In a Pew Research Center survey of 11,004 U.S. adults fielded December 12 to 18, 2022, only 30 percent correctly identified all six listed everyday products and services as using artificial intelligence. Fitness trackers (68 percent correct) and customer service chatbots (65 percent) were recognized most often, and email spam filters (51 percent) least. Everyday AI is mostly pattern-matching software built into familiar tools, and Google Maps travel-time prediction shows how one such system works.
Key takeaways
- Pew's 2022 survey found 30 percent of U.S. adults answered all six AI-recognition questions correctly, and 31 percent got two or fewer right.
- Education mattered: 53 percent of adults with a postgraduate degree scored six of six, against 14 percent of those with a high school diploma or less.
- Google Maps ranks routes by predicted travel time from a graph neural network, which Google's research team said raised ETA accuracy by up to 50 percent in several cities.
- A spam filter is a classifier that scores each message from patterns learned on labeled examples.
- Pew measured U.S. adults only, and the data is from late 2022, so the survey describes awareness at that time.
What Pew asked, and what people got right
The report, Public Awareness of Artificial Intelligence in Everyday Activities, was published February 15, 2023. Respondents were asked whether six items use AI. The share who answered correctly for each one:
| Everyday item | Correctly said it uses AI |
|---|---|
| Wearable fitness trackers that analyze exercise and sleep | 68 percent |
| Customer service chatbots | 65 percent |
| Product recommendations based on past purchases | 64 percent |
| Security cameras that recognize faces | 62 percent |
| Customized music playlist recommendations | 57 percent |
| Email services that sort messages as spam | 51 percent |
The average respondent got 3.7 of six right. A further 38 percent answered three to five correctly. Younger adults did better: 75 percent of those ages 18 to 29 recognized chatbots as AI, against 45 percent of those 65 and older. On self-reported contact, 27 percent said they interact with AI almost constantly or several times a day, 28 percent about once a day or several times a week, and 44 percent less often.
Why the quiet cases go unrecognized
Spam filtering and playlist suggestions run without any visible prompt. A person types nothing and sees only a result: a message lands in a junk folder, or a song appears in a queue. Chatbots and voice features announce themselves, and the survey results track that difference. Recognition therefore measures how visible a feature is as much as how common it is.
A worked example: how a spam filter scores a message
This is an illustrative example with invented counts, not data from any real mail service. Suppose a filter has seen 1,000 labeled messages: 100 spam and 900 legitimate. The word "prize" appeared in 40 of the spam messages and in two of the legitimate ones. The filter asks how likely a message is to be spam given that it contains the word.
- Spam share of the sample: 100 of 1,000, or 10 percent.
- Chance a spam message contains "prize": 40 of 100, or 40 percent.
- Chance a legitimate message contains it: two of 900, about 0.22 percent.
- Combining these with Bayes' rule gives 0.04 divided by the sum of 0.04 and 0.002, about 95 percent.
Real filters weigh many signals together, and the method above is a simplification. The pattern is the same: counts from labeled examples become a probability, and a threshold decides which folder a message goes to. A legitimate prize-draw notice would also be scored high, which is why filters make mistakes in both directions.
A documented case: predicting travel time in Google Maps
In a post dated September 3, 2020, DeepMind described a traffic model built with Google. Road segments are treated as nodes in a graph, linked when they are consecutive or meet at an intersection. A graph neural network predicts travel time for each "supersegment," a group of adjacent segments that share a lot of traffic. Google Maps then combines these predictions with candidate routes to rank them by estimated time of arrival (ETA). Before the change, the post said, predicted ETAs were already accurate for more than 97 percent of trips. After it, accuracy improved by up to 50 percent in cities including Berlin, Jakarta, Sao Paulo, Sydney, Tokyo, and Washington, D.C.; across the cities shown, the gains ranged from 16 percent in London and Copenhagen to 51 percent in Taichung City.
The same group published the method at the CIKM 2021 conference in a paper on ETA prediction with graph neural networks. The reported gains are the developers' own figures for a comparison with the previous production model, and the improvement is in prediction error, not in how fast anyone actually drives.
Reading everyday AI claims
- Ask what the system predicts: a travel time, a spam probability, a next song.
- Ask what it was compared with, and who ran the comparison.
- Expect errors at the edges, since each of these tools works from probabilities.
- Check the date and population of any survey, because Pew's figures cover U.S. adults in 2022 and no matching Canadian figure was verified here.
FAQ
What counts as AI in everyday life?
Software that learns patterns from data to classify, rank, or predict. Spam filters, route-time estimates, product suggestions, and face-recognizing cameras fit that description. A fixed rule, such as a timer, does not.
How many people knew they were using AI in 2022?
In Pew's U.S. survey, 30 percent correctly identified all six examples, and the mean score was 3.7 of six. Knowing that a tool uses AI was therefore uneven across the public.
Did AI really improve Google Maps arrival times?
Google's research team reported better ETA accuracy after adding a graph neural network, with gains of up to 50 percent in some cities. These are the developers' figures, and the quoted range depends on the city.
Is everyday AI always accurate?
No. Each of these tools produces a probability or an estimate, so it will sometimes misclassify a message or misjudge a route. Even a large improvement leaves some error.
Related reading
- AI and its environmental footprint
- How predictive analytics works
- AI and the workplace
- AI use in New York
More in AI Basics.
Last verified: October 2026
Sources
- Public Awareness of Artificial Intelligence in Everyday Activities, Pew Research Center, February 15, 2023
- Traffic prediction with advanced Graph Neural Networks, DeepMind, September 3, 2020
- Derrow-Pinion A et al., ETA Prediction with Graph Neural Networks in Google Maps, CIKM 2021, arXiv:2108.11482