A Noncompete Cut Recruiter Mobility by Half
A freelance recruiter opened a short-term offer from a finance firm. The wage was visible. Farther down, a noncompete restricted what work could follow after the project ended. Every version of the contract also contained a nondisclosure agreement.
Across roughly 14,000 real offers, a 2026 field experiment turned that scene into evidence. Bo Cowgill, Brandon Freiberg, and Evan Starr worked with two finance firms, randomizing wages and the presence, salience, and duration of the noncompete.
Removing the clause increased movement between the competing employers by 36% to 52%. It raised the recruiters’ combined earnings from the two firms by 12% to 17%. Another presentation of the experiment, reported in the OECD Employment Outlook 2026, found that a clause placed deep in the contract reduced subsequent employment with the competing firm by 57%.
“By half” is shorthand for that 57% outcome. It does not mean every noncompete cuts every kind of mobility by the same amount.
Most arguments about noncompetes start after the evidence has become tangled. An employer says it trained the worker, disclosed a customer list, or revealed a process. A worker says the restriction was boilerplate, appeared late, or covers jobs unrelated to the secret. A court sees the dispute only after a move. Wage data then mix people who selected into different firms, roles, and contracts.
Randomization separated part of that knot. The offer changed while the worker pool and job stayed comparable. One management choice emerged from the contract language: protect a defined asset or restrict a person’s next market. Those actions overlap less than many standard templates assume.
Separating company assets from worker experience becomes more expensive when companies compete for AI skills. PwC’s 2026 Global AI Jobs Barometer, based on more than one billion job advertisements across 27 countries and territories, reports an average 62% wage premium for jobs requiring AI skills. Its method does not measure noncompetes. Scarcity turns a mobility restraint into a capacity decision.
A practical response begins with an offer-level review: name what needs protection, test a narrower tool, show the restriction before the candidate gives up other options, price any remaining restraint, and measure what happens to hiring and mobility. Universal boilerplate cannot do that work.
A finance contract tests a quiet restraint
Before turning the result into a slogan, narrow its scope. Participants were freelance recruiters taking short contracts from two finance firms. They were not a representative sample of salaried employees, software engineers, or executives. The paper is a working paper, not a final judgment for every labor market.
Its design is still rare. Across the offers, the researchers changed four features: the wage, whether a noncompete appeared, how visible it was, and how long it lasted. Because all contracts also included an NDA, the experiment compared confidentiality protection alone with confidentiality protection plus a mobility restriction.
Keeping the NDA in every version matters. If a recruiter could access a candidate list, hiring strategy, or process, the firms did not abandon information protection. They tested whether preventing later work for a rival added value beyond an agreement not to disclose secrets.
Recruiters also occupied a revealing position in the labor market. Their paid work combined employer information with portable skill. They learned which profiles converted, how a firm framed a role, and which sourcing channels performed. Judgment about candidates, search technique, and the ability to run a process traveled with them. A broad clause treats both categories as if they were the same property.
They are not. A customer list exported from a private system is company information. A recruiter’s ability to conduct an interview is human capital. A current compensation plan may be confidential. General knowledge that one job pitch failed is experience. Contract review gets difficult at the boundary, but difficulty does not erase the boundary.
Consider two versions of a protection request. The first says a recruiter may not download, reveal, or use a named firm’s candidate database, private compensation bands, or unreleased hiring plan. The second says the recruiter may not work for another finance firm for a period after the contract. The first follows the information. The second follows the person.
Narrow cases may justify both kinds of protection. A senior commercial leader may carry current pricing strategy and customer relationships into an immediately substitutable role. A founder selling a business may agree to a restraint as part of the transaction. A worker may receive paid gardening leave while time-sensitive information loses value. Those cases require facts that boilerplate does not supply.
Short-term recruiting makes the timing visible. A project might last weeks while a restraint lasts longer. The worker could finish the assignment but remain unable, or believe they were unable, to take the next available contract from the other firm.
For a freelancer, that timing converts a sentence into an income event. A salaried worker might encounter the restriction in a search months later; the freelancer’s next project is the earnings path. The clause sits quiet during the contract, then becomes active exactly when the worker needs another buyer.
