Opens in a new tab
needsahuman.

Will AI replace office clerks, general?

A little.

Much of the day is phone calls, walk-ins, mail, errands, and records checks that software alone cannot finish. This job scores 65 out of 100 on (higher is safer). Today AI could do about 29% of the work by itself, people do 13% with AI’s help, and 58% still needs a person.

Updated 3 October 2026 43-9061 4159, 4132 2026-Q4
Office and Administrative SupportOffice Clerks, General43-9061 · 2026-Q4
29% AI does it13% AI helps58% needs a human
Your job's name, lit by the work that still needs a human.Needs a human 58%AI helps 13%AI does it 29%

AI does it: AI can do the task largely by itself. AI helps: a person still does it, faster with AI. Needs a human: AI can do little of it yet.

Why the clerk’s desk hasn’t emptied out

General office clerks do a bit of everything, and that is the whole point. One hour is typing and proofreading records. The next is answering the phone, routing a caller, opening the mail, and walking a signed form to the person who has to see it. The honest answer to whether AI will replace office clerks is that the job keeps shedding tasks while the role itself holds on, usually with fewer people doing it.

Much of the work is glue work. A clerk spots an invoice that came in with the wrong account number, asks the sender about it, and fixes the record before it reaches accounting. Software can flag the mismatch. Someone still has to make the call. That mix of small judgment calls and physical handling is why the job sits across the whole other office and administrative support workers group rather than inside one neat task.

Scale matters too. BLS counts about 2,464,940 people in this occupation, with median pay of $45,010 a year (BLS, 2025). BLS also projects employment down about 6% between 2025 and 2035. That is the pattern across administrative support work: hiring thins out, often at the first rung, long before any one desk is handed to a machine.

What software finishes, what it assists, what stays with people

29% of task time is work current tools can finish on their own. Keying data from a scanned form into a database is the clearest case. Sorting and compiling records by rule is another: by date, client, or account number, work that used to eat an afternoon.

13% is shared work, where the tool drafts and a person checks. Searching files and records to answer someone’s question fits here, as does preparing routine forms and letters from a template. Both still need a clerk who knows which version of a document is the real one. On our can-AI-do-it measure, explained on the coverage method page, this job reads 39 out of 100.

58% stays with a person. Greeting walk-ins and handling the phone call that has gone sideways sit there. So does carrying mail, deliveries, and signed paperwork around a building, plus keeping the copier, scanner, and postage meter running. A real share of the handling work is physical, and the hardware class that would cover it is mobile robots, which are neither cheap nor common in a mid-size office.

What has actually been tested

Very little, directly. Our evidence grade for this job is D on an A to D scale, and a D means no study has measured AI output against a qualified office clerk doing this job’s real task mix. So there is no parity number here, and we do not publish one.

Benchmarks do cover pieces of the work: pulling fields out of documents, transcribing audio, filling standard forms. A clerk’s day is the assembly of those pieces, inside one organization’s filing rules, with people walking up to the desk. A timed comparison on a live inbox, a real records system, and a week of phone traffic would settle it. Until something like that exists, the score leans on task data, which is set out on our methodology page.

When the picture could change

Most likely between 2035 and 2053 (8 in 10 of our scenarios). Two things could pull that earlier. More records are born digital, so there is less paper to move, and routing, filing, and routine replies are becoming default features of office software rather than a tool someone has to buy. When the cost of the software side sits far below a salary, the argument gets made quickly.

Two things hold it back. The errands are stubborn: mail, deliveries, machines, and the badge-and-signature end of records work. And accountability sticks to people. When a payroll file or a client record is wrong, an employer wants a named person who checks it, which is why records retention and signature rules slow handoffs. How the window itself is built is on the replacement-year method page. For the wider trend in clerical headcount, see our list of jobs AI is expected to shrink.

How office clerks stay needed

Lean into the parts that stay on your side of the desk. Handle the caller or visitor who does not fit the script. Own the records system, so you are the one who knows which file is correct and who is allowed to see it. Keep the physical chain moving: mail, deliveries, signed paperwork, and the machines that jam.

Two skills matter more than any job title. The first is tool fluency: knowing what the office software can route, extract, and draft, and knowing where it quietly gets things wrong. The second is process ownership: writing the checklist, fixing the step that keeps failing, and training the next hire on it. That is work a tool does not do for you, and it is how clerks move into coordinator and office manager roles.

