AI Anxiety at Work: How to Turn Fear Into a Practical Plan -

AI anxiety is a useful description for worry about how artificial intelligence may affect work. It is not, by itself, a clinical diagnosis. If a new assistant raises the question, “Can it see customer information?”, the first step is not to predict what it means for an entire job. Name the concern, identify who can answer it, and choose one proportionate next action.

Some concerns call for personal learning. Others require an employer to explain data access, task ownership, training, evaluation, or human review. Separating those questions makes the next step clearer.

This guide uses a systems approach: classify the concern, review the task rather than an entire occupation, and keep consequential decisions with accountable people. It also distinguishes dated survey findings from clinical evidence and historical robot-displacement research from evidence about generative AI.

Do not automate a vague fear. Identify the concern, its owner, and the decision that would make it more answerable.

AI anxiety at work: turn a broad fear into a specific next step

Start with a short concern statement:

I’m worried about [specific issue] because [known fact or uncertainty].

Then ask what kind of answer would help. A skills question may need training. A question about customer data needs an answer from the tool, privacy, or policy owner. A concern about performance evaluation needs a clear explanation from the responsible manager or HR team. Persistent or severe personal distress belongs with a qualified mental-health professional, not a workplace productivity tool.

Name the concern before choosing a response

“AI anxiety” can describe several different workplace questions. Use the category to choose the owner and the next action:

Job-displacement concern: Will my tasks or role change?
Skills concern: What will I need to learn to do my work well?
Privacy concern: What information can the tool access, retain, or expose?
Evaluation concern: Will AI affect performance reviews, pay, or promotion?
Accuracy concern: Who checks the result, and what happens if it is wrong?
Autonomy concern: Can a person review or override a consequential decision?
Information-overload concern: How can I separate useful information from alarming coverage?

Observable signs may include recurring worry about tasks or skills, unanswered questions about data use, or feeling overloaded by AI coverage. These descriptions are not diagnostic criteria. Ask what information or action would make the concern smaller or more answerable. If distress persists, becomes severe, or interferes with daily functioning, consider qualified professional support.

Decision point

A training plan cannot answer an unclear data-permission question, and a policy statement cannot teach a missing skill. Route the concern to the person or function that controls the relevant decision.

What workplace surveys do, and do not, say about AI anxiety

Survey findings describe the people asked, the questions they answered, and the time the survey was conducted. They do not establish how every worker feels today, prove that a feared job change happened, or diagnose respondents.

EY’s 2023 AI Anxiety in Business Survey surveyed 1,000 employed US office or desk workers who were at least somewhat familiar with AI. EY reported that 75% were concerned AI could make certain jobs obsolete, 65% were anxious about AI replacing their job, and 63% were anxious about lacking access to AI learning opportunities. These are different measures, and the results are not a current estimate for all workers.
The American Psychological Association’s 2023 Work in America survey found that 38% of workers worried AI might make some or all of their duties obsolete. Among workers with that worry, 51% said their work negatively affected their mental health, compared with 29% among workers without the worry. APA also reported that 46% of worried workers intended to look for another employer within the next year, compared with 25% of workers who were not worried. These are associations and stated intentions, not evidence that AI worry caused mental-health outcomes or job changes.
A 2021 US survey study in Socius found that 13.7% of respondents reported that a robot had replaced their job. The study involved 1,959 respondents and measured reported robot replacement, not generative-AI displacement.

EY also reported that 81% of respondents supported more organizational self-regulation and 78% supported a larger government regulatory role. Those figures describe opinions in that 2023 survey, not a universal policy consensus.

Before repeating any percentage, check the fieldwork date, geography, population, denominator, and exact measure. Concern that certain jobs may become obsolete is not the same as expecting one’s own job to disappear. Stated worry is not observed job loss, and robot-replacement evidence is not generative-AI evidence.

Review tasks instead of predicting which jobs will disappear

Roles contain different tasks, and the same task can carry different risks in different settings. A task review can help a team decide whether to keep work manual, test assistance, or pause for policy or expert review. It is a work-design discussion, not a way to rank employees or predict who should lose a job.

For each proposed task, record its frequency, how explicit its rules are, whether its inputs are structured, the cost of an error, the need for human relationships, and who remains accountable. Repetitive, structured, reversible work with low error costs may be a candidate for a bounded AI test. Those characteristics do not prove that a task or role will be automated.

Keep it manual when the task depends on substantial judgment, sensitive information, or unclear authority.
Test assistance when the task is bounded, inputs are approved, errors are easy to catch, and a named person can review the output.
Stop for review when errors could affect health, safety, legal rights, employment, pay, or reputation, or when data permissions are unclear.

For a proposed test, agree on a measure before starting, such as reviewer-rated quality, time spent, or rework required. Keep the existing manual process available. If the test creates more checking work or misses important context, revise it or stop.

Organizations assessing whether a bounded workflow is appropriate can explore AI agent consulting as an optional next step.

