Firms’ Reactions to Information about Workers’ AI Acceptance

We run two consecutive online surveys with firm managers and one parallel worker survey. The first firm survey and the worker survey are designed to provide descriptive evidence about workers’ acceptance of AI and firms’ beliefs about these acceptance rates.
The experiment is implemented in the second firm survey, conducted about two weeks later. Based on the gaps identified between firms’ beliefs and factual worker acceptance rates, we design an information treatment to correct potential misperceptions. In the second firm survey, managers in the treatment group receive sector-level information about actual worker acceptance of AI.

After information provision, we elicit firms’ beliefs about their workers’ acceptance and collect different measures of expected AI adoption among firms. We then estimate treatment effects on these outcomes.

2026-01-12

2026-03-12

Firms’ beliefs about their workers’ AI acceptance and AI implementation expectations.

We will elicit firms’ beliefs about their workers’ AI acceptance and their own firm’s expected AI adoption using the survey questions below. We will summarize these responses into two standardized indices, which serve as our two primary outcomes:
1. An index of firm beliefs about workers’ AI acceptance (questions 1.a and 1.b)
2. An index of firms’ AI adoption expectations (questions 2.a, 2.b and 2.c)
Besides estimating treatment effects on the indices, we will also estimate effects on the individual index components.

Questions:

1. Beliefs about own employees’ attitudes about AI.
(a) Question: “[Insert selected statement regarding AI here; see section “Experimental Design” for details] What is your estimate of the approximate percentage of employees in your company who would ‘strongly agree’ or ‘somewhat agree’ with this statement?”
Answer categories: Respondents answer this question using integers from 0% to 100%. Respondents also indicate the certainty with which they hold these beliefs on a 7-point scale.


(b) Further beliefs about employees’ responses to AI (these outcomes are elicited in an item battery)
Question: “To what extent do you agree to the following statements?”; statements: “I am concerned that AI could endanger my employees’ jobs.”, “I believe that our employees are capable of making greater use of AI.”, “I expect that our employees will quickly adapt to using AI at work.”, “I believe that our employees will actively support the use of AI in the company.”
Answer categories: “Strongly agree,” “Somewhat agree,” “Neither agree nor disagree,” “Somewhat disagree,” “Strongly disagree.”

2. Expectations about AI implementation
(a) Statements about support of AI adoption
Question: “To what extent do you agree to the following statements?”; statements: “I support the use of AI among my employees.”, “I will work to ensure that AI is used more extensively in our company.”
Answer categories: “Strongly agree,” “Somewhat agree,” “Neither agree nor disagree,” “Somewhat disagree,” “Strongly disagree.”

(b) Information acquisition about AI implementation in firm
Question: “Would you like to receive links to information and support services related to the use of AI in companies at the end of this survey?”
Answer categories: „Yes“, „No“; in case of “Yes”, respondents are provided a respective list of links at the end of the survey.

(c) Investment expectations
Question: “What percentage of your company’s annual revenue do you plan to invest in artificial intelligence (AI) on average per year over the next five years?”
Respondents chose one of the following answer categories: “0% (no investment planned)”,”0.1% - < 0.5%”; “0.5% - < 1%”; “1% - < 2%”; “2% - <5%”; “5% - <10%”, “10% and more”

We will analyze effects on (i) economic-performance expectations and (ii) actual AI implementation. In addition, we will perform heterogeneity analyses by firm different firm characteristics.

As secondary outcomes, we investigate the effects on economic-performance expectations, which are more indirectly related to perceived acceptance of AI applications among workers. We summarize these outcomes into one economic-performance expectations index (summarizing questions 3.a and 3.b) as this secondary outcome (secondary outcome 1). In addition, we will investigate the effects on the individual index components.

