SAN FRANCISCO, September 18 — Ask three of the world’s leading AI chatbots what is happening right now, and you may not get the present moment back. That is the finding of a first-person test in which a writer put the same current-events question to OpenAI’s ChatGPT, Anthropic’s Claude and Google’s Gemini, and found that at least some of the answers were shockingly outdated. The test was a hands-on demonstration rather than a formal benchmark.
The writer posed one question to each assistant and judged the replies by how current they were. The comparison is anecdotal by design, and the article presents no independent study, statistic or third-party quote.
The evidence is the writer’s own transcripts and observations.
Why a chatbot can lose track of the present
All three products are large language models, systems trained on vast bodies of text that generate responses by predicting likely word sequences. Training runs take time and are expensive, so a model’s built-in knowledge carries a cutoff date. Without a live search or browsing tool, the model cannot know about events after that point.
Even with browsing switched on, results depend on how the system is configured and prompted. That is why two users asking the same question of the same chatbot can get different answers depending on settings, region, and whether web access is enabled.
The distinction matters. A model’s static training knowledge is not the same thing as a retrieval-augmented or search-enabled answer, and the gap between them is exactly what the writer’s experiment exposed.
Three products, one crowded race
ChatGPT is the conversational AI product developed by OpenAI, the San Francisco-based AI research company. Claude is the assistant built by Anthropic, an AI safety company founded in 2021 by former OpenAI researchers including Dario and Daniela Amodei. Gemini is Google’s flagship assistant, announced in December 2023 as the successor to Google Bard and positioned across Google’s products.
OpenAI, Anthropic and Google compete on model capability, and up-to-date information and reliable citations have become selling points in that contest. The test lands squarely inside that race.
For anyone who turns to a chatbot for news and current events, the practical stakes are straightforward. An outdated or confidently wrong answer can mislead, and the burden of verification falls on the user. The takeaway is cautionary and practical rather than alarmist.
Chatbots remain useful tools. They are not reliable real-time news sources, and time-sensitive claims are best checked against primary sources.
What to watch next is whether search-enabled answers close the gap the writer found, and how each company handles the trade-off between a model’s frozen training knowledge and the live web. Until then, the question “what is happening right now” may still return yesterday.