Randomized wages added another test. If workers knowingly accepted restricted jobs in return for higher pay, the offer data would reveal a compensating premium. The researchers found no evidence that workers chose noncompete jobs for higher wages. The restriction was not behaving like a clearly priced benefit exchange.
Compensation may still change whether a restraint is fair or lawful. Many countries explicitly require it, and negotiated executive arrangements can look different. In these offers, however, the market did not automatically price the lost option.
Recruiters moved after the clause disappeared
Mobility sounds abstract until it is placed in sequence. A recruiter accepts Firm A’s short project. Firm B later has work. Without a noncompete, the recruiter can evaluate the second offer on pay, timing, and fit. With the clause, the recruiter must also estimate whether Firm A will object, whether the restriction applies, and what a dispute might cost.
Behavior changed in that sequence. Removing noncompetes increased movement between the firms by 36% to 52% and total earnings from both firms by 12% to 17%. Different measures and specifications produce the ranges, so keep them as ranges.
Viewed from the opposite direction, the OECD reports that placing the clause deep in the offer reduced subsequent competing-firm employment by 57%. The numbers are compatible descriptions of related comparisons, not interchangeable universal constants.
Earnings changed because the second opportunity mattered. Firm A could pay less in total when a worker’s route to Firm B narrowed. Firm B also paid less because fewer recruiters arrived. The clause affected more than retention at the first firm; it changed competition between two buyers of labor.
Lawsuits are not required for a noncompete to change behavior. A rational worker includes the probability of enforcement, legal cost, delay, reference risk, and reputational conflict. Even when each probability is uncertain, the combined expected cost may exceed the value of a short contract.
Cowgill, Freiberg, and Starr interpret the results through inattention and uncertainty about enforcement. Many workers appeared unaware of the noncompete until a post-employment communication from the firm made it salient. The clause therefore has two lives: nearly invisible at acceptance, unmistakable when a competing opportunity appears.
Consent gets harder to interpret under this pattern. A signature proves that text was presented in a contract. It does not prove that the worker understood how it would attach to the next opportunity, expected enforcement, or received a price for surrendering the option.
Retention metrics change too. Suppose restricted employees have lower exit rates. A dashboard may label the figure loyalty or workforce stability, yet a lower exit rate can also reflect a blocked destination. Better work and a threatened move look identical until the company examines rejected offers, release requests, and post-employment contacts.
Hiring suffers on the other side. The OECD reports that the experiment’s clause reduced the overall number of job applicants as it reduced later competing employment. A firm gains leverage over signed workers while shrinking its future pool of experienced hires if competitors use the same tool.
AI talent makes that feedback loop easier to see. A company may want to keep an engineer who has learned its model pipeline or an operator who deploys an agent into a regulated workflow. If every firm broadly restricts those people, knowledge and experience stop moving toward the teams best positioned to use them. The same company that likes its own restraint may dislike the rival clause blocking its next hire.
Treat the AI connection as an inference from mobility evidence. The experiment did not study AI workers. PwC’s reported AI-skill premium is a job-ad association, not proof that noncompetes caused the premium. Together, the sources identify a decision worth testing: in a scarce market, how much recruiting capacity does a firm lose when it treats mobility as a threat by default?
Recruiting operations supply the first answer. Track candidates who disclose a restriction, offers delayed for review, roles abandoned, start dates moved, and people excluded before interview. Compare those costs with the specific losses the company’s own clauses are meant to prevent.
A talent team need not infer the business interest from a legal template. The hiring manager requesting a restraint names it. If the answer is a current client list, a narrow confidentiality and non-solicitation provision may address it. If the answer is general fear that a trained worker might leave, the business has identified a retention problem, not a secret.
The NDA already carried the secret
Employers have a serious argument. Training costs money. Customer relationships take years to build. Product road maps, pricing, source code, security controls, and acquisition plans lose value when disclosed. Failure to protect them may reduce investment or keep workers away from important knowledge.
Noncompetes are sometimes defended as the condition that makes sharing possible. Give the employee broader access, pay for training, and allow them to work on sensitive projects because they cannot immediately carry the result to a competitor. The alternative, in this account, is a company that withholds opportunity.
By the time an NDA case begins, an unreleased price or model design may already be with a rival. Damages are hard to prove, and the competitive moment may be gone before a court acts. Access controls do not erase what a senior employee remembers. A customer relationship cannot be returned like a laptop. Prevention may look safer than suing after an irreversible disclosure.
