AI-based propaganda is proliferating rapidly. Let's check it out.

While browsing X / Twitter recently, there were some new followers gained and plenty of low-engagement accounts tweeting about controversial topics. After scrolling through the feed for a while there was an interesting tweet that got the brain thinking about a topic that's both close to the heart and extremely prevalent for the modern age.

What topic you might ask? Propaganda. Or, more specifically. Propaganda that was enabled by the usage of language learning model AI’s.

Let’s have a look at the offending tweet. We’ll keep it cut down to the content only, to protect the not-so-innocent, and use a screen capture so we aren’t “targeting” anyone specifically or driving traffic to questionable accounts.


Looks pretty mild doesn’t it? But that tweet tells us something pretty interesting. Source: Author

Those who are familiar with ChatGPT, you’re probably well aware of the “I am an AI language model” responses that are returned when an AI system does something outside its programmed limitations. This is that response, in the wild posted to an actual and current Twitter / X account. The technology behind this isn’t particularly groundbreaking.

However, this begs the question. Can we use things like AI-based returns as another way to help identify spam and bot accounts? Or does searching for threads and phrases like this just confuse the matter making it harder to identify legitimate accounts? Let's do some digging and find out.

Our Search Phrase

Our Search string can be as complex or as simple as you’d like with varying levels of success with each return. For the purposes of today's article, we’ll use the phrase that originally caught the eye and then search on that to see what we find.

Given it’s a reasonably simple phrase we should expect at least a few hits back although, at the same time, it’s fair to assume that we’re using a pretty broad search term as well. We should probably expect to see some troll posts as well from people using the phrase as a sarcastic response or joke.

So to be clear, the phrase we’ll be using for this article will be. “As an AI language model”. Nothing more, nothing less.

What Did We Find?

Well as you’d expect, we found quite a bit. And, while it’s fair to assume that not all of these accounts are bots, there are a few common factors that a lot of bots will have in common.

We’ll cover those in a moment, but in the first capture, we can see a few accounts worth investigating further. We can also see that we’ve had the joke approach as well, with the phrase appearing in a bio. It’s good practice if you’re looking to identify bot accounts, to keep an open mind and use multiple factors that gauge the probability of an account being used by a human.


We aren’t looking to target accounts within an article as that is what the report button is there for. So we’ve taken out account details to protect the not-so-innocent but as we look further we can see two accounts that look like they may be posting using the API enabled with ChatGPT Inputs.

Some of the other factors that help give away a non-genuine or bot account are as follows.

1. Verified Checkmark: In the old days, this meant something. Today, it just means that someone paid. It’s pretty common to use the checkmark to help establish authenticity.
2. Low Account Age: Typically, bot accounts are either reasonably new with little posting history or old with a long time spent dormant. Analysing the type of engagements an account might have can help establish legitimacy
3. Low Follower Count: We get it, everyone starts from 0 after all. However, you’ll mostly find that bot accounts that spray propaganda or disinformation have reasonably low follower counts more often than not. Also, they’ll often follow thousands of accounts as well.
4. Lacks Genuine Engagement: Many bot accounts are used to spread propaganda or disinformation as their primary purpose and because of this you’ll typically see little in the way of genuine interaction. Limited diversity or repetitive content can be a good gauge of an account's status.
5. 
   

No Checkmark but it meets the other criteria. Source: Author

Sometimes, you’ll find accounts that meet all of these criteria and are still genuine. This is one of the many reasons why using multiple sources to make an informed decision is a smart thing, as for the most part, it’s pretty rare you can use a single sign consistently as a gauge of an account's status.

In the investigative world, jumping to conclusions too soon is never a good thing.

A New Era

In this article, we’ve simply tried to apply some curiosity and some search strategies to help make identifying malicious accounts a bit easier. We could, of course, go deeper and look at things like post content or posting habits to identify further activity or establish links between accounts or networks, but that's a bit outside the scope of today's article.

The truth is, as we approach an election year in the United States, we’ve released this article to get people thinking about what their positions may be around a few important issues that have influenced things like elections in the past. And while they might be controversial for some, they are important topics for us to have as a society.

1. Should we believe what we see on social media?
2. What impact does pairing AI-based systems with social media have on things like propaganda and disinformation?
3. What position should social media companies have when approaching disinformation and propaganda?

When encouraging discussion on these issues it’s important to understand that the concept of both propaganda and disinformation are not in the least new, and are something that humanity has dealt with since the birth of the information age.

What is new in these circumstances is the ability of AI-based systems to either amplify or drown out a perspective or opinion, and do it with little human engagement needed to carry this out on a large scale.

While it’s smart to assume that humans will continue to be involved in the deployment and assessment of tools such as these the deployment of AI-powered systems alongside means that each person becomes more effective, while the message being sent can be sent more clearly with less management needed. The days of the AI revolution may be in the starting phases, but if we are to understand and mitigate some of these issues then it’s better to start doing that sooner rather than later.

Over To You

So what can you do to deal with, and raise awareness around some of these important issues?? Quite a few things.

1. Talk the Talk: Encourage people within your circle to have discussions around some of these important topics.
2. Zero Tolerance: Typically, we’re all about tolerance and letting things be but in this case it's time to discriminate. Downvote, Mute, Report, Block or don't engage with bot drive accounts and be sure the platform owners know they are there.
3. Don’t Buy In: Bots and Propaganda accounts go for the heartstrings or rage baiting to draw in engagement and amplify negative messages. By denying them that, you’re making it harder to get their message out an amplified, which is exactly what we want.

While some of these issues will take time and a coordinated approach around managing them both now and into the future, it’s our opinion that the time to have these discussions is now. What do you think?

If you found this article insightful, informative, or entertaining, we kindly encourage you to show your support. Clapping for this article not only lets the author know that their work is appreciated but also helps boost its visibility to others who might benefit from it.

🌟 Enjoyed this article? Support our work and join the community! 🌟

💙 Support me on Ko-fi: Investigator515

📢 Join our OSINT Telegram channel for exclusive updates or

📢 Follow our crypto Telegram for the latest giveaways

🐦 Follow us on Twitter and

🟦 We’re now on Bluesky!
🔗 Articles we think you’ll like:

1. Signals From Space: The International Space Station
2. Ukraine OSINT: Strava Strikes Again

✉️ Want more content like this? Sign up for email updates