You’ll also now start seeing more information written by artificial intelligence, in one of the most consequential changes to your searches and the internet in Google’s 25-year history.

Google said Tuesday that beginning this week, everyone in the United States will have access to its “AI overviews” that Google has been slowly expanding for the past year. For some types of searches, Google will use the same type of AI that powers OpenAI’s ChatGPT to summarize information or spit out digestible answers to complex questions – all right at the top of the page.

The more Google’s search pages expand beyond the familiar list of website links, the harder it becomes to know what all of this stuff is, and whether the information is reliable.

“I don’t think the average person knows what the different [search] features are,” said Kyle Byers at Semrush, a company that studies how companies and people use Google.

To break down what’s in your search results, we did four Google searches and explained what we saw and whether it was helpful.

We want this anatomy of search results to demystify Google, which for a generation has been the starting point for our questions, curiosities and needs.

In a statement, Google said its search “connects people to helpful information for billions of queries every day” and that Google is “constantly innovating and building helpful new features to help people find exactly what they’re looking for.”

The more we learned about how Google search works, the more we believed it’s an incredible resource – but it isn’t always ideal or trustworthy, particularly as Google leans more on AI.

1. “What is the best Mexican restaurant near me with great margaritas, a nice atmosphere and at least four stars on Yelp?”

Google would not typically give you a cogent answer to this kind of multipart, personalized quest for Mexican restaurants in San Francisco. But Google says its AI-generated replies are ideal for some types of complex searches.

If you don’t see AI-generated replies like this yet, they are coming. Look for something like “AI Answer” or “Get an AI Overview for this search?”

We asked a human expert, San Francisco Chronicle associate restaurant critic Cesar Hernandez, to evaluate the Google AI’s restaurant suggestions. “I wouldn’t recommend any of those spots,” he said.

Hernandez said Chuy’s Fiestas and El Buen Comer are more what we’re looking for. He said Bombera in Oakland has great margaritas and food if we’re willing to travel farther.

Restaurant recommendations are subjective, of course. Google’s AI answer is likely synthesizing a huge volume of customer reviews on Yelp and a bunch of other information online.

It was useful to get a snappy answer without combing through reviews or articles. But the type of AI used by Google and ChatGPT essentially generates an average or typical response to your search. That might be great or good enough – or banal or wrong.

(Google said it is adding a feature to turn off search features like this and mostly see the classic web links.)

2. “Mexican restaurants near me”

This search is more typical. We saw what Google calls its “places” panels that pull information from Google’s databases of businesses, their locations in Google Maps, customer reviews written in Google, operating hours and more.

This is handy information. But some Google critics, including frequent antagonist Yelp, say Google is doing you a disservice by showing you at least some information from Google’s computer systems – even if that isn’t the best source.

For an alternative, try searching for Mexican restaurants in Apple Maps. You’ll see reviews from Yelp or other sources.

Google said its local business listings help generate billions of interactions each month with customers, including phone calls and dining reservations.

3. “Best vacuums for pet hair”

Searching Google for products is a minefield if you’re not sure what to buy, said Gisele Navarro, managing editor of product-review site HouseFresh.

You’ll see “sponsored” results at the top – companies that paid Google for ads that put their listings in a prominent place. Google makes no promise that those ads point you to the best vacuums.

Further down the page we saw small photos of other vacuum options. Those are from Google Shopping, mini-stores within Google that compete with Amazon. You’ll probably see Amazon listings when you do product-related searches, too.

(Amazon founder Jeff Bezos owns The Washington Post.)

Google, Amazon and the vacuum sellers are making no assurances that the products you see when you search for “best vacuums” are actually the best ones.

Navarro has also written about chronic deceit when Google prioritizes product reviews from some large websites that put little or no effort into testing products.

And Google has recently given more prominence in many product-related searches to forums like Reddit and Quora where people give advice to one another. The quality of those forums can be great – or outdated and spammy.

Google said its ads and product search results aim to be helpful and relevant, and that Google only makes money from ads when they’re useful enough for you to click on them.

Navarro believes Google is now so unreliable when you’re researching products that she tells friends and family members not to use the search engine for that.

Navarro recommends searching with alternatives like Kagi, DuckDuckGo or the Brave web browser. She said Consumer Reports, RTINGS.com, her own site and the Shortcut newsletter consistently publish trustworthy product reviews.

4. “Why did Toni Morrison change her name?”

We’re showing you nearly identical Google responses that came from two different parts of Google search.

First, we saw a Google AI-generated answer with links to four websites.

Google’s AI gave us a correct answer this time but experts generally say this type of AI is not reliable if you’re looking for facts like this Toni Morrison question. It’s best to click on one of the regular links Google shows you or search Wikipedia.

Sometimes AI will botch an answer or make one up. A recent X post showed a Google AI-generated reply for how to safely pass a kidney stone. The answer suggested drinking plenty of urine. (Google said it fixed the problem.)

We also saw almost the same answer about Morrison from a feature that Google calls a “featured snippet.” The text wasn’t AI-generated. It’s a highlighted section from a website that Google considers a reliable answer to your question.

You might see these snippets at the top of Google, as we did here, or under the “People also ask” questions.

This Morrison snippet was correct and sourced to a trustworthy news organization. Occasionally, though, Google highlights scam phone numbers for businesses or misleading information in snippets.

The company said it’s “extremely rare” for scam numbers to appear in snippets and that it has stricter rules on the information in those fields.