A targeted restraint has its strongest case here. Test it by identifying the investment or information that would not be shared without the clause, the period during which it retains competitive value, the roles capable of using it, and the narrower tools that fail.
One result cuts against automatic escalation. Every contract in the finance experiment had an NDA. Removing the noncompete produced no evidence of secret leakage, and the researchers report that their design could reject even small effects. Within this setting, the extra restriction changed mobility and earnings without adding observable protection against leakage.
Scope again matters. Recruiters on short projects may handle different secrets from laboratory scientists or deal executives. Leakage measurement might miss harms that emerge later or outside the observed firms. No single experiment proves that an NDA is always sufficient.
The result breaks the habit of treating sufficiency as untestable. Before adding a noncompete, inventory the protective tools already in place: role-based access, data-loss prevention, confidentiality clauses, invention assignment, customer non-solicitation, return-of-property procedures, paid notice, and targeted post-employment reminders. Map each tool to a named risk.
Access design is often more reliable than a promise about future employment. A recruiter who never needs to export an entire candidate database does not receive that capability. A machine-learning engineer works in a controlled environment where model weights, production credentials, and unreleased evaluation data remain logged. A salesperson loses CRM export access before departure. These controls do not eliminate misconduct, but they address the asset directly.
Memory and relationships still require judgment. If a role gives one person current merger terms, a live bid, or authority over a small set of strategic accounts, technical controls alone may be inadequate. The review explains why residual risk survives those controls and why the proposed duration matches the information’s useful life. A generic reference to competition leaves the gap open.
Time also narrows the problem. Pricing that changes weekly needs a different protection period from a long-lived trade secret. A launch plan loses sensitivity after release. A customer relationship may persist, yet an industry-wide restriction reaches jobs that never touch the customer.
Training needs its own file. The OECD chapter reviews studies linking stronger noncompete use with greater training intensity in some occupations. It also says evidence that workers receive higher long-term earnings from that training is limited. In the OECD-Bocconi employer survey, noncompete use did not differ significantly by past innovation, growth, or training activity. More than 40% of firms reporting innovation or widespread training used no noncompetes at all.
Those mixed findings make a company-specific record essential. Name the training that goes beyond normal onboarding, its cost, the portable credential or capability created, and what the worker receives. A paid retention agreement, milestone bonus, or narrowly designed repayment term for optional training may express a valuable investment more directly. Local law determines what is allowed.
A broad restriction also weakens the protection story. If a clause applies to workers with no access to confidential information, the template appears to follow headcount rather than risk. The OECD found that 31% of firms using noncompetes applied them to all employees regardless of role or seniority.
Good information security starts with classification. Contract design needs the same discipline. Protect source code as source code, client commitments as client commitments, and a worker’s general skill as the worker’s skill.
Workers cannot price a clause they miss
Timing changes the economics of an offer. A candidate compares opportunities while several doors remain open. Once the person accepts, resigns, rejects other offers, or schedules a start date, those alternatives begin to close. A restriction disclosed at the signature stage arrives after part of the bargaining market has disappeared.
OECD survey results put numbers around that sequence. Among workers bound by a noncompete, 21% became aware of it only when signing. Another 18% learned about it after accepting the job but before signing. Only 44% said they read the clause carefully; 23% skimmed it, and 9% signed without reading it. Among those who read it, 47% said they did not fully understand it.
Only 25% tried to negotiate. Some saw the clause as reasonable. Others believed it was non-negotiable, were explicitly told negotiation was unavailable, worried about creating tension, or lacked another offer. Those reasons describe bargaining conditions, not mere reading habits.
Late notice also defeats price discovery. A worker cannot compare a restricted $120,000 offer with an unrestricted $115,000 offer if the restriction appears after the second offer expires. A company cannot claim the wage includes compensation for the clause unless the candidate could see and value both parts of the exchange.
Almost half of employees in the OECD survey reported receiving no compensation for their noncompete. Legal requirements vary sharply. Some countries require a percentage of pay during the restricted period. Common-law systems may treat acceptance of employment as sufficient consideration. Several jurisdictions ban clauses for lower-paid workers or limit duration and scope.
