Google's AI search is producing millions of wrong answers every day

[Daniel Sims]

Cutting corners: Most search engines now present users with AI-generated overviews by default, sparking controversy over concerns about accuracy and lost click-through traffic. While testing suggests that Google's AI overviews are accurate most of the time, the enormous volume of queries the search engine processes each day likely still results in millions of incorrect responses.

According to The New York Times, testing suggests that approximately one in 10 Google AI search overviews contains false information. Given that the search engine processes roughly 5 trillion queries per year, users could be exposed to more than 57 million inaccurate answers each hour – nearly 1 million per minute.

The figures come from AI startup Oumi, which the Times asked to evaluate Gemini's accuracy using SimpleQA, a widely used generative AI benchmark. After analyzing 4,326 Google searches, Oumi found that Google's AI assistant, Gemini version 2, produced accurate overviews 85 percent of the time in October. By February, Gemini 3 had improved that figure to 91 percent.

However, Oumi can evaluate large volumes of results only by relying on AI tools, which may also introduce errors. In addition, Google sometimes generates different AI overviews for the same query, even when it is repeated seconds apart.

A Google spokesperson called Oumi's testing flawed, arguing that it does not reflect real-world search behavior. The company's internal testing indicates that Gemini 3, when operating independently of Google Search, hallucinates 28 percent of the time.

Sourcing presents another challenge. Google attempts to support its AI overview results with relevant links, but those sources often do not substantiate Gemini's claims – whether accurate or not.

In some cases, an incorrect AI overview is immediately followed by a link containing correct information; in others, an accurate overview cites a source with inaccurate information; and sometimes the linked pages contain no relevant information at all. Notably, discrepancies between AI overviews and their sources increased after the February update, rising from 37 percent of searches with Gemini 2 to 56 percent with Gemini 3.

Researchers also found that AI overviews are susceptible to manipulation. In one example, a BBC journalist published a blog post containing false information and later found that Google repeated those claims the following day.

Tellingly, Google and other AI companies acknowledge the technology's tenuous relationship with the truth in the fine print. Microsoft's terms of service describe its Copilot AI tool as intended for entertainment purposes, not for making important decisions. Google's AI overviews advise users to double-check responses, while xAI acknowledges that hallucinations can occur.

Permalink to story:

Google's AI search is producing millions of wrong answers every day

 

[Squid Surprise]

So the point of this article is?

Yes, we know AI isn’t 100% accurate… humans aren’t either… the difference is, humans (and human nature) won’t change - whereas AI is improving all the time.

Let’s see the stats in 5 years… then 10…

 

[daffy duck]

As ordinary as DuckDuckGo's BING powered search is, it's still far more desirable than Google's AI megaslop as you can turn off all AI.

We are in a deplorable state when their are only about 3 independent search engines and Brave search is way too limited. No one outside of Russia would use Yandex, so it's not an option.

 

[loki1944]

What a shock

 

[Lew Zealand]

How was technically correct information that was irrelevant to the original query counted?

Because that's been the ongoing problem with AI answers and there have been no improvements. Ask a Q, AI focuses in on an easy and common answer containing most of your terms but it's irrelevant information for your actual question.

It's like arguing with someone who knows only how to invent strawmen instead of addressing the topic.

 

[James321]

Why not allow people to easily disable it?

 

[pnntmp]

As ordinary as DuckDuckGo's BING powered search is, it's still far more desirable than Google's AI megaslop as you can turn off all AI

We are in a deplorable state when their are only about 3 independent search engines and Brave search is way too limited. No one outside of Russia would use Yandex, so it's not an option.

Do you mean Yandex is an option inside Rusia???
There may be 3 independent old-school classic search engines, but these will disappear soon anyway. Models, which are now commonly used as search engines, are more.

Google's AI answers are pretty good, BTW, incl. in regard to accuracy. I have no idea how these results are relevant to real world search (probably not at all), but the chance to find false information using the classic search is way above 10%.

 

[ScottSoapbox]

Why not allow people to easily disable it?

You can.

It’s called using another search engine.

I personally prefer Brave search.

 

[ScottSoapbox]

A Google spokesperson called Oumi's testing flawed, arguing that it does not reflect real-world search behavior. The company's internal testing indicates that Gemini 3, when operating independently of Google Search, hallucinates 28 percent of the time.

Where is this 28% coming from? It doesn’t make sense in context of the article or intuitively. AI is bad but not that bad. @Daniel Sims

 

[TooQik]

I'm curious as to how AI can actually deliver an accurate overview of information from multiple unrelated sources in the first place.

For example, if an AI overview generator parses two conflicting pieces of information, how does it determine which piece of information is correct? As a human, we would choose based on our knowledge and experiences - AI doesn't have either of these to draw on, it simply has data.

Or is it a numbers game, where the most frequent piece of information witnessed becomes the "right" information?

I personally don't trust AI overviews. This is based on my experiences with them where I've been presented with information that I know is wrong, as well as information that is demonstrability wrong when put into practise.

 

[Burty117]

the chance to find false information using the classic search is way above 10%.

Only if you’re really bad at using search engines, when the AI refuses to give me useful information to my question, the fall back is normal search, because I’ll 99% of the time find the information I’m after, AI just keeps repeating itself.

Don’t get me wrong, AI prompting is becoming a skill in itself, so asking AI the question the “right way” so it gives you the answer you need is a skill I lack.

But I don’t lack the skills to use a regular search engine effectively, which is probably why I can easily fall back on it and get the answers I need.