What to do: pick one recurring task on your desk this month, document it end to end, and automate the keying part yourself before someone else does.

Nearby moves worth a look: Data Entry Keyers, Secretaries and Administrative Assistants, Except Legal, Medical, and Executive, and Receptionists and Information Clerks. You can put any two of them side by side on our compare jobs tool. If you are early in your career, our guide to AI and entry-level jobs covers the hiring side of this, and the full job rankings show where clerical roles sit against the rest of the labor market.

Frequently asked questions

Is AI going to replace admin jobs?

Not as whole jobs, on the evidence so far. The pattern is task erosion: keying, sorting, and routine drafting move to software, while calls, visitors, physical paperwork, and responsibility for records stay with people. The practical effect shows up in hiring. Employers keep a smaller admin team using better tools rather than closing the function. The task list above shows which parts move first.

Which office clerk tasks is AI taking first?

The rules-based ones. Pulling fields from a scanned form into a database, sorting and compiling records by date or account, and producing standard letters from a template. These have clear inputs, clear outputs, and little negotiation. Tasks that involve a person on the other side of a desk, or paper that has to be carried and signed, move much more slowly.

What skills should an office clerk learn now?

Learn the software your employer already pays for, in depth: records systems, spreadsheets, document workflows, and whatever assistant features are built in. Then learn to document and improve a process, because that is what gets you a coordinator title. Communication under pressure helps too. The tasks listed above as needing a person are the ones worth practicing deliberately.

Is general office clerk still a reasonable job to start in?

It can be, with eyes open. BLS projects the occupation shrinking about 6% between 2025 and 2035, and median pay was $45,010 a year (BLS, 2025). Treat it as a way into an organization rather than a destination. Clerks who learn the records system and the process end up in scheduling, operations, bookkeeping, or office management roles.

Which jobs will not be replaced by AI?

No job is untouched, but work that mixes physical presence, in-person judgment, and accountability holds up best: skilled trades, hands-on care, emergency response, and supervision. Office work is not all one thing either. Roles built on routine keying face more pressure than roles built on handling people and owning a system. Our rankings page lets you check any job against the same three questions.

Each ridge is a slice of the job's task time.Needs a human 58%AI helps 13%AI does it 29%
The job’s mark

No two jobs leave the same print

Every job gets its own fingerprint, drawn from its code. The amber ridges are the share of task time that still needs a person. Below them, the same ridges are written out in ones and zeros: slate for the work AI helps with, white for the work AI can do.

Office Clerks, General, O*NET-SOC 43-9061. 58% of the job’s task time still needs a human, so 58 of every 100 ridges are amber; slate is what AI helps with, white what AI can do.

What AI can and cannot do

The tasks that make up the job, from , and where AI stands on each today: , (a person does it, with AI speeding it up) or . 58% of the still needs a human.

Each block is one task; its height is its share of working time.Needs a human 58%AI helps 13%AI does it 29%
The job's task list: the parts AI can do are blacked out.Needs a human 58%AI helps 13%AI does it 29%
Operate office machines, such as photocopiers and scanners, facsimile machines, voice mail systems, and personal computers.Needs a human
Answer telephones, direct calls, and take messages.AI does it
Communicate with customers, employees, and other individuals to answer questions, disseminate or explain information, take orders, and address complaints.AI does it
Maintain and update filing, inventory, mailing, and database systems, either manually or using a computer.Needs a human
Compile, copy, sort, and file records of office activities, business transactions, and other activities.Needs a human
Review files, records, and other documents to obtain information to respond to requests.AI does it
Open, sort, and route incoming mail, answer correspondence, and prepare outgoing mail.Needs a human
Compute, record, and proofread data and other information, such as records or reports.AI helps
Complete work schedules, manage calendars, and arrange appointments.AI does it
Type, format, proofread, and edit correspondence and other documents, from notes or dictating machines, using computers or typewriters.AI helps
Inventory and order materials, supplies, and services.Needs a human
Deliver messages and run errands.Needs a human
Collect, count, and disburse money, do basic bookkeeping, and complete banking transactions.Needs a human
Complete and mail bills, contracts, policies, invoices, or checks.Needs a human
Process and prepare documents, such as business or government forms and expense reports.AI does it
Monitor and direct the work of lower-level clerks.Needs a human
Prepare meeting agendas, attend meetings, and record and transcribe minutes.AI helps
Train other staff members to perform work activities, such as using computer applications.Needs a human
Count, weigh, measure, or organize materials.Needs a human
Troubleshoot problems involving office equipment, such as computer hardware and software.Needs a human

Is it better than a person? The evidence

No direct test against people in this job yet. Every study is , and vendor studies are labelled as such.

