When Google added a new button to search the web using generative artificial intelligence, the feature debuted with such glaring flaws, like a tendency to make things up, that I was skeptical it would become my go-to tool for finding information online.
But a year later, I confess that I became a convert. While the technology remains imperfect, I increasingly click the button marked “AI Mode” on Google.com to enter requests and instantly complete tasks that would have taken many minutes with the old search.
It took me some experimentation to get better AI search results, and the key was telling Google to work with a small amount of information instead of scouring the web for answers. I recently used Google’s AI search to identify an auto part to fix, pick out spices at the grocery store, and uncover internet scams. In a normal web search, I would have to run a few keyword queries and read a few articles to do this, but Google’s AI has pretty much automated the process.
Big caveat: I recommend avoiding using AI mode as a normal search engine and looking for direct answers to questions. Google’s AI often spews misinformation – last week it incorrectly said a car diagnostics app could tell me if my vehicle was ready for a smog check, a non-existent feature.
A New York Times analysis found that roughly 10 percent of the answers generated by Google’s artificial intelligence were incorrect. (Of course, an optimist would say that 90 percent were right.) Google processes more than five trillion searches a year, which means tens of millions of wrong answers every hour; in our analysis, Google wrestled with basic facts, including historical data, names and news about celebrities.
Google said the vast majority of its AI answers were accurate. She disputed the Times study, saying it doesn’t reflect what people are searching for on Google. Regarding my issue with the car diagnostic app, it also stated that the AI search technology can misinterpret web content.
While we wait for AI Mode to get better at answering direct questions, we can still use it for other purposes where AI technology is already better than the old keyword search. I’ll go through my favorite examples.
Grocery shopping
In my experience testing generative AI, the technology is most reliable when you direct it to gather information from trusted sources instead of from all over the web. The same goes for Google’s AI search.
Take my recent visit to a local Korean grocery store as an example. I was shopping for gochujang, a fermented bean paste, and the store had several brands. I remembered seeing the episode “America’s Test Kitchen” which featured the popular gochujang brands of the cooking show, but I couldn’t remember which ones were the best.
So I uploaded a photo of all the brands of gochujang carried in the store and wrote this prompt: “Which one does America’s Test Kitchen recommend?”
Google’s AI immediately identified a top shelf bean paste that had received rave reviews from the show, and I placed it in my grocery cart. It was much faster than googling the episode and rewatching it to find the recommended tag.
Ordering spare parts
The search AI also excels at identifying objects, which came in handy during a recent repair I did on my car.
A few months ago a relative driving my car hit a pole in the garage and broke the plastic side mirror cover. I wanted to order a spare part. So I took a picture of my car and the damaged mirror, drew a circle around the broken plastic, uploaded the image to AI mode and wrote “Identify this part”.
Google’s AI immediately identified the part, described it as a mirror base cover, and showed where it could be ordered online. Following this, I asked for a video demonstrating how to replace the mirror part, and AI pulled up the relevant YouTube clip.
When the part arrived a few days later, I followed the video and completed the repair in 15 minutes.
Searching for flight deals
While planning a summer vacation, I realized that using Google AI to find flight deals is better than browsing travel aggregator sites like Google Flights. That’s because, in addition to the price, I could adapt the criteria to my personal situation – traveling with a toddler.
For a trip to Hawaii, I asked AI Mode to find the best flight deals in July.
“Check out the best flight deals to Oahu in July,” I wrote, “Keep in mind I’m traveling with a toddler so extremely late arrivals are out of the question.”
Google’s AI suggested morning or afternoon flights that arrived in Oahu by 4 p.m. They weren’t the cheapest tickets available, but they were the best value for my family.
Product research
I also used AI search to factor in my personal preferences when narrowing down options when shopping for products like shoes.
I said I was looking for the best shoes for my foot condition, plantar fasciitis. The shoes on Google’s AI list that podiatrists recommend for people with plantar fasciitis.
Next, I asked the AI to present my options in a graph and Google created a table containing the main features and price of each shoe. I ended up ordering a pair made by Brooks that was on sale.
Fraud detection
In addition to identifying objects, artificial intelligence is useful for detecting signs of Internet fraud.
When I suspect an online store might be fraudulent when shopping, I simply copy and paste the website into AI mode to ask if the website is legit. The AI quickly alerted me to potentially fraudulent activity by pointing out red flags like a strange web address, incredibly low prices, and bad customer reviews.
Using artificial intelligence for fraud detection has benefited me in other surprising ways.
Last month I ordered an engine part from eBay for my wife’s car (yes, I’ve been having a lot of problems with the car lately), which the seller identified as original equipment from Bosch.
The part arrived in an intact Bosch box and looked real. But when I installed it the check engine light came on.
Confused, I took a picture of the part and the box and uploaded it to the AI search tool. It immediately flagged the item as counterfeit because of one detail I overlooked: The part number printed on the auto part differed from the part number on the box label by a single digit.
When I shared the photos and this observation with the eBay seller, he promptly refunded my money.