3. Economic-performance expectations
(a) Productivity gain expectations
Question: “Do you expect that the use of AI will have positive effects on your company’s productivity over the next five years?”
Answer categories: “Yes”, “No”
If “Yes”: “What is your estimate of your company’s productivity increase over the next five years due to the use of AI?”
Answer categories: “0% - <1%”, “1% - < 5%”, “5% - < 10%”, “10% - < 15%”, “15% - < 20%”, “20% or more”

(b) Growth expectations
Question: “What is your expected annual average growth in revenue, workforce, and investment from 2026 to 2030?“
Answer categories: „Revenue growth (in %): …“, „Workforce growth (in %): …”, “Investment growth (in %)”

We will also match firms’ survey responses with measures of their actual AI implementation and innovation at a later stage. This allows us to investigate whether the information treatment affects real outcomes (secondary outcome 2).

We also plan to explore heterogeneities by the following firm characteristics:
(i) Firms’ prior beliefs about workers’ acceptance of AI in their sector.
(ii) Firms’ current level of AI integration.
(iii) Economic sectors (i.e., High-tech manufacturing, Other manufacturing, Knowledge-intensive services, Other services, Wholesale and retail trade, Construction)

Additionally, we will conduct exploratory analyses on further heterogeneities by firm characteristics (e.g., size, age).

The study is designed to identify how correcting firms’ misperceptions about workers’ AI acceptance affects firms’ beliefs and AI adoption expectations.
In the worker survey, we elicit acceptance rates using the following question:

“How much do you agree with the following statements?

(a) The use of AI makes my work more pleasant.
(b) The use of AI makes my work more interesting.
(c) The use of AI improves the quality of my work results.
(d) I have a positive attitude toward the use of AI in my work.
(e) I have concerns about the use of AI in my work.”

Respondents stated their agreement on a five-point scale: “Strongly agree,” “Somewhat agree,” “Neither agree nor disagree,” “Somewhat disagree,” “Strongly disagree.”

In the first firm survey, we elicit firms’ beliefs about the average worker acceptance in their sector. The question reads:

“We conducted an online survey in November 2025 among several thousand employees in Germany, asking how much they agree with the following statements about the use of AI at work:
(a) The use of AI makes my work more pleasant.
(b) The use of AI makes my work more interesting.
(c) The use of AI improves the quality of my work results.
(d) I have a positive attitude toward the use of AI in my work.
(e) I have concerns about the use of AI in my work.
Respondents were asked to choose one of the following answer options for each statement: “Strongly agree,” “Somewhat agree,” “Neither agree nor disagree,” “Somewhat disagree,” “Strongly disagree.”
In the following, we would like to know how you think these respondents answered.”

Question: „What is your estimate of the share of employees in your sector who would answer ‘strongly agree’ or ‘somewhat agree’ to the following statements?” For each statement, managers could select a percentage value from 0% to 100% in 10%-steps. Upon completing this beliefs-battery, respondents indicate the certainty with which they hold these beliefs on a 7-point scale.

Comparing firms’ beliefs and workers’ actual responses allows us to identify the dimensions where beliefs are most inaccurate.

In the second firm survey, conducted about three to four weeks later, we implement a randomized information treatment. Managers are randomly assigned to a treatment or control group. The treatment group receives information about the average worker acceptance rate of AI in their sector, presented both textually and graphically.

Treatment text:
“INFORMATION: In the recent survey of employees in your industry in Germany, 50% “strongly” or “somewhat” agreed with the following statement:

“[selected statement will be provided here]”

You yourself had estimated agreement at 22%.”

Before the information is provided, we elicit firms’ prior beliefs about the sector average. After the information treatment, we elicit firms’ beliefs about their own workers’ AI acceptance, as well as their expectations about AI implementation (primary outcomes), as well as additional secondary outcomes (see above).

Not available

By a computer

At the individual firm manager level

No

4,000 firms in the second survey (maximum number, contingent on successful re-contact rates)

4,000 firms in the second survey (maximum number, contingent on successful re-contact rates)

2,000 firms per treatment arm