When could it be replaced?

When AI could largely do this job: 2035–2053

Most likely between 2035 and 2053 (8 in 10 of our scenarios). A range from our of how fast AI improves, how fast employers take it up and what holds it back, not a forecast that the job ends. “” has a strict meaning here. Today’s answer is at the top of the page; this is how it could change.

The sand is the human working years left, measured in the same 40-year glass for every job, so a safe trade starts nearly full and an exposed job with a thin layer.

The sand is the human working years left, in the same 40-year glass for every job.Years still needing a humanYears run out

How this job could shift, year by year

Where the job could sit on our scale each year to 2060, across the ten behind its .

Today
Will AI replace this job?
A little.
By 2045
80%
of our scenarios have AI largely doing this job by 2045 (Largely.)
0% still have it mostly needing a person (A little. or Nah.)
By 2060
100%
of our scenarios have AI largely doing this job by 2060 (Largely.)
0% still have it mostly needing a person (A little. or Nah.)

We run this job as ten scenarios spread across its replacement range. In each, the score moves towards the bottom band (Largely: AI could largely do the job) by the year that scenario reaches it, slowly at first and faster later, as adoption usually goes. Each bar splits the ten by the band they put the job in. The model stops at 2060. How the timeline works

Share of this job's scenarios in each verdict band, today to 20600%25%50%75%100%2026: 100.0% of scenarios: AI could do a little of this job (A little.)100%Today2030: 30.0% of scenarios: AI could do a little of this job (A little.)30%2030: 70.0% of scenarios: AI could partly do this job (Partly.)70%20302035: 40.0% of scenarios: AI could partly do this job (Partly.)40%2035: 30.0% of scenarios: AI could mostly do this job (Mostly.)30%2035: 30.0% of scenarios: AI could largely do this job (Largely.)30%20352040: 10.0% of scenarios: AI could partly do this job (Partly.)10%2040: 30.0% of scenarios: AI could mostly do this job (Mostly.)30%2040: 60.0% of scenarios: AI could largely do this job (Largely.)60%20402045: 20.0% of scenarios: AI could mostly do this job (Mostly.)20%2045: 80.0% of scenarios: AI could largely do this job (Largely.)80%20452050: 100.0% of scenarios: AI could largely do this job (Largely.)100%20502055: 100.0% of scenarios: AI could largely do this job (Largely.)100%20552060: 100.0% of scenarios: AI could largely do this job (Largely.)100%2060
Will AI replace the job?Largely.Mostly.Partly.A little.Nah.
Share of this job's scenarios in each band, year by year. Updated with every release.
Show the data
YearLargelyMostlyPartlyA littleNah
Today (2026)0.0%0.0%0.0%100.0%0.0%
20300.0%0.0%70.0%30.0%0.0%
203530.0%30.0%40.0%0.0%0.0%
204060.0%30.0%10.0%0.0%0.0%
204580.0%20.0%0.0%0.0%0.0%
2050100.0%0.0%0.0%0.0%0.0%
2055100.0%0.0%0.0%0.0%0.0%
2060100.0%0.0%0.0%0.0%0.0%

What’s stopping AI taking over?

The things that keep this work with people, strongest first. Each is scored 0 to 100 from work context, licensing and the evidence we have.

Clients want a personFace-to-face contact is rated 4.6 and physical closeness 3.6 out of 5; caring for or serving people is 2.6 out of 5 in importance.
Evidence gapNo study yet compares AI with people doing this job, so employers have no proof it is good enough.
LiabilityMistakes are rated 2.4 out of 5 for consequence and decisions 3.8 out of 5 for impact; someone has to answer for them.
Physical work50% of the task time is physical; robots have been shown on 95% of that time.
RegulationWorkers rate responsibility for others' health and safety 2.5 out of 5.
LicensingUsual entry requirement (BLS): high school diploma or equivalent, then short-term on-the-job training.

What would it cost to hand the work to AI?

The share of the year AI could handle (820 of 2,080 hours a year), priced two ways. Both are ranges, not quotes.