A contract copied unchanged from London to Toronto to Mexico City carries different rules into the same sentence. One operating principle still travels: disclose a proposed restraint with the first written offer, in plain language, before the candidate surrenders alternatives. Legal validity remains local.
Plain language answers practical questions. Which employers or activities count as competition? Which geography applies to remote work? When does the clock start? What happens after a layoff? Who issues a release? Is the worker paid during the restricted period? Which confidentiality duties continue after release?
Give the candidate a complete copy and time to seek advice. Route requested changes to a named owner rather than making the recruiter improvise. Recruiters explain process and timing; they are not the legal interpreter.
Managers need a boundary too. A manager who requests a clause must not punish a candidate for asking what it means. Record the request, the business rationale, and the decision. Aggregate the data by role, level, location, and manager to find one department using restrictions far more broadly than peers.
Salience deserves separate attention. Bold text or a summary box improves awareness but does not cure an overbroad term. The field experiment randomized salience because visibility may change both acceptance and behavior. A well-disclosed restriction may still be unnecessary; a narrow restriction may still arrive too late.
Post-employment communication is part of the system. The working paper found that many workers appeared unaware until the firm contacted them after the project. If the first concrete explanation arrives when a rival job is in hand, the company has shifted interpretation to the moment of maximum leverage.
Employer-initiated departures need an explicit branch. A worker who is laid off may lose income and then discover that the old employer still claims a right over the next job. Some jurisdictions limit or invalidate restraints after certain dismissals; others do not. The offer record states whether an employer-initiated separation triggers automatic release, paid protection, or a fresh review. The exit notice repeats the answer.
A better exit process states what remains protected, what work is restricted if any, how to request a written release, and when the company will answer. If a product has launched or a client assignment ended, release does not depend on a worker finding counsel.
Fifteen countries show the same contract friction
Causal leverage comes from the narrow field experiment. Breadth comes from the OECD, whose 2026 chapter combines employer surveys across 15 countries with employee surveys covering 31,762 workers in 14 of them and 6,112 firms across the full employer sample.
Employers reported that around one-fifth of private-sector employees were definitely covered by noncompetes, rising to roughly one-third when probable coverage was included. The exact definite and probable estimates vary by presentation and respondent. The useful signal is scale plus uncertainty: many workers and firms do not have a crisp shared record of whether a restraint exists.
Use has spread beyond senior employees. The OECD found noncompetes in knowledge-intensive services such as information, finance, real estate, and professional services, but also in retail, transport, hotels, and restaurants. It documented clauses among lower-educated and lower-paid workers and among people without access to confidential information.
Standardization explains part of that reach. Nearly one-third of clause-using firms said they applied noncompetes to everyone. A template reduces drafting cost, but it also removes the fact pattern that is supposed to justify the restraint.
Small firms face a real constraint here. They may lack employment counsel in every market and may rely on a contract supplied years earlier. The OECD notes that small firms using clauses often cover a larger share of workers, while many employers misunderstand compensation requirements. A lightweight review process therefore matters as much for a 40-person startup as for a multinational.
Worker representation changes the process in some countries. The OECD found that sectoral bargaining adds limits in places including France and Sweden, while collective agreements play little role elsewhere. A global company cannot treat consultation, compensation, and individual consent as interchangeable. Where employee representatives have a formal role, review starts before the template reaches a candidate.
Formal enforceability does not settle behavior. The OECD found that workers may avoid a conflicting offer even when a clause appears unlikely to be upheld. The expected cost of finding out can be enough. That chilling effect means a legally weak clause can remain economically strong.
Firm behavior travels outward. Workers moving between companies carry tacit knowledge: how to deploy a tool, organize a team, diagnose a customer problem, or recognize a failed approach. Hiring lets more productive firms expand. Startups form when people leave incumbents with enough experience to build something new without taking protected material.
An OECD working paper published July 17 merged cross-country noncompete prevalence with firm data from Orbis. A 10-percentage-point increase in industry-level prevalence was associated with 1.9% lower aggregate business-sector productivity. The authors link the association to weaker labor reallocation and slower knowledge diffusion.
They explicitly do not claim a purely causal relationship. Industries that use more clauses may differ in other ways. Orbis coverage and survey estimates carry limitations. Do not convert the 1.9% association into the forecast loss from one company contract.
