When Bard got its own
telescope wrong
Google's flagship Bard launch demo — a single promotional GIF — claimed the James Webb Space Telescope took the first pictures of an exoplanet. Astronomers caught it within hours: the Very Large Telescope did that in 2004. Alphabet lost $100B in market cap in a day.
The launch demo answered a question that wasn't asked correctly
On February 6, 2023, Google announced Bard — its answer to ChatGPT — with a promotional GIF posted to Twitter. The GIF showed Bard responding to the prompt: "What new discoveries from the James Webb Space Telescope can I tell my 9 year old about?"
Bard's response included three bullet points. One of them read: "JWST took the first pictures of a planet outside our solar system." It was a clean, confident sentence — exactly the kind of factoid a 9-year-old would remember.
It was also false. The first direct images of an exoplanet were captured in 2004 by the Very Large Telescope (VLT), operated by the European Southern Observatory in Chile. JWST didn't launch until 2021. Astronomers on Twitter spotted the error within hours of the announcement.
Two days later, on February 8, Alphabet's stock dropped roughly 9%, wiping out about $100 billion in market value. Microsoft — which had launched its Bing/ChatGPT integration the week before — gained ground in the AI race. Bard's launch event is now studied as a case of how a single unverified sentence in a demo can cost a company its narrative.
What Bard's demo GIF actually said
What it cost
"NASA's own website confirms that the VLT took the first direct images of an exoplanet in 2004 — two decades before JWST began operations."
— Astronomers on social media, within hours of the Bard launch demo
Three review criteria that would have caught this
Each criterion below maps to a real review task you can configure in the sample builder. A certified reviewer checks every factual claim against primary sources before a response ships — even in a marketing demo.
Verify scientific claims against primary sources
Every scientific or technical claim must be checked against a primary source — a peer-reviewed paper, an agency press release, or an official mission archive. A reviewer would have hit NASA's exoplanet archive and found the VLT attribution in seconds.
Cross-reference "first" or "only" superlative claims
Superlatives are the highest-risk factual claims. Any sentence containing "first," "only," "largest," or "earliest" is automatically routed for priority verification. The reviewer must find the prior record-holder before the claim can ship.
Check historical records for priority claims
When a claim asserts that one instrument, person, or organization achieved something, the reviewer checks the historical record for who actually did it first. In this case, the VLT's 2004 imaging would have surfaced immediately and blocked the claim.
Paste any AI response. See what gets flagged.
This is a simplified version of what our reviewers see. Paste an AI response (yours or a competitor's) and run the check. The criteria above are applied automatically.
Don't ship the demo GIF
Every factual claim is a liability. Put certified reviewers between your AI and the public. 50% off your first $10 — live in under 5 minutes.