How AI Changes Freelance Work Without Making Every Freelancer Replaceable

The short answer: AI changes freelance tasks and demand unevenly

AI puts real pressure on some freelance services, especially low-context digital outputs that a client can specify, compare, and check cheaply. But that pressure does not settle the fate of a freelancer as a whole. Projects also involve briefing, context, choices, checking, communication, and responsibility. The International Labour Organization’s [2025 Global Index of Occupational Exposure](https://www.ilo.org/resource/news/one-four-jobs-risk-being-transformed-genai-new-ilo%E2%80%93nask-global-index-shows) estimates potential exposure across occupations, not actual job losses or any freelancer’s odds; it reports that transformation is more likely than full replacement. Meanwhile, the UCL study [Winners and losers of generative AI](https://discovery.ucl.ac.uk/id/eprint/10204198/) finds divergent changes in online freelance postings: demand fell for skill clusters it classified as substitutable, while complementary and unaffected clusters moved differently. Those results concern platform postings, not every client relationship or a guarantee that workers can reach growing work. Treat them as reasons to inspect, not predict. Review recent paid projects and pipeline, identify which deliverables clients still buy, then test an affordable workflow upgrade or adjacent offer before making a larger change. Human judgment may help define value, but neither experience nor review guarantees demand. The next step is task-level evidence, not automatic retraining or passive reassurance.

Sources: One in four jobs at risk of being transformed by GenAI, new ILO–NASK Global Index shows; Winners and losers of generative AI: Early Evidence of Shifts in Freelancer Demand

Start with tasks and transactions, not the label ‘freelancer’

An occupation label is too broad to tell a freelancer which part of their work may change. Start with the last three to five projects and write down what clients paid for. Separate repeatable production from briefing, context gathering, choosing among alternatives, integration, verification, communication, revisions, and responsibility for delivery. The International Labour Organization’s “Generative AI and Jobs: A Refined Global Index of Occupational Exposure” supports this task-level view. Its 2025 index combines nearly 30,000 occupational tasks with expert validation, AI-assisted scoring, and harmonized employment data. The summary estimates that 25% of global employment is in occupations potentially exposed to generative AI, rising to 34% in high-income countries. The ILO stresses that these figures describe potential exposure, not observed job losses; transformation is more likely than full replacement because many tasks still involve people. For one representative project, make a ledger with a row for each task. Note its inputs, acceptable result, how errors are found, and what happens if one reaches the client. Add paid time or revenue shares only when records support an estimate. Also note recurrence, review effort, client permissions, and who owns the consequence. Unknowns stay blank; do not turn rows into a personal replacement percentage. Consider two copy assignments. A short piece from a complete supplied brief, fixed format, and familiar subject may mainly require producing text against clear criteria. Researched editorial work may also require locating and checking sources, resolving conflicting facts, and deciding what the reader needs. These differences affect the task bundle; they do not prove editorial judgment cannot be automated or that a buyer will keep paying the same freelancer. Similarly, resizing an approved visual into specified variants is narrower than developing an art-directed campaign from incomplete inputs and integrating assets into client channels. Buyers can still change budgets or suppliers. Ask whether needed inputs are available, errors can be caught at acceptable cost, permissions allow the workflow, domain context affects choices, and someone can hand off a dependable result. Do not treat “human” as a synonym for protected. The UCL study “Winners and losers of generative AI: Early Evidence of Shifts in Freelancer Demand” shows why task distinctions matter commercially. Researchers grouped online-platform postings into 116 skill clusters and classified them as substitutable, complementary, or unaffected by large language models. They report that postings for substitutable skills such as writing and translation fell 20–50% relative to a counterfactual trend, with the sharpest decline in short-term jobs. This concerns categorized postings on one platform, not every freelancer’s paid hours or the fate of an individual contract. Pair the ledger with your own transactions: record what was scoped, bought, revised, and paid.

Sources: One in four jobs at risk of being transformed by GenAI, new ILO–NASK Global Index shows; Winners and losers of generative AI: Early Evidence of Shifts in Freelancer Demand

Why freelance demand can split by deliverable and time horizon

The evidence points in more than one direction because freelance demand is measured in different ways: contracts and earnings, postings across skill clusters, or marketplace categories and contract values. Some deliverables lose demand while other work expands or changes. That does not show every freelancer can move from a shrinking service into a growing one.

