Google Bard Inaccurate Info: What Happened?

Yes. Google’s Bard chatbot shared inaccurate information in a public demonstration in February 2023, and it has also sometimes produced incorrect or misleading answers in use. The episode highlighted a basic limitation of generative AI: Bard could make a response sound confident without making it true.

The 2023 Bard demonstration

Google announced Bard on February 6, 2023, describing it as an experimental conversational AI service powered by its language models. In an example shared by Google, a user asked what new discoveries from the James Webb Space Telescope could be explained to a nine-year-old. Bard replied with several bullet points, one of which said the telescope had taken the very first pictures of a planet outside our solar system.

That claim was wrong. The first confirmed image of an exoplanet—meaning a planet beyond our solar system—was made in 2004 by the European Southern Observatory’s Very Large Telescope, as NASA also notes. The JWST has since observed exoplanets and contributed important research, but it did not take the first such picture.

The error was visible in Google’s own promotional material before Bard was broadly available. It drew attention because the answer was presented smoothly and without any signal that the claim was uncertain. The mistake was not evidence that every Bard response was false; it was a clear example of why a fluent answer needs fact-checking.

Why chatbots can give wrong answers

Bard was a large language model: a system trained to generate text by predicting likely sequences of words, informed by patterns in its training and, depending on the product and task, information it can retrieve or use from other sources. That process is not the same as looking up every claim in a verified reference book.

As a result, a chatbot can produce what is often called a hallucination: a plausible-sounding statement that is fabricated, inaccurate, or unsupported. It may combine a true detail with a mistaken date, name, quotation, or causal explanation. Confidence and clear prose are characteristics of the generated response, not proof that the underlying information has been checked.

Errors can arise for several reasons:

• Incomplete or conflicting information: Sources available to a model may not cover a subject well or may disagree.
• Ambiguous questions: A chatbot may assume what a user means instead of asking for clarification.
• Outdated knowledge: A model may not know about later events, or may fail to use current sources effectively.
• Faulty synthesis: It may join separate facts in a way that creates an incorrect conclusion.
• Misleading prompts: A question that presupposes a false claim can lead the system to repeat or build on it.

What happened to the Bard name?

Bard was Google’s name for its conversational AI product after the 2023 announcement. In February 2024, Google renamed Bard to Gemini, the same name used for its family of AI models. The rebrand means that current references to Google’s chatbot generally use “Gemini,” while “Bard” is relevant to the product’s earlier history.

The name change did not mean that the original error was corrected simply by changing the label. Google has continued developing its AI products, but no generative chatbot can be assumed error-free. Features, available models, and connections to Google services can change over time, so an old Bard example should not be treated as a complete description of what Gemini can do now.

How to check an answer

For everyday brainstorming or drafting, an occasional mistake may be easy to spot. For facts that matter—such as historical claims, technical instructions, or information used in a report—verify the answer independently. A good check focuses on the specific claim, not just whether the overall response sounds reasonable.

1. Identify checkable claims. Pull out names, dates, numbers, quotations, and statements of cause or priority.
2. Ask for sources, then open them. A citation or link is a lead, not a guarantee. Confirm that the source exists and actually supports the claim.
3. Compare reliable references. For scientific or historical claims, look for primary sources or established institutions, and check whether independent sources agree.
4. Check dates and context. A statement can be accurate for an earlier period but no longer current, or true only under particular conditions.
5. Ask the chatbot to distinguish certainty from uncertainty. This can make gaps easier to notice, but it does not replace checking.

For the telescope example, checking NASA’s account of exoplanet imaging against the European Southern Observatory’s record would have revealed that the “first pictures” claim was incorrect.

What the mistake does—and does not—show

The mistake showed that Bard could produce a factual error in a public-facing answer and that polished wording was not a dependable accuracy signal. It also raised questions about how companies demonstrate emerging AI systems: a carefully chosen example can illustrate a capability, but it cannot establish that the system is reliable across subjects.

It does not show that every answer from Bard or Gemini is inaccurate, nor does one demonstration establish how often a product makes mistakes. Accuracy varies with the question, the subject, the information available, and the model or feature being used. A chatbot can be useful for generating ideas, summarizing material, or offering a starting point while still requiring human review.

One common misconception is that a chatbot is a search engine that simply retrieves a correct page and repeats it. Some AI products can draw on web information or show sources, but their generated wording may still misstate what a source says. Conversely, a plausible answer without a citation is not automatically false. The practical approach is to evaluate the evidence behind important claims.

Frequently Asked Questions

Did Google Bard give a wrong answer in its launch demonstration?
Yes. In a February 2023 promotional example, Bard incorrectly said the James Webb Space Telescope took the first pictures of a planet outside our solar system. The first confirmed exoplanet image came from the European Southern Observatory’s Very Large Telescope in 2004.

Is Bard still available under that name?
No. Google renamed Bard to Gemini in February 2024. “Bard” refers to the earlier name of Google’s conversational AI product.

Does Gemini make factual mistakes too?
It can. Like other generative AI systems, Gemini may give inaccurate, incomplete, or outdated information. Important claims should be checked against trustworthy sources.

Why did Bard sound certain when it was wrong?
Language models generate likely text rather than guaranteeing that each statement has been verified. Natural, confident phrasing can therefore accompany an incorrect answer.

How can I tell whether a chatbot answer is reliable?
Check its specific claims against sources that directly support them. Verify citations, dates, and context, and compare independent authoritative references when the information matters.

Was the exoplanet claim the only problem with Bard?
The example is a prominent documented error, but it should not be treated as a complete record of every mistake. Chatbots can err in many ways, and one example alone cannot establish an overall error rate.

Related

• How AI Bridges the Gap Between Developers and Testers
• The Future of Television at CES: What Comes Next?
• AI Font Generators: How They Work and What to Check