AI model usage, a year
$80–$8,200
A person’s wage for the same hours
$11,940–$25,480

AI cost covers model usage only: no integration, licences, oversight or the human time still needed to review the work. Human cost is the wage for the same hours, without benefits or overheads. As of 2026-10.

Robots and humanoids

AI software can only take the work at a screen. The rest needs a robot that can do it.

51%
of the task time is physical work
Mobile robots
the kind of robot the physical work would need
Commercial in warehouses, hospitals and some outdoor sites; hands are still limited.

Source: Anthropic Economic Index, 'What work can robots do?' (30 September 2026); O*NET 31.0 task weights.

Which AI skills does this job lean on?

The job’s task time split by what an AI model would need to be good at, and where models stand today.

Each star is a task, grouped by the AI skill it leans on.Needs a human 58%AI helps 13%AI does it 29%
Writing · 12% of time
Strong
Drafts, edits and translates most routine documents at professional quality.
Analysis · 10.6% of time
Strong
Reliable on structured data and rules; uneven on judgement calls with thin information.
Coding · 2.6% of time
Strong
Agents complete many routine software tasks end to end; larger systems still need people.
Vision and design · 0% of time
Good
Reads documents, images and layouts well; specialist imaging needs dedicated, approved tools.
Speech · 14.4% of time
Good
Voice agents handle routine calls and live interpreting; complex or sensitive calls still go to people.
Planning and agents · 25.4% of time
Emerging
Multi-step agents work in narrow, well-tooled workflows; open-ended coordination is unreliable.
Physical manipulation · 27.3% of time
Early
Robots handle structured, repetitive handling; general dexterity outside fixed settings is not commercial.
Care and persuasion · 7.6% of time
Limited
Can script, coach and advise; trust, presence and accountability still need a person.
Amber matter holds its orbit, slate circles the inner disc, white falls in.Needs a human 58%AI helps 13%AI does it 29%
How exposed is it?

Still needs a human: 65/100↑ safer

The amber matter on the outside holds its orbit: that is the task time that still needs a person. The slate matter circling the inner disc is the work AI helps with. The white matter is the share AI can do; it spirals in and is gone over the edge.

The matter follows the job’s task split: 58% needs a human, 13% AI helps, 29% AI does it. Still needs a human: 65/100 ↑ safer. Will AI replace them? A little.

What the AIs say

We asked four AI assistants the same question: will AI replace this job in the next 10 years? One word (Yes, Partly or No), then one sentence. Our own answer, from the data: Still needs a human: 65/100 ↑ safer. Will AI replace them? A little.

ChatGPTPartly

AI will automate many routine clerical tasks, but humans will still be needed for judgment, communication, exception handling, and oversight.

gpt-5.5-2026-04-23 · asked 2026-10-03
ClaudePartly

AI will automate many routine clerical tasks, but human clerks will likely still be needed for complex judgment calls, interpersonal communication, and overseeing automated systems.

claude-sonnet-5 · asked 2026-10-03
GeminiPartly

While AI will automate the vast majority of routine data-entry and administrative tasks, human workers will still be needed to manage complex workflows, handle sensitive exceptions, and provide interpersonal communication.

gemini-3.8-flash · asked 2026-10-03
PerplexityPartly

AI will automate many routine clerical tasks and reduce some jobs, but human judgment, coordination, and accountability will keep office clerks from disappearing entirely.

sonar · asked 2026-10-03

Assistants answer from what they learned in training (Perplexity also searches the web), so they can be confidently wrong, and the same question can get a different answer tomorrow. Our score is built from task data and graded evidence. Answers collected through DataForSEO.

Cite this page

NeedsAHuman.com (2026). Will AI replace Office Clerks, General? A little. Still needs a human: 65/100, higher is safer; release 2026-Q4. https://needsahuman.com/jobs/office-clerks-general/ (accessed 4 October 2026).

Scores change with each , so cite the release. The data is open under : credit NeedsAHuman.com with a link. Open data · Press

Put the badge on your site

The badge updates itself with each release and links back to this page.

Sources

  • Tasks and work context: 31.0, ().
  • Jobs, pay and projections: US , and 2025–35.
  • How AI is used today: ; Microsoft Research, .
  • What AI can do: our task ratings ( r1) and the quality evidence register.
  • UK names and employment: coding index and .

How each score is built: methodology. Every figure on this page: open data. Release 2026-Q4.