February 2023: Bard's $100 Billion Typo

Written in 2026, looking back at Bard’s launch demo with the benefit of knowing how it ends.

The clip is a few seconds long. A promo for Google’s new chatbot, Bard, answers a question about the James Webb Space Telescope and says it took the very first picture of a planet outside our solar system. It didn’t. That honour goes to a different telescope, years earlier, and astronomers pointed it out almost immediately. By the time markets closed, Alphabet had shed something in the region of a hundred billion dollars of value over a single sentence in a marketing graphic.

I remember reading it on my phone in Bali, coffee going cold, thinking: that is not a company that meant to launch this yet.

The tell was the speed

ChatGPT had arrived in late November 2022 and by February it was the only thing anyone in my world could talk about. Clients, friends, my own group chats. Google, which had been quietly sitting on this kind of language model for years, suddenly had to look like it was in the race. So it shipped a demo. Fast. And the fast was the problem, because the answer inside it hadn’t been checked by anyone who knew about telescopes, or apparently by anyone at all.

What struck me at the time, and looks even clearer now, is that the mistake itself was small and human. Anyone can get a space fact wrong. The story wasn’t the fact. The story was that the most powerful search company on earth was moving so quickly, so nervously, that it let an unchecked claim into a launch it had staged on purpose. You do that when you are scared of being left behind, not when you are calm.

A scared giant explains a lot of what came next

I keep coming back to that word, scared, because I think it’s the honest read and because it predicts so much of the following three years. If Google had felt secure, it would have taken its time, tested Bard quietly, launched when it was ready. Instead it reacted. And a company reacting out of fear tends to keep reacting.

From where I sit in 2026, almost everything Google did to search after this moment makes more sense if you assume the starting emotion was fear rather than confidence. The rush to put AI answers directly in results. The willingness to disrupt the click, the thing that made the whole business work, before it had a replacement for the click. The constant relabelling of features. None of that reads like a company executing a plan. It reads like a company that got a fright in the winter of 2022 and never fully got its breath back.

Why I wrote this down

At the time I thought the Bard demo was mostly a funny, expensive typo, a good story for a Monday. I filed it under embarrassing rather than important.

I was underrating it. What I was actually watching was the moment the incumbent stopped setting the pace and started answering someone else’s. Once you see that, the pattern holds for years. Every scramble that followed, the ones that scrambled my own clients’ traffic, traces back to the same posture.

I don’t feel smug about it, if anything I feel a bit sheepish that I laughed first and thought second. But it taught me to watch what a big company does when it’s frightened, not what it says when it’s rehearsed. The rehearsed part is easy to fake. The fear leaks out in the timing, and the timing here was screaming.