In project markets, a client can post a bounded task, alter its brief or budget, and hire by contract. In The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market, Hui, Reshef, and Zhou study Upwork, a platform for remote, short-term projects. Using freelancer histories from 2022 through April 2023 and a difference-in-differences design around ChatGPT’s November 2022 release, they compare more-affected services, including writing, with less-affected ones. The authors report 2% fewer monthly jobs and 5.2% lower monthly earnings for the more-exposed group relative to the comparison group after the release. They report a similar direction for design and image services after image-model releases.

These short-term averages from one platform are not a forecast for a particular freelancer. The study does not observe every income source or each contract’s task mix. It does show that capability changes can coincide with lower contract flow and earnings, not only higher productivity.

A different view appears in Winners and losers of generative AI: Early Evidence of Shifts in Freelancer Demand, recorded by UCL Discovery. The researchers divide postings from a leading freelance platform into 116 skill clusters and classify them as substitutable, complementary, or unaffected by language models. The record reports that postings for substitutable skills such as writing and translation fell 20–50% relative to a counterfactual trend, especially for short-term jobs lasting one to three weeks. Postings increased in complementary and unaffected clusters, but complementary demand was uneven: machine-learning programming postings rose 24% and AI-powered chatbot development nearly tripled, while novice-worker demand declined in general.

The UCL study measures postings and relative trends, not completed contracts, total earnings, or net employment. Its platform categories answer a different question from the first study, which follows freelancer-level jobs and income in more- versus less-exposed services. Posting declines in substitutable clusters can coexist with other outcomes by category, period, or worker group. The percentages should not be averaged: one unit is freelancer-month outcomes, the other posting volume by skill cluster against an estimated counterfactual.

Upwork’s Generative AI and the Future of Work: How New Technologies Are Impacting Freelancers adds a third, platform-specific account. It describes lower earnings in writing and translation, particularly for lower-value work, alongside growth in some technology and business-operation categories and a mixed creative-sector picture. Using proprietary marketplace data through 2023 and synthetic controls, the report helps show category movement on that platform. Its definitions and commercial setting bound interpretation; growth in selected areas does not establish that lost demand is offset.

Together, these sources support neither “AI only helps freelancers” nor “freelance work is disappearing.” Deliverable, duration, price tier, worker group, platform, and time window matter. Faster production may help a freelancer complete more work, while clients buy fewer units of a standardized output. Growth in a neighboring skill cluster shows demand there; it does not establish that the same person can win those contracts, acquire the prerequisites quickly, or preserve income during a transition.

For a practical decision, track two things separately: whether clients still buy the service you sell, and whether an adjacent service is reachable through your expertise, samples, relationships, and available learning time. Growth in AI-related postings is evidence about those postings, not a general instruction to become a machine-learning engineer. If the underlying client need persists but the purchased bundle changes, test a nearby offer against actual briefs and responses. If current work weakens and the adjacent path requires costs or an income gap you cannot absorb, include that constraint in the comparison. These studies sharpen what to watch; your own pipeline helps decide whether a specific experiment is worth making next.

Sources: The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market; Winners and losers of generative AI: Early Evidence of Shifts in Freelancer Demand; Is Generative AI a Job Killer? Evidence from the Freelance Market; How Generative AI Adds Value to the Future of Work

Separate tool capability, buyer adoption, and paid demand

A tool's ability to produce a plausible first draft establishes neither dependable delivery under a real brief nor a buyer's decision to use it. A client might test a tool, retain a freelancer while asking for more output, reduce contract size or frequency, or purchase a different service. Evidence at one step cannot stand in for the next. The International Labour Organization's 2025 global exposure index uses task-level data, expert input, AI-assisted scoring, and harmonized employment data to measure potential occupational exposure, while warning that the figures are not actual job losses. This global measure helps explain why tasks matter; it does not establish what a particular freelance client will buy.

