As Israel continues its offensive in Gaza, another battle unfolds in the digital world — one between human voices and automated bots on social media.
Lebanese analysts Ralph Baydoun and Michel Semaan from the research consultancy InflueAnswers have been tracking the behavior of what appear to be “Israeli” bots since October 7. Initially, pro-Palestinian voices were dominant online. However, not long after, a wave of pro-Israeli responses started flooding these spaces.
“If a [pro-Palestinian] activist shares a post, we’re seeing a surge of pro-Israel responses — often within minutes or hours,” said Semaan. “Almost every post is swamped by accounts that behave alike and appear convincingly human — but they’re not.”
These are bots.
Understanding Bots
Bots — short for “robots” — are automated programs designed to carry out repetitive online tasks. They can be useful or malicious.
Good bots help with alerts, content discovery, and online customer support. Bad bots manipulate online engagement, spread lies, enable fraud, and harass users.
Sowing Chaos and Uncertainty
A study by U.S. cybersecurity firm Imperva revealed that nearly half of all online traffic in 2023 came from bots. Malicious bots made up 34 percent — their highest level ever recorded — while beneficial bots accounted for 15 percent. This increase was partly attributed to the growing use of AI for generating media.
According to Baydoun, the bots pushing pro-Israel content don’t necessarily try to win trust but aim to create doubt and disrupt pro-Palestinian narratives.
These so-called “bot armies” — sometimes numbering in the thousands or more — are deployed in vast misinformation campaigns to distort public opinion. The more sophisticated bots become, the harder it is to tell them apart from real people.
Jillian York, director of International Freedom of Expression at the Electronic Frontier Foundation, warned that these AI-generated bot swarms severely impact honest communication and silence genuine voices by overwhelming them.
The Evolution of Bots
Earlier bots were basic — rule-based programs that could only perform specific, repetitive tasks. As social media emerged in the 2000s, bots gained popularity for automating actions like adding friends or posting content.
In their early forms, bots used simplistic, repetitive language and were easy to detect. “Bots back then would just repeat the same message. You could instantly tell it wasn’t a real person,” said Semaan.
By the 2010s, advances in Natural Language Processing (NLP) — the AI field that enables machines to understand and generate human language — made bots more capable.
During the 2016 U.S. elections, University of Pennsylvania researchers found that bots accounted for a third of pro-Trump tweets and nearly a fifth of pro-Clinton tweets during the debates.
Later, Large Language Models (LLMs) — powerful NLP systems trained on billions of data points — further transformed bots into more advanced tools.
Rise of the Superbot
Superbots are bots powered by cutting-edge AI, including LLMs like ChatGPT. These systems allow users to craft realistic social media replies with simple prompts.
Baydoun and Semaan built a superbot themselves to better understand the mechanics. Here’s how it works:
Step 1: Target Identification
- The bot seeks out prominent accounts (verified or with high engagement).
- It filters posts by hashtags like #Gaza, #Genocide, or #Ceasefire, and tracks post popularity metrics.
- The link and content are then passed to the next step.
Step 2: Prompt Creation
- The content is entered into an LLM such as ChatGPT.
- Sample instruction: “Pretend you are a social media user with strong views. Respond to this tweet using a bold, pro-Israeli tone.”
- The model generates a reply, which can be adjusted before continuing.
Step 3: Post Engagement
- The bot posts the reply, often adding brief delays to appear human.
- If the original user replies, the bot can sustain the dialogue indefinitely.
- Multiple bots can attack many accounts in parallel, flooding the discussion.
Spotting a Superbot
While older bots were easy to spot, today’s LLM-powered bots are sophisticated, often simulating personality and reasoning. “It’s becoming nearly impossible to know if content is written by a person or an AI,” said Semaan.
However, Baydoun and Semaan say some signs still help identify bots:
- Profile: Bots may use AI-generated images with odd facial details. Their usernames may have random numbers or odd capitalisation. Bios are often vague but appeal to a specific audience.
- Account Age: Most bot accounts are created recently.
- Followers: Bots often follow other bots to appear popular.
- Posts: They frequently repost unrelated topics like music or sports, while replies tend to follow a distinct agenda.
- Activity Level: Bots respond rapidly, often within minutes.
- Posting Schedule: Bots post continuously, without breaks.
- Language Use: Their replies can sound overly formal or unnaturally structured.
- Targeting: Bots usually aim at influential or high-follower accounts.
What’s Coming Next?
A Europol report [PDF] warns that by 2026, 90 percent of online content could be AI-generated. Deepfakes — AI-generated media mimicking real people — have already played a role in the 2024 Indian elections, raising fears for the upcoming U.S. elections in November.
Digital rights advocates are sounding alarms. “We’re not just worried about censorship by governments. Bots and human-led propaganda can drown out real voices,” said York.
She frames the situation as a challenge to free speech itself — if genuine opinions are buried under waves of misinformation, expression becomes meaningless.
Activists are urging tech companies to take responsibility for protecting elections and public discourse, but York describes it as an uphill battle. “It’s a David and Goliath fight. These companies have all the power and resources.”
AI-driven bots have now become a new, covert weapon in the information warfare surrounding Gaza.
