Clients are increasingly comfortable with professionals using AI. What they are far less comfortable with is removing the professional from the process altogether.
Businesses are becoming comfortable with AI surprisingly quickly. They use it to draft copy, analyze information, generate code, create designs, summarize documents and automate parts of everyday work.
That does not mean they are equally comfortable handing the final result to AI.
In an October 2024 survey of Upwork clients, 66% expressed high levels of trust in work delivered by freelancers using AI tools, while only 26% expressed the same level of trust in work delivered by AI alone, according to the research on AI’s changing role in professional work.
That is a 40 percentage point gap. Roughly two and a half times as many clients reported high trust when a freelancer remained involved.
The obvious interpretation is that clients trust humans and distrust AI. The evidence points to something more interesting. Clients appear increasingly comfortable with AI being part of professional work. What they are less comfortable with is AI becoming the only party responsible for the result.
That distinction could shape how professional services are bought and sold as AI becomes more capable.
Clients are not rejecting AI
The trusted option in the survey was not human-only work. It was work delivered by a freelancer who was using AI.
That matters because it changes the argument. Clients do not appear to be demanding that professionals avoid the technology. They appear to want someone competent using it, checking it and taking responsibility for what gets delivered.
The same tension appears in broader research. A 2025 global study of more than 48,000 people across 47 countries, led by the University of Melbourne in collaboration with KPMG, found that 66% of respondents used AI regularly, while only 46% were willing to trust AI systems.
Adoption and trust are not moving at the same speed.
People can find AI useful without wanting to depend on it completely. Someone may use ChatGPT every day and still want a lawyer to review an important contract. A developer may rely heavily on an AI coding assistant while still expecting an experienced engineer to decide whether the code is safe to deploy. A client may encourage a designer to use generative tools while expecting that designer to make the final creative decisions.
The professional does not necessarily need to perform every step manually. The value may increasingly lie in making sure the result is dependable.
Human involvement changes what the client is buying
When a business hires a freelancer, it is easy to think the transaction is mainly about production. The client wants an article, a website, a presentation, an analysis or a design.
Professional services have always involved more than producing the final artifact.
Clients also pay someone to understand an incomplete brief, notice when information is wrong, make judgment calls, adapt to context, explain decisions and solve problems when the first version does not work.
AI can reduce the amount of manual production involved without eliminating those responsibilities.
A client building a financial model may use AI to generate formulas or summarize financial documents. Someone still needs to know whether the assumptions make sense. A company commissioning market research can use AI to organize large amounts of information, but it still needs someone who can distinguish a credible source from a weak one and recognize when the evidence does not support the conclusion.
The same problem is increasingly visible in software. A developer can generate substantial amounts of code with AI, but someone still needs to understand how that code fits the wider system and whether it introduces problems that will appear later.
This is already creating paid work. Jobs asking professionals to repair or refine AI-generated output grew 70% year over year, a shift Flexable examined in its report on the growing market for fixing AI-generated work.
The two findings fit together. AI makes it easier to generate work, but generation does not always finish the job. The 66% trust figure suggests clients recognize the difference.
Human oversight can increase trust, but context matters
There is experimental evidence that knowing a person remains involved can change how people respond to AI-assisted work.
A study involving 1,542 participants and AI-assisted care planning found that when people were explicitly told humans remained in the decision loop, the share who trusted the AI-assisted care plan more than a non-AI plan increased by 8.95 percentage points compared with the condition where that assurance was absent.
More recent research points in the same direction, while also showing that the effect depends heavily on the task. A 2025 preregistered experiment involving 2,100 participants examined trust in human involvement and AI-supported decisions across medical diagnosis, hiring, transportation and financial investment. Decisions involving human intervention were generally perceived as more trustworthy, although financial investment provided an exception.
There is no universal rule saying people will always prefer a human.
Trust depends on what the work involves.
For highly structured tasks where accuracy can be measured clearly, people may be comfortable relying heavily on an algorithm. When the work involves taste, ambiguity, interpretation, personal consequences or subjective judgment, human involvement often carries more weight.
That difference becomes even clearer when researchers look at what happens after people are told that AI contributed to the work.
The trust penalty is stronger when the work feels personal
A 2026 meta-analysis covering 13,582 participants examined 21 experimental effect sizes to understand how AI disclosure affected trust, attitudes and behavioral intentions.
Across the studies, revealing AI involvement produced a significant negative effect. The size of that penalty, however, depended on the setting.
It was considerably stronger in subjective contexts such as advertising, services and chatbot communication than in more objective informational settings.
That distinction is particularly relevant to freelance work because a large share of professional services cannot be judged with a simple right-or-wrong answer. Brand positioning, design, copywriting, consulting recommendations and research interpretation all depend on context and judgment.