A practical chain has five links. First is capability: can a tool perform a bounded task, such as drafting, sorting supplied material, or making an initial image variation? Second is reliability: does the result meet this client's facts, format, context, quality threshold, and error costs, including checking time? Third is adoption: does the client permit the tool, have suitable data, and trust the process? Fourth is purchasing or scope change: does the client alter the brief, budget, turnaround, or hiring decision? Fifth is labor demand and income: do those changes affect contracts, repeat work, rates, or earnings over time? Each transition can fail or take another path. A capability demonstration says little about permission; adoption does not tell us whether the client will buy more, less, or different freelance work.

Upwork's 2025 report, “AI Trends on the World’s Work Marketplace: How AI Is Reshaping the Way Humans Work,” illustrates why these links should not be collapsed. Drawing on platform activity across more than 130 work categories and billions in freelancer earnings, it describes augmentation, substitution, and category evolution together. It reports selected lower-complexity writing, market-research, data-extraction, and photography tasks as facing substitution, while also reporting AI-related activity and growth in some other categories. This is evidence about one marketplace's recorded activity, not a forecast for every freelance market: platform clients, category definitions, and commercial incentives bound the result. A rising category can coexist with a weaker market for a particular deliverable and does not show which affected workers captured new work.

The report also cites an October 2024 internal survey of 673 clients: 66% expressed high trust in freelancer-delivered work using AI tools, compared with 26% for AI-only output. Those answers describe stated trust among surveyed platform clients, not contract conversion, earnings, or an individual freelancer's retention. A separate Upwork analysis, “How Generative AI Adds Value to the Future of Work,” used platform data and synthetic control around ChatGPT's November 2022 public release. It reports 2.4% more platform job posts overall and 1.3% higher earnings per new contract, alongside reduced demand in writing and translation, particularly lower-value work. Overall and category findings can coexist because a marketplace average may hide losses in particular services. Both reports are platform-specific; the earlier analysis concerns short-term outcomes around one release. Neither settles what is happening in direct-client, local, regulated, or later-period work.

For your own decisions, track signals close to paid work: client questions and brief changes, contract frequency and size, unpaid review time, revisions, proposal conversion, repeat work, and realized earnings. Compare similar services and client types over a defined period, noting other plausible causes such as seasonal demand, client budget changes, or your availability. If clients ask for more output at the same price, that is a scope signal; falling brief volume is a pipeline signal; rising review time affects margin. None alone proves AI caused the change. A short ledger is a decision aid, not a causal study or personal displacement probability. Its practical value is showing whether a bounded workflow test, revised offer, or closer monitoring fits what buyers actually request and pay for.

Sources: AI Trends on the World’s Work Marketplace: How AI Is Reshaping the Way Humans Work; How Generative AI Adds Value to the Future of Work; One in four jobs at risk of being transformed by GenAI, new ILO–NASK Global Index shows