 

[JamesBlond]

Don't tell the left those answers are wrong you will be hated.... just tell them its a different answer that you were looking for

 

[pnntmp]

Only if you’re really bad at using search engines, when the AI refuses to give me useful information to my question, the fall back is normal search, because I’ll 99% of the time find the information I’m after, AI just keeps repeating itself.

Don’t get me wrong, AI prompting is becoming a skill in itself, so asking AI the question the “right way” so it gives you the answer you need is a skill I lack.

But I don’t lack the skills to use a regular search engine effectively, which is probably why I can easily fall back on it and get the answers I need.

It's not about being good or bad at using search engines.
A classic search can yield results that vary from correct information to total BS. If you don't know the subject enough to make a difference, the chance to see BS is way above 10%.

 

[Endymio]

AI focuses in on an easy and common answer containing most of your terms but it's irrelevant information for your actual question.

It's like arguing with someone who knows only how to invent strawmen instead of addressing the topic.

I'm sorry, but your statement is incorrect. It was the Sumerians who invented straws around 3000 BC, and women use them as often as men.

 

[bloomsun]

I don't use search engines anymore at all. I use Kimi K2.5, which I prefer before any other AI.

 

[lonetac]

It's not about being good or bad at using search engines.
A classic search can yield results that vary from correct information to total BS. If you don't know the subject enough to make a difference, the chance to see BS is way above 10%.

fact of the matter is AI search overview is often wrong. its just searching for terms and scrubbing the internet. There is no intelligence behind it determining if the information is accurate.

If I made 1 million pages with wrong information and you google that term, its going to feed you that wrong information.

Ive even seen the AI overview add context and insert incorrect information into its overview. For example, when asking it about a licensing overview for a certain product, it assumed information that was never stated by the vendor.

The overviews might appear to be intelligent but its just repeating what it found on the internet and then adding its own fluff to it.

Again, it has no idea if what its saying is true or not because AI isnt actually intelligent. Its just repeating whats on the internet.

There is no way in the world that the google search AI is only wrong 10% of the time, its wrong like at least 40-50%.

If you are relying on google AI overviews, you have miserably failed yourself and are often repeating non sense that AI fed you without validating.

The overviews are meant to give you a starting point, its not the ending point.

 

[Watzupken]

I think the way they present the issue makes it sound like a small problem, I.e. it is only 1 mio out of 57 mio queries. In my experience, the wrong or made up response is "nicely" embedded in the output of our queries, which is actually problematic. That means I generally need to still skim through whatever supporting or documents to ensure the output is accurate.

 

[lonetac]

I think the way they present the issue makes it sound like a small problem, I.e. it is only 1 mio out of 57 mio queries. In my experience, the wrong or made up response is "nicely" embedded in the output of our queries, which is actually problematic. That means I generally need to still skim through whatever supporting or documents to ensure the output is accurate.

exactly. It inserts its own fluff into the responses and makes its response unreliable. its only a starting point.

its not smart, it has no intelligence to know if its correct. Its just repeating what it found and adding in a few words to make it sound correct. Then when you dig into it, its wrong.

 

[Squid Surprise]

I honestly don’t see the problem… it’s not like the AI overview needs to be used… is it that hard to scroll down? The “normal” search results are still easily found underneath the AI overview you know…

It’s basically just an extra result for the lazy - and it’s still usually right…

 

[BadThad]

"Google sometimes generates different AI overviews for the same query, even when it is repeated seconds apart."

Absolute fact, AI changes it's mind. AI is a tool, nothing more, it cannot be relied on without human verification. I've seen 100% contradiction between two different AI "answers".

 

[BadThad]

fact of the matter is AI search overview is often wrong. its just searching for terms and scrubbing the internet. There is no intelligence behind it determining if the information is accurate.

If I made 1 million pages with wrong information and you google that term, its going to feed you that wrong information.

Ive even seen the AI overview add context and insert incorrect information into its overview. For example, when asking it about a licensing overview for a certain product, it assumed information that was never stated by the vendor.

The overviews might appear to be intelligent but its just repeating what it found on the internet and then adding its own fluff to it.

Again, it has no idea if what its saying is true or not because AI isnt actually intelligent. Its just repeating whats on the internet.

There is no way in the world that the google search AI is only wrong 10% of the time, its wrong like at least 40-50%.

If you are relying on google AI overviews, you have miserably failed yourself and are often repeating non sense that AI fed you without validating.

The overviews are meant to give you a starting point, its not the ending point.

Exactly! Excellent comment!

 

[Burty117]

It's not about being good or bad at using search engines.
A classic search can yield results that vary from correct information to total BS. If you don't know the subject enough to make a difference, the chance to see BS is way above 10%.

So if what you say is true, explain to me why AI is any better?

 

[pnntmp]

So if what you say is true, explain to me why AI is any better?

If for no other reason, because it saves time - you don't have to browse lots of pages.

 

[Squid Surprise]

So if what you say is true, explain to me why AI is any better?

Why does it have to be “better”?

You still get the “normal” search results below… now you simply have another option - and 90% of the time, it works…

 

[lonetac]

Why does it have to be “better”?

You still get the “normal” search results below… now you simply have another option - and 90% of the time, it works…

I would say the overview is just a starting point for you to take your investigation further and should only be trusted once 100% verified.

As a master googler over the years, its not hard for me to find what I need on google without AI however the overviews will often link popular threads or starting pages to get you going.

90%, not in my experience but hey, its 2026 and 95% of stats are made up on the spot. We can skew a % to say anything we want.