How Far Do You Verify Generative AI for Business Use? 1 in 4 Submits 'As Is'! 14.1% of Users Experience Business Errors

NEXER Inc., in collaboration with RSG Inc., a recruitment agency specializing in construction and real estate, conducted a survey on 'Verification of Generative AI Responses Used in Business' and is pleased to announce the results.

■ Generative AI is convenient, but can its responses truly be trusted?

Generative AI has become a familiar tool for improving business efficiency. Its applications are countless, including document creation, drafting emails, and information gathering. However, the answers provided by AI are not always accurate. So, to what extent do we verify AI's responses before using them in our work?

This time, in collaboration with RSG Inc., we conducted a survey on 'Verification of Generative AI Responses Used in Business' targeting 500 men and women nationwide.

*When quoting the content of this press release, please comply with the following:

・State that the source is 'a survey by NEXER Inc. and RSG Inc.'

・Provide a link to RSG Inc. (https://rsg-c.jp/)

・Provide a link to RSG Tenshoku Navi (https://rsg-tenshokunavi.jp/)

"Survey on Verification of Generative AI Responses Used in Business" Survey Outline

Survey method: Online questionnaire

Survey period: April 17, 2026 – April 24, 2026

Respondents: Men and women nationwide

Valid responses: 500 samples

Questions:

Question 1: Do you use generative AI for business?

Question 2: Have you ever copied, submitted, or sent AI responses as is?

Question 3: To what extent do you verify the accuracy of AI responses?

Question 4: In what situations do you verify AI responses? (Multiple answers allowed)

Question 5: Have you ever made a mistake by using AI responses as is?

Question 6: Please describe the specific nature of the mistakes that occurred by using AI responses as is.

*As a general rule, percentages are rounded to one decimal place, so the total may not always be 100%.

■ 15.6% responded that they "use generative AI for business"

First, we asked if they use generative AI for business. The result was 15.6% for "yes" and 84.4% for "no." Although its adoption is progressing, it seems that only a portion of the workforce uses it daily in the workplace.

■ 25.6% responded that they "have copied, submitted, or sent AI responses as is"

Next, we asked those who use generative AI for business if they have ever copied, submitted, or sent AI responses as is. The result was 25.6% for "yes" and 74.4% for "no." This means that approximately one in four people has experience using AI responses without any modifications. Of course, if the response content is perfect, there is no problem. However, AI output may contain misinformation or outdated information, and using it without verification carries a certain degree of risk.

■ 42.3% responded that they "only verify important content" regarding the accuracy of AI responses

Next, we asked those who use generative AI for business to what extent they verify the accuracy of AI responses. The most common answer was "only verify important content" at 42.3%. This was followed by "always verify in detail" at 38.5%, "lightly check" at 16.7%, and "hardly verify" at 2.6%. Combining "always verify in detail" and "only verify important content," it is clear that over 80% of users perform some form of verification. On the other hand, it is also true that about 20% only lightly check or hardly verify, indicating that users' stances on AI responses vary greatly.

■ 55.1% responded that they verify AI responses "when they feel a sense of discomfort"

Next, we asked those who use generative AI for business in what situations they verify AI responses. The most common answer was "when they feel a sense of discomfort" at 55.1%. This was followed by "when the content is highly specialized" at 53.8%, "when it includes numerical values or factual information" at 44.9%, "when used for external documents/emails" at 37.2%, "before submitting to a boss/client" at 24.4%, and "other" at 1.3%. Since AI can present misinformation in natural-sounding sentences, the ability to feel a sense of discomfort may depend on the user's experience and knowledge.

■ 14.1% responded that they "have experienced business errors" by using AI responses as is.

Finally, we asked those who use generative AI for business if they have ever experienced business errors by using AI responses as is. The result was 14.1% for "yes."