What market evidence says—and where reassurance breaks

Freelance AI evidence shows both task shifts and losses. Upwork’s 2025 “AI Trends on the World’s Work Marketplace” describes substitution in selected low-complexity categories alongside growth in some AI-related and more complex work. Independent studies of online freelance markets report adverse short-term outcomes in some exposed services. The evidence shows uneven change with material downside. The clearest counterweight to a simple “experienced workers will be fine” argument is “The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market.” This October 2024 working paper analyzes roughly 80,000 searchable Upwork freelancer profiles and monthly job and earnings histories from January 2022 through April 2023. Its difference-in-differences comparisons report that, after ChatGPT’s release, more-exposed freelancers had 2% fewer monthly jobs and 5.2% lower monthly earnings on the platform than less-exposed groups. For image-generation releases, the reported declines were 2.1% in monthly jobs and 5.2% in earnings. Past performance did not moderate the effects; results were consistent with top freelancers being disproportionately affected. These short-term group estimates are not an individual forecast: they cover platform work after early tool releases, not all income or later outcomes. They challenge the idea that reputation alone insulates a service when buyers can substitute for it. Another study helps explain why platform gains coexist. In “Winners and losers of generative AI: Early Evidence of Shifts in Freelancer Demand,” a peer-reviewed study examines online job postings grouped into 116 skill clusters. The UCL record reports that postings for skills classified as substitutable, including writing and translation, fell 20–50% relative to the estimated counterfactual trend, with the sharpest declines in short-term jobs. Complementary and unaffected clusters saw increased demand, but the pattern was uneven: the study reports growth in machine-learning programming and AI-powered chatbot development while novice-worker demand declined generally. These are posting trends classified by the researchers, not realized earnings or proof that affected freelancers moved into growing clusters. This shows divergence, not a ready-made transition route. Upwork’s own “How Generative AI Adds Value to the Future of Work” reaches a more positive platform-level picture for the period it studies. Using a synthetic-control design on Upwork work groups and data from 2021 through the third quarter of 2023, it reports a 2.4% increase in total job posts and 1.3% higher earnings per new contract overall. Category findings were mixed: writing and translation earnings fell, particularly for lower-value work, while other groups gained. Upwork’s later “AI Trends on the World’s Work Marketplace” describes continued substitution in selected transactional writing, research-extraction, and photography tasks, alongside growth in AI-related and some nontechnical categories. Its 130-plus categories and internal client trust research describe Upwork’s system; they cannot establish durable demand or income across freelance markets. One study measures jobs and earnings for more- versus less-exposed freelancers after releases; another classifies posting trends across skill clusters; Upwork’s analyses aggregate its own contracts, categories, and survey responses over different periods. Aggregate growth can coexist with subgroup losses. A posting is not a contract; contract counts, earnings, and stated trust measure different things. Comparison groups, categories, periods, and platform populations differ, so averaging results into one forecast would hide useful variation. The practical correction is modest but important: visible accountability is not a proven protective moat. A client may value checking or judgment, but the evidence here does not show that offering a human review step causes retention or protects revenue. If contracts, prices, or margins fall persistently in your own client mix, experience does not make that pattern irrelevant. Stable demand can justify a bounded workflow or offer test, but neither platform growth nor survey-reported trust guarantees you will capture it.

Sources: The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market; Winners and losers of generative AI: Early Evidence of Shifts in Freelancer Demand; AI Trends on the World’s Work Marketplace: How AI Is Reshaping the Way Humans Work; How Generative AI Adds Value to the Future of Work

Can human review or expertise make a freelance offer more durable?

Human review or expertise can make a freelance offer more useful when it changes the result: a decision fits the brief, material errors are caught, requirements are reconciled, or work can be integrated and explained. A person beside generated output does not establish that value. Review takes time and may add cost; clients must care about the risk reduced or outcome improved. Ask: “What consequential failure does my contribution prevent, and can the client see and value that difference?” That contribution may begin before production. A client can describe a deliverable without knowing which question to ask, what evidence applies, or which constraint makes a polished answer unusable. Expertise can clarify the goal, request context, and expose conflicts. During production, a freelancer may compare alternatives against requirements, test claims, reconcile edits, or check that separate pieces work together. At handoff, explaining a tradeoff or recording a limitation can make the result usable. These are possible sources of value, not tasks inherently beyond automation; their importance depends on the assignment, tools, client rules, and cost of error. “Review” can describe different work. Checking a citation against an authoritative source may catch a consequential mistake. Cosmetic edits may add little if the client wants a low-cost draft and accepts its limits. A service boundary should say what is checked, against which materials or standards, what is outside scope, and who approves the result. It should not imply that review guarantees correctness. A concrete but narrow signal for verification work appears in Upwork’s *Monthly Hiring Report: AI Amplifies Demand for Human Skills*. Fact-checking and video editing were among its ten most in-demand AI-related skills hired on its U.S. marketplace from August 1–25, 2025. This is a dated snapshot of new contract activity, not proof of durable demand across markets or a protected career path. The counterevidence matters for anyone making expertise the whole strategy. *The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market* analyzes public histories for roughly 80,000 Upwork freelancers from January 2022 through April 2023, using a difference-in-differences design around model releases. In more exposed writing services, it reports 2% fewer monthly jobs and 5.2% lower monthly earnings relative to less-exposed occupations after ChatGPT. Past performance and employment history did not moderate the adverse effect; there is suggestive evidence of larger effects for top freelancers. This short-run result from one project-based platform does not measure every client relationship or later market. It does challenge the claim that experience or reputation alone shields a freelancer from changed demand. Upwork’s *AI Trends on Upwork: How AI Is Reshaping the Way Humans Work* reports surveyed clients expressed high trust in freelancer work using AI tools more often than AI-only output, while also identifying substitution in selected low-value transactional categories. A trust response is not a contract, and a platform trend does not guarantee an individual review offer will sell. Human involvement matters commercially when it improves a result the buyer recognizes and will purchase. Make a review-led offer testable: name one risk or decision, define the checks included, and show an artifact such as a source trail, test result, or exception note. Ask clients whether it would change their choice, reduce rework, or ease approval. Track checking time, revisions, acceptance, repeat requests, and whether price covers the work. If clients do not value a useful check, revise the scope or investigate an adjacent service. If review adds cost without improving reliability or usefulness, calling it “human review” will not make it durable. Sell a bounded outcome and honest responsibility, not a claim to be human or AI-proof.

