OPINION:
When people interact with artificial intelligence platforms, they often believe the technology possesses higher wisdom or a more objective form of knowledge, but beneath the mystique lies a simple truth: AI does not understand; it guesses.
AI platforms powered by large language models, such as ChatGPT, Gemini, Claude, Grok and Perplexity, are trained on vast collections of online text. They use these collections to generate probabilistic predictions, which are shaped not only by past data but also by how users question, challenge and interact with the platforms.
Although they excel at synthesizing information from a wide range of sources, these models cannot independently evaluate the reliability of those sources, which becomes problematic when they are asked questions about current events. If biased articles dominate the media, as they have regarding Israel, AI summaries and analyses will reflect that.
As AI shifts from a novelty tool to a primary method for understanding the world, users must recognize how deeply these systems depend on their training data and why active, critical engagement matters.
Multiple recent studies demonstrate a clear rise in the use of generative AI tools for searching, learning and making sense of information. Today, 60% of Americans use AI as a search engine, and 25% get news from AI. That number is sure to rise, especially among young people who are already served a deluge of anti-Israel content across social and traditional media.
HonestReporting’s review of New York Times reporting showed that 46% of Israel-related articles expressed empathy exclusively for Palestinians, while only 10% expressed empathy for Israelis — even in the weeks after the Oct. 7, 2023, massacre. When patterns like these dominate the dataset, AI platforms will treat that imbalance as a neutral baseline rather than an aberration.
Users of AI are not powerless. Just as every news article becomes fodder for AI, so does every interaction with AI. It’s learning from us, and we can help improve it.
When an AI-generated answer feels incomplete, ask yourself: Is the terminology misleading or vague? Does the response feel one-sided? Is context missing? If so, engage with the AI on those grounds.
A recent Guardian article accuses Israel of “weaponizing water” while omitting Hamas’ destruction of infrastructure, including turning water pipes into missiles, and Israel’s attempts to repair pipelines and restore desalination capacity.
If AI is trained on such coverage with no pushback, its answers will mirror those omissions — so start pushing back. Try asking about water in the Gaza Strip and then asking the engine to take HonestReporting’s analysis of the article into account and ask it to answer again. You’ll see the difference for yourself.
Here are some techniques you can use to get AI to present a more complete picture of the news.
First, establish standing instructions to draw from a variety of sources and perspectives. Don’t just ask “What happened?” Ask: “What do multiple sources across the political spectrum say about this?” Tell your AI you require citations, source diversity and full transparency about where it’s getting its information. You can even ask it to remind you if certain media outlets are state-run, nonprofit or privately held.
Then, supply evidence yourself and ask for the strongest counterargument. Share relevant articles and expert statements and ask the AI to reevaluate its answers. This helps shape the models with more accurate responses, and it’s a good reminder that the presence of countervailing information doesn’t necessarily mean AI takes it into account.
Finally, share your successful corrections with others to help build a community actively challenging algorithmic bias. The more people who are aware of AI’s biases and limitations, the more responsibly they will handle the content it generates.
Passively accepting AI responses has consequences. Just as you create a corrective signal when you challenge a biased answer with evidence, accepting a flawed response without pushback reinforces existing imbalances.
Models will reproduce errors until someone provides them with better information. Your corrections are data points that do more than improve one answer; they shape the trajectory of future models.
Think of yourself as an AI accountability partner, someone who helps systems become more accurate by supplying missing information, demanding context and challenging inaccurate narratives.
AI is already influencing global perceptions about Israel. The question is whether we allow biased inputs to define that future or whether we actively intervene.
Just as HonestReporting’s readers have changed how major media outlets cover Israel and its recent war against Hamas, AI users can shape how algorithms synthesize and present information about the Jewish state and the conflicts it may encounter in the future.
We all can help shift AI toward fairness and accuracy. The models are listening to us; we just need to speak up.
• Didi Shammas-Gnatek is the AI project leader at the nonprofit media watchdog HonestReporting.
Please read our comment policy before commenting.