It remains a useful macro check. A hiring policy may look rational from one firm’s side and costly when every firm copies it. The company retains an option over its workers while losing access to experienced people restrained elsewhere. Local protection becomes market congestion.
Heterogeneity also appears in a March 2026 Chicago Fed working paper. Using NLSY97 work histories, the authors found noncompetes associated with slower wage growth over four years for lower-education workers but faster wage growth for higher-education workers. Job-tenure estimates were imprecise.
Different population, contract, and method mean the observational result does not cancel the randomized recruiter result. It raises a distribution question: who receives training, bargaining power, compensation, or access in exchange for the clause, and who receives only the restriction?
An AI company may have both groups on the same payroll. A highly paid researcher may negotiate equity, scope, and a paid restriction. A data annotator, sales development representative, or support contractor may see the same template without the same leverage. Company-wide incidence hides that difference.
Measure by cohort. Track role, access to protected information, compensation for the restraint, negotiation, release, enforcement contact, rejected applicants, time to fill, pay growth, and destinations. The aim is not to manufacture a favorable average. It is to see whether the clause is attached to a legitimate interest or to a worker with weak bargaining power.
Federal policy moved from a ban to cases
U.S. employers cannot use the phrase “the FTC rule” as a complete 2026 answer. The nationwide Noncompete Rule issued in 2024 was vacated in federal court. In September 2025, the FTC dismissed its appeals and accepted the vacatur. In February 2026, the agency removed the rule from the Code of Federal Regulations to conform to the court decisions.
No nationwide FTC ban is currently in force. That does not create nationwide permission.
State statutes and case law still differ by worker, wage, profession, notice, consideration, duration, geography, protected interest, and circumstances of departure. Federal antitrust enforcement also continues through individual matters under Section 5 of the FTC Act.
Rollins shows the new posture. On June 22, the FTC finalized a consent order requiring the pest-control company to stop enforcing noncompetes against more than 18,000 employees nationwide and to notify current and former workers that they could compete, including by starting a business.
According to the agency’s complaint, the clauses covered nearly all employees, typically lasted two years, and often reached a 75-mile radius around more than 700 locations. Covered roles included technicians and customer-service staff. The FTC alleged that workers had little ability to negotiate, received no extra compensation, and sometimes faced cease-and-desist letters or lawsuits.
Those are matter-specific allegations and order terms. They do not turn every two-year clause or every 75-mile radius into an automatic federal violation. The final order has force for Rollins. Other companies must assess their own facts and applicable law.
Policy moved in a sequence that a ban/no-ban headline misses. A broad national rule failed in court. The agency then used complaints, warning letters, and consent orders against arrangements it viewed as unfair or anticompetitive. Enforcement attention moved from the existence of any clause to the workers covered, bargaining process, compensation, scope, and competitive effect.
For a multi-state employer, this creates two review layers. Legal counsel maps what a jurisdiction allows. Business owners decide whether an allowed restriction is necessary, targeted, and worth its hiring cost. Passing the first layer does not answer the second.
Reviews expire. Laws change, salary thresholds move, and remote workers complicate the location analysis. A clause drafted for an office in one state may reach a person now working elsewhere. Acquisition templates can leave different employee groups under different terms.
Do not ask a recruiting system to decide enforceability from a location field. The system routes offers, preserves the approved version, enforces notice periods, and records acknowledgments. Local legal analysis still determines whether a court would uphold a fact-specific restraint.
Past clauses need an owner as well. When policy or law changes, identify who receives notice, which clauses are released, which confidentiality duties remain, and how the company answers a reference or competitor inquiry. A silent policy update leaves the chilling effect intact for workers holding older paper.
Operational analysis cannot replace legal advice. A company considering, changing, releasing, or enforcing a noncompete needs qualified counsel in each relevant jurisdiction. The operating work makes that advice better by supplying the role, asset, access, timing, compensation, and outcome instead of sending counsel an empty template.
An offer review before the signature
Return to the next recruiter offer. The manager wants the freelancer to see candidate data and the finance firm’s hiring plan. Legal has a standard noncompete. Talent Operations has a start date two days away. The candidate has another project waiting.