A practical response for individuals and teams

Match the action to the concern rather than trying to master every AI tool or predict every labor-market change.

Skills uncertainty: Choose one task relevant to your role and one learning goal. Try a low-risk example only with an approved tool and non-sensitive inputs.
Workplace uncertainty: Ask the responsible manager or policy owner how AI changes task ownership, data use, output review, training, and escalation. Request a written answer where it affects how work is done.
Privacy uncertainty: Check organizational rules before entering information. Do not put confidential or sensitive data into a tool unless that use is explicitly authorized.
Media overload: Set a deliberate limit on AI coverage. For a work decision, check primary sources and the date and scope of evidence rather than relying on a viral summary.
Persistent personal distress: Seek qualified support rather than treating an AI productivity tool or a workplace experiment as care.

01Name itWrite one concern and separate known facts from unanswered questions. The individual or team raising it owns this first description.

02Route itChoose the right owner: self-directed learning, a manager, HR, a privacy lead, or a policy owner. Keep personal concerns private unless the person chooses to share them.

03Choose one actionLearn one relevant skill, ask for a policy answer, limit coverage, or propose a controlled test. The action should address the named concern.

04Review the resultFor a work test, the task owner checks quality, data permissions, and the agreed measure against the current process.

05Escalate when neededSend unresolved operational questions to an accountable workplace owner. Persistent or severe personal distress belongs with a qualified professional.

Use workplace AI as a bounded assistant, not an authority

ClickUp documents Brain features for drafting, summarizing, simplifying, translating, generating action items, and creating supported ClickUp items. Users can review, replace, insert, copy, or continue working with a response. Brain may use context from work the user has permission to access. Availability and limits vary by plan, role, workspace, and rollout.

The documented product capability supports a user-initiated sequence: provide an approved task brief, ask Brain to draft or summarize, review the response, edit or reject it, and then insert suitable text or create a supported item. The documentation does not establish that Brain diagnoses anxiety, verifies research, predicts job loss, or makes employment decisions.

For example, a user could provide approved, non-sensitive notes and ask Brain to draft action items for a routine internal project update. The user compares the suggestions with the notes, removes unsupported items, confirms the owner and due context, and only then adds suitable work to the relevant task or Doc. If the task needs confidential data or an unreviewed consequential decision, stop and use the appropriate policy or expert review.

Three illustrative patterns for a controlled workflow

Concern routing: A worker voluntarily submits a privacy concern through an approved form. A human or approved tool may summarize the source wording and suggest a category. A reviewer checks that the summary preserves the concern, removes unnecessary personal or customer data, and routes the question to the privacy or tool-policy owner. If the volume is low, a deterministic form and human routing are preferable to AI classification.
Task exposure review: A manager proposes using AI for a specific routine task. The team records frequency, rule clarity, data format, error cost, human-accountability requirements, and a named reviewer. The decision is to remain manual, test assistance, or stop for review. A manual fallback remains available, and the pilot ends if permissions, authority, or the agreed success measure is unclear.
Evidence organization: A user asks Brain to summarize an approved source or create a draft action list. The reviewer checks the source URL, publication date, percentages, denominator, and wording before the result enters a shared document. A source summary is not treated as verified research, and a concern label is never used to make an employment or mental-health judgment.

These are editorial workflow designs, not ClickUp-published fields or a guaranteed JSON, API, export, or deduplication contract. If a team stores records, define the row grain first. For example, one row may represent one distinct concern submission, while a separate evidence row represents one claim from one source version. Do not combine a person, a survey result, and a CRM contact or deal event in one record.

For evidence records processed concurrently, use a database-enforced unique key or transactional upsert. A proposed citation-level key could combine the normalized source URL, source version, and a normalized claim or citation hash. A survey-result key should also include the study, survey wave, geography, subgroup, metric, and denominator definition. A lookup followed by create is not sufficient when concurrent workers can process the same record.

Pre-flight check for a voluntary AI test

Use one defined task and inputs approved for the chosen tool.
Name the person who will review the output and the destination for approved work.
Set a quality, time, or rework measure before the test.
Keep a manual fallback and stop if permissions, authority, or error costs are unclear.
For evidence records, preserve the source version, metric, denominator, and row grain before saving.

Read ClickUp’s official guidance on Brain writing and review options and creating supported items with ClickUp AI. Organizations considering a reviewed ClickUp workflow can also see ClickUp setup and automation.

When concern needs more than a work experiment

Workplace questions about monitoring, data access, performance evaluation, or changes to a role need an answer from someone accountable for that policy or decision. A tool-generated summary is not that answer. Decisions affecting employment, pay, promotion, health, or legal rights must remain with accountable human decision-makers.

Ordinary, situation-specific concern does not establish a mental-health condition. If anxiety is persistent, severe, or interfering with daily functioning, contact a qualified mental-health professional. Do not infer a diagnosis from a survey response or an AI-generated label.