:35 Social media platforms, if they do publicly provide a number,

8:39 will generally claim it's low.

8:41 Twitter reported in 2022 that fewer than 5% of its

8:44 daily active users were false or spam-based on an internal review,

8:48 and while platforms do have access to data that outside researchers don't,

8:51 they also have an incentive to under-report bots.

8:55 They're publicly traded companies,

8:56 and they want to report a high number of active, real users,

9:00 but by using those bot detection algorithms

9:02 on data sets from social media platforms,

9:05 researchers can make estimates of their own.

9:08 The only problem is that those estimates are all over the place.

9:12 (quirky chiming music) I talked to one

9:16 researcher who made one of those estimates.

9:18 I would say that the population on the bots is probably about 20%.

9:22 And I talked to another who doesn't think that estimates can be made at all.

9:26 I would advise uncertainty (chuckles).

9:28 I think I'm suspicious of any particular estimate.

9:31 So why do the estimates vary so much?

9:34 Bot detection algorithms are trained to look for signature bot-like behaviours,

9:38 but those algorithms differ.

9:40 They might disagree on whether a given account is a bot.

9:43 The data sets they're tested on also differ,

9:45 so the estimates aren't all really of the same thing.

9:49 It's an extremely frustrating answer to land on.

9:52 Bots are on social media platforms,

9:54 but the experts studying those bots don't all agree about whether or not

9:59 we can know just how many there are on any given platform.

10:03 They also don't agree about whether we can reliably

10:07 tell if something is a bot or not, at all.

10:10 But then I talk to someone who shook up my whole perspective on bots.

10:14 The mistake that a lot of researchers make is

10:18 they say a bot is an account that meets X, Y, and Z parameters.

10:23 You know, it tends to be extremely active, for instance,

10:27 it tends to have few followers and few followings,

10:31 it tends to have been created recently.

10:34 You know, you can name a number of different metrics that you

10:37 could put on something that is more likely to be a bot.

10:41 Bot detection algorithms rely on this idea that bots have

10:44 telltale signs where they just act differently than humans do.

10:46 Darren made the case that humans just

10:49 act too unpredictably to ever know for sure.

10:52 There's all kinds of reasons that a real person is

10:55 gonna be using a burner account and might be running,

10:58 you know, more than one account on more than one computer,

11:01 and if I can't point to a monetizable

11:03 reason (chuckles) why somebody is running a fake account,

11:06 I'm usually very hesitant to say that something is a bot.

11:12 There are other reasons to be sceptical when you see estimates like these.

11:15 For one thing, the estimates don't separate benign bots, like news updates,

11:19 from malicious ones that are the ones actually worth worrying about.

11:22 And even if, say, 20% of accounts on a platform are bots,

11:26 that doesn't mean that 20% of the content you see will be from bots.

11:30 Those bots could be concentrated in specific communities,

11:33 or they could be programmed to do simpler

11:35 tasks like engaging with content, not creating it.

11:38 The way we engage with those, they're not actually talking to us (chuckles),

11:42 they're just affecting a number,

11:44 they're affecting the probability that something might appear in our stream,

11:49 whatever platform we might be on, so, you know,

11:52 we're not engaging with those as though we're engaging with another human.

11:55 And that's most bots, they're just there to affect a number.

11:59 Darren told me that generative AI is changing that, it's

12:01 enabling bots to interact online in more convincingly human ways.

12:05 (quirky music) All of this left me with a pretty unsatisfying answer.

12:11 There are probably more bots than what platforms claim

12:14 and fewer bots than what the most pessimistic third-party estimates claim.

12:18 Whatever that number is, we aren't outnumbered by bots on social media,

12:22 and the internet isn't technically dead,

12:24 but I do think I might have learned why it feels that way,

12:27 because even if the grand majority of accounts online are still human,

12:30 what we actually wind up seeing online is

12:32 more and more shaped by forces that aren't human.

12:35 Early social media had this promise of authenticity and community, connection.

12:41 When algorithmic feeds came along, they tweaked that promise.

12:43 They promised to surface and curate the very best stuff for you,

12:48 but they also made it possible for other people to optimise

12:51 their way into your eyeballs by making things tailored for engagement,

12:55 and bots could exploit that system by making

12:57 things look more popular than they actually are,

13:00 so humans had to act a little bit more like

13:03 bots to make it to the top of the pile.

13:06 The reasons behind why we see what we

13:08 see online have gotten more and more abstract,

13:11 and that leaves us susceptible to big ideas

13:14 about just how inhuman the internet has become.

13:18 I think it's a knee-jerk reaction that we've almost been programmed- Programmed.

13:22 (both laugh) Yeah.

13:24 to put out there, that, oh, this is weird to me,

13:26 so it's a bot, and no, people are weird (laughs).

13:30 At the end of the day, people are weird.

13:32 We're weird in real life, we're even weirder online.

13:35 I think that, you know,

13:37 there's a real danger to looking at difference online and assuming

13:44 that that difference indicates that you're not a real person,

13:47 and to dismiss various attitudes and beliefs as, oh, that's just the bots,

13:52 when really it's just people that are different than us.

13:56 Making this video, I've learned that this idea that the internet

13:59 is dead and that social media is mostly made up of bots,

14:03 it's not literally true, but I don't think that the internet has to be

14:07 made up mostly of bots for it to feel dead,

14:11 it just has to feel like it's not entirely for us anymore.

14:15 Hey, thank you so much for watching.

14:18 We had a lot of fun making this video.

14:20 I'm recording this right now from the New York office of NewPress.

14:25 It is, it's a work in progress (laughs).

14:29 NewPress is the organisation that I'm helping build,

14:31 along with video journalists, Johnny Harris, and Sam Ellis, behind Search Party.

14:35 We're gonna use it to shape our stories out in the open

14:37 and be as transparent as possible every step of the way.

14:41 If that sounds like something that you wanna see out there in the world,

14:43 you can join the waitlist right now at newpress.com.

14:47 We're just getting started.

14:48 Okay, see you next time.