Pausing the offer does not require a committee or a philosophical debate. One compact review, owned by people who answer different parts of the decision, is enough.
| Review field | Evidence to record | Decision it supports |
|---|---|---|
| Protected asset | Named dataset, client relationship, pricing, code, plan, or optional training | Whether a legitimate business risk exists |
| Worker access | Systems, records, customers, projects, and time sensitivity | Whether this role can create the stated harm |
| Narrower protection | NDA, access control, invention assignment, non-solicit, paid notice, return of property | Whether mobility must be restricted at all |
| Restricted activity | Specific work, competitors, customers, geography, and duration | Whether scope follows the risk rather than the industry |
| Notice and understanding | Date shown, plain-language summary, review period, questions, and counsel opportunity | Whether the worker saw the term while alternatives remained |
| Compensation | Incremental pay, paid restriction, equity, or other consideration | Whether the lost option has an explicit price |
| Approval and release | Business owner, legal reviewer, expiry, release contact, and response time | Who can narrow or end the restriction |
| Outcome | Acceptance, negotiation, lost candidate, later move, enforcement contact, and verified loss | Whether the clause works as claimed and what it costs |
Use the table as a management frame, not contract language. Local counsel determines which fields are legally required and which tools are available. The business supplies the evidence.
Start with the asset. “Confidential information” is too broad for an operating record. Name the candidate database, unreleased workforce plan, client pricing, model weights, deployment playbook, or acquisition file. State when its value decays.
Then map access. The recruiter may need candidate names and interview notes but not executive compensation data. A software engineer may work on an internal evaluation harness without access to customer contracts. Role design shrinks the protected surface before a contract restricts a person.
Test alternatives in order. An NDA addresses disclosure. Access controls reduce possession. A customer non-solicit protects relationships without blocking unrelated work. Paid notice keeps a worker away from current strategy while continuing income. None is automatically harmless or legal everywhere. Compare each with the named risk.
If a noncompete remains, make the scope readable. “Competitor” cannot silently expand to every firm in a growing category. Remote work makes a radius especially blunt. Duration follows the life of the information, not a round number inherited from another role.
Show the term early. Put the full provision and a plain-language summary with the first written offer. Give a defined review window. Keep another candidate’s deadline from becoming the hidden reason this person cannot ask a question.
Price the restriction separately where appropriate and lawful. A line called base salary cannot credibly pay for every obligation that appears later. When compensation is required during a restricted period, the payroll and termination process must know how to trigger it.
Create a release path before an exit. The worker needs an address for a competing role description and a response deadline. Silence may destroy an offer even if a release arrives weeks later. Record the answer and any conditions.
Enforcement requires a loss statement. Before a cease-and-desist letter, the business owner identifies the protected asset, the evidence of threatened use, and why a narrower response fails. Legal reviews that record. A desire to deter other departures is not evidence that this worker took a secret.
Finally, measure outcomes twice: for workers the company restrains and candidates it cannot hire because of another firm’s clause. Include acceptance, negotiation, abandoned offers, delayed starts, release rates, legal spend, verified leakage, customer loss, pay growth, and mobility. A clause that creates many contacts but no verified loss may be generating activity rather than protection.
Incoming candidates need a parallel path. Record the rival clause, role overlap, response deadline, and whether the company narrowed duties, delayed the start, sought a release, or abandoned the hire. Do not ask the candidate to guarantee enforceability. Company counsel owns that risk decision, and the hiring manager sees the time and capacity lost while it is made.
For the finance-firm recruiter, the review may end with the NDA already in every experimental contract. Access controls keep bulk candidate data inside the system. A client non-solicit covers relationships the recruiter personally handled. The broad competing-firm restriction drops out.
Another role may produce a different result. A senior executive receives a paid, time-limited restraint tied to a live transaction. The file records who approved it and when it expires. A support worker with no confidential access receives no clause at all.
Different roles produce different outcomes. Variation shows that the company is matching protection to facts instead of attaching a standard loss of mobility to every badge.
Two finance firms gained a rare view of the counterfactual through the 2026 experiment. Most employers will not randomize contract restrictions. Cohort comparisons, exception reviews, and expired-risk checks still reveal clauses that no longer map to a protected asset.
Before the recruiter signs, the offer now has three visible lines: what information belongs to the firm, what work remains open to the recruiter, and who will answer if the boundary is disputed.
The firm keeps its secret. The worker keeps a market.