Is AI e-fooling us all? Most can’t tell real email from fake

This story version has been formatted as an on-air script for broadcast outlets. See the original research story here.

IN A RECENT GLOBAL SURVEY, MOST PEOPLE COULDN’T DIFFERENTIATE BETWEEN A PHISHING MESSAGE WRITTEN BY A-I AND AN AUTHENTIC, HUMAN-WRITTEN EMAIL.

THE SURVEY FROM YUBICO OF 18,000 EMPLOYED ADULTS FROM AROUND THE GLOBE TESTED THEIR AWARENESS WHEN IT COMES TO CYBER DANGERS, LIKE A-I AND PHISHING, AND FOUND STARTLING ROOM FOR IMPROVEMENT. 

BUT INTERESTINGLY, AGE DID NOT SEEM TO PLAY A ROLE IN AWARENESS, AS THERE WERE NO SIGNIFICANT DIFFERENCES BETWEEN GENERATIONS IN BEING ABLE TO CORRECTLY RECOGNIZE THE PHISHING ATTEMPT (GEN Z 45%, MILLENNIALS 47%, GEN X AND BABY BOOMERS, BOTH 46%), HIGHLIGHTING THE FACT THAT NO GROUP IS EXEMPT FROM NEEDING EXTRA CYBER-CAUTION IN THE AGE OF A-I. 

CONDUCTED BY TALKER RESEARCH, THE RESULTS FOUND THAT MORE THAN FOUR IN 10 PEOPLE (44%) HAVE CLICKED ON A LINK OR OPENED AN ATTACHMENT FROM A PHISHING MESSAGE IN THE LAST YEAR, WITH 13 PERCENT EVEN ADMITTING THEY’VE DONE SO IN THE LAST WEEK.

Survey methodology:

Talker Research surveyed 2,000 employed adults from each of the following countries: the United States, the United Kingdom, Australia, India, Japan, Singapore, France, Germany and Sweden; the survey was commissioned by Yubico and administered and conducted online by Talker Research between Aug. 15 and Aug. 27, 2025.

We are sourcing from a non-probability frame and the two main sources we use are:

• Traditional online access panels — where respondents opt-in to take part in online market research for an incentive
• Programmatic — where respondents are online and are given the option to take part in a survey to receive a virtual incentive usually related to the online activity they are engaging in

Those who did not fit the specified sample were terminated from the survey. As the survey is fielded, dynamic online sampling is used, adjusting targeting to achieve the quotas specified as part of the sampling plan.

Regardless of which sources a respondent came from, they were directed to an Online Survey, where the survey was conducted in English; a link to the questionnaire can be shared upon request. Respondents were awarded points for completing the survey. These points have a small cash-equivalent monetary value.

Cells are only reported on for analysis if they have a minimum of 80 respondents, and statistical significance is calculated at the 95% level. Data is not weighted, but quotas and other parameters are put in place to reach the desired sample.

Interviews are excluded from the final analysis if they failed quality-checking measures. This includes:

• Speeders: Respondents who complete the survey in a time that is quicker than one-third of the median length of interview are disqualified as speeders
• Open ends: All verbatim responses (full open-ended questions as well as other please specify options) are checked for inappropriate or irrelevant text
• Bots: Captcha is enabled on surveys, which allows the research team to identify and disqualify bots
• Duplicates: Survey software has “deduping” based on digital fingerprinting, which ensures nobody is allowed to take the survey more than once

It is worth noting that this survey was only available to individuals with internet access, and the results may not be generalizable to those without internet access.