Sources: The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market; Monthly Hiring Report: AI Amplifies Demand for Human Skills; AI Trends on the World’s Work Marketplace: How AI Is Reshaping the Way Humans Work

Compare an upgrade, an adjacent move, and a larger change

Upgrade first when clients still buy the outcome and a tool can reduce friction without shifting unacceptable risk onto you. Move adjacent when clients still need the result but buy a different bundle. Investigate a larger career change when exposed work dominates, several business indicators worsen, and feasible upgrades or adjacent options do not fit your income and life constraints. Compare choices; not mandatory stages. | Path | Evidence and constraints | First bounded test | | --- | --- | --- | | Upgrade current offer | Clients still request the outcome; tool use is permitted and review manageable | Test one low-risk task; count checking, revisions, total time, and client response | | Adjacent service | Clients' needs persist but briefs or buying criteria change; expertise transfers | Ask clients about the need; make one relevant sample or small scoped offer | | Larger change | Pipeline, repeat work, contract value, or margins weaken across clients or time | Compare real prerequisites, training cost and time, income floor, location, health, and care duties | An upgrade is useful only if the whole job improves. A writer might use an approved tool to organize notes or prepare a rough outline, while retaining responsibility for sources, argument, factual checks, and client voice. Verification and revisions can erase drafting time saved. If confidentiality, quality, or client rules make the workflow unsuitable, apparent efficiency is no business gain. An adjacent move changes what the client buys while retaining some existing expertise. A writer might test research and editorial strategy; a designer, art direction and production integration; a routine researcher, source verification and decision support. These are hypotheses, not guaranteed transitions. Look for buyer evidence: changed briefs, requests, sample reviews, or small paid work. Identify missing capabilities and how to demonstrate them. Choose a course, project, feedback, or self-study for the gap; a credential alone does not prove readiness. Market signals can guide investigation, not select a move for an individual. “Winners and losers of generative AI: Early Evidence of Shifts in Freelancer Demand,” in UCL Discovery, reports an analysis of postings in 116 skill clusters. Substitutable clusters such as writing and translation fell 20–50% relative to a counterfactual trend, especially for short jobs; complementary and unaffected clusters increased overall, with mixed results among complements. This is one platform’s posting activity, not proof a writer can enter a growing cluster. Upwork's “Monthly Hiring Report: AI Amplifies Demand for Human Skills” lists fact-checking and video editing among its top AI-related skills hired in the United States during August 1–25, 2025. This dated signal suggests task recombination, not durable demand or a universal reason to retrain as an AI engineer. The ILO's “One in four jobs at risk of being transformed by GenAI, new ILO–NASK Global Index shows” describes potential task transformation, not a personal displacement forecast. A larger change is worth investigating when commercial indicators persistently weaken and nearer options fail a practical test. One slow week or lost contract is not a trend: compare inquiries, repeat work, scope, price, and margin across time and clients, allowing for seasonality and client-specific causes. Compare options with income needs, location, entry requirements, training time and cost, health, and family responsibilities. A degree may fit a target occupation requiring it; it is a mismatch if the goal is using AI in an existing service. A short course can structure learning but does not itself prove job readiness. Learn for the target work: problem framing, data literacy, evaluation, domain knowledge, and verification endure longer than tool interfaces. Prefer the smallest reversible experiment that could change your decision. If finances or care make experimentation costly, gather client evidence before reducing paid work. Move from workflow test to adjacent offer, or investigate a larger change, only when evidence and constraints support it. No path guarantees employment.

Sources: Winners and losers of generative AI: Early Evidence of Shifts in Freelancer Demand; Monthly Hiring Report: AI Amplifies Demand for Human Skills; One in four jobs at risk of being transformed by GenAI, new ILO–NASK Global Index shows

What can you learn from a 30-day test?

A 30-day test cannot predict your career or prove AI caused a market change. It can answer a narrower question: does one permitted workflow save time after checking, do clients accept a service variation, and is there reason to investigate an adjacent offer? Treat the month as observation, not forecast. A faster draft shows task capability; it does not show buyers adopted the tool, changed purchases, or reduced demand for your work.

Choose one frequent, low-risk task within work you understand, such as outlining from a brief or making routine variations from approved material. Avoid work where mistakes could harm a client or you cannot verify the result. Before changing the workflow, record a baseline from comparable recent projects: production and checking time, revisions, delivery against the agreed standard, and client response.

Set boundaries first. Check client instructions, contract terms, confidentiality, and data-handling rules. Use only permitted material and define what remains under your review. Stop if you cannot reliably catch material errors, checking erases time saved, the client disallows the workflow, or quality falls below the agreed bar. Quicker initial output is no gain if full delivery takes longer or shifts unpriced risk onto you.

Test buyer response separately. Describe a bounded service variation to suitable clients: what is included, what you verify, and what the client receives. Track replies, scope requests, contracts, and fees separately; a few conversations do not represent a market. Upwork's “AI Trends on the World’s Work Marketplace” reports substitution in selected lower-complexity categories alongside growth or augmentation elsewhere on its platform. This supports checking your buyers, not assuming a new offer will sell.

At day 30, compare like with like. Did total delivery time fall without quality or permission problems? Did clients accept the changed scope on workable terms? Does the underlying need remain, even if buyers want a different bundle? A yes to the first supports refining the workflow, not claiming career security. Persistent need with weak response may justify testing an adjacent offer. If pipeline, contract size, and margins worsen across comparable work, this test cannot explain why; compare larger changes against income needs, location, training time, cost, health, and family responsibilities.

The evidence calls for restraint. “The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market” studies freelancers on one large online platform after model releases and reports average employment and earnings declines in more affected occupations; past performance did not insulate workers. Its short-term platform results do not predict an individual's outcome or explain one client's decision. Upwork's category trends likewise cannot show that a freelancer can capture expanding work. Personal notes are narrower still: they guide the next test, not settle attribution or long-term demand.

Choose among three next steps: keep or refine the workflow if it passes time, quality, and permission checks; test an adjacent service if client needs persist but the purchased bundle shifts; or compare a larger change if commercial indicators keep worsening and feasible upgrades do not address them. Preserve paid work where possible. The free task-level checker can organize which work deserves attention, but its signals are not a validated probability of displacement. A personalized roadmap can compare scenarios against constraints; it cannot guarantee income or employment.

Sources: AI Trends on the World’s Work Marketplace: How AI Is Reshaping the Way Humans Work; The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market

Your next move: inspect one real offer before changing careers

Review one recent project this week. List the tasks the client paid for, what a tool might draft, where checking or judgment mattered, and what the client valued. Compare that offer with repeat contracts, project size, revision load, and your visible pipeline. The ILO’s “One in four jobs at risk of being transformed by GenAI, new ILO–NASK Global Index shows” supports looking at task bundles; its occupational exposure figures describe potential exposure, not observed job losses. “The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market” reports short-term declines in jobs and earnings among more-exposed Upwork freelancers. That platform result is not an individual forecast or evidence about every freelance market.

If demand is steady, test one permitted workflow change that could save production time without worsening verification or quality. If clients still need the outcome but buy a different bundle, investigate an adjacent service that uses skills or relationships you already have. If pipeline, contract value, or margins keep weakening, compare a larger change before committing. Include geography, salary floor, training time and cost, health, and family responsibilities. A 30-day test can inform the choice but cannot prove what caused a market shift. The free task checker may help organize change-pressure signals; it does not provide a validated probability of displacement.

Sources: One in four jobs at risk of being transformed by GenAI, new ILO–NASK Global Index shows; The Short-Term Effects of Generative Artificial Intelligence on Employment: Evidence from an Online Labor Market