A client may be completely comfortable with AI contributing to the process while still wanting a person to decide whether the final work is appropriate.
Research on content creation points in a similar direction. Two experiments examining consumer reactions to human and AI collaboration found that negative reactions could be reduced when people understood that humans retained control over AI-generated content.
The important distinction is not whether a human name appears somewhere in the workflow. Clients need a reason to believe that meaningful judgment was actually exercised.
Trust should not be confused with better performance
The 66% figure has an obvious trap.
Because clients report more trust in human and AI collaboration than in AI alone, it would be easy to conclude that human and AI teams consistently produce better work.
The wider evidence does not support that claim.
A major systematic review and meta-analysis published in Nature Human Behaviour examined 106 experiments and 370 effect sizes comparing humans alone, AI alone and humans working with AI.
Human and AI combinations performed better than humans working alone on average. They did not, however, outperform whichever of the human or AI system was already the strongest option. On average, the combined system actually performed worse than the better standalone performer.
Task type mattered considerably. Human and AI collaboration showed more promise in creation tasks, while many decision tasks produced weaker results relative to the best standalone option.
That nuance matters for clients because human involvement is not automatically valuable. An inexperienced person approving an answer they do not understand does not make that answer more reliable.
The value comes from meaningful oversight.
A professional needs enough expertise to recognize poor output, enough context to judge whether the result fits the situation and enough independence to reject what the AI recommends when necessary.
Otherwise, “human in the loop” becomes little more than a label.
Knowing when not to use AI may become a valuable skill
Much of the conversation around AI skills focuses on how effectively someone can use the technology. Businesses want people who can prompt well, automate workflows, use agents and produce more in less time.
Those abilities matter, but the trust research points to another skill that may become equally important: knowing when the tool should not be trusted.
A strong freelancer may accept an AI output quickly in one situation, review it carefully in another and decide not to use AI at all for a third.
That judgment is difficult to express through an AI proficiency badge. It is also difficult for clients to evaluate before hiring someone.
This is where professional reputation becomes more important. The client is not simply choosing between a human and a machine. The client is choosing whether to trust a particular professional to use the machine appropriately.
For readers new to independent work, Flexable’s guide to what freelancing is and how the model works explains how responsibility sits directly between the professional and the client.
That relationship makes trust commercially important.
A freelancer who consistently delivers reliable work can earn repeat projects, referrals and longer client relationships. Someone who sends plausible AI output without checking it properly may save time on one assignment while damaging the relationship that could have produced many more.
Flexable’s research into inconsistent work among freelancers shows why continuity matters. For independent professionals, client trust can have a direct effect on how predictable future work becomes.
AI disclosure is becoming part of the client relationship
Clients cannot evaluate human oversight if they have no idea how AI contributed to the work.
That does not mean every minor interaction with AI requires a disclosure statement. Using a tool to correct grammar is different from using it to generate the central recommendation in a consulting report. Asking a coding assistant to explain an error is different from delivering a largely generated application that received very little review.
The useful question is not simply whether AI was used. It is how much the final work depended on it and what the professional did after receiving the output.
This becomes more important because transparency around workplace AI use remains weak. Flexable’s analysis found widespread reluctance among employees to reveal their AI use, even as the technology becomes more common in everyday work.
For clients, useful disclosure can be much more practical than a generic statement saying AI was involved. A freelancer might explain that AI was used for first-pass research, code generation or ideation, while the sources were independently verified, the code was tested or the final creative decisions remained human.
That tells the client something meaningful about the process.
Saying “I used AI” reveals very little. Explaining what AI did and what the professional personally checked provides information the client can actually evaluate.
Clients may care more about accountability than authorship
The most useful interpretation of the 40-point trust gap may have less to do with who physically created the work and more to do with who stands behind it.
Imagine two deliverables of equal quality. One was produced entirely by a professional. The other was completed much faster because the professional used AI extensively, reviewed the output, corrected the problems and took responsibility for the final result.
The second process does not necessarily reduce the value to the client. It may increase it if the work arrives faster, costs less or gives the professional more time to focus on the difficult parts.
Problems arise when responsibility becomes unclear.
If an AI-generated recommendation turns out to be wrong, the client still needs someone who can explain why the recommendation was made. If generated code creates a security vulnerability, the business needs a developer capable of understanding and correcting it. If an AI-assisted campaign damages a brand, saying that a model generated the material does not answer the question of who approved it.
Professional services have always contained an element of accountability. AI is making that part of the transaction more visible.
The output may increasingly be produced through a combination of human and machine, but somebody still needs to own the result.
The trust gap may change what freelancers sell
As AI improves, freelancers may need to rethink how they communicate their value.
Saying “I can produce this for you” becomes less distinctive when a client knows software can generate a plausible first version in seconds.
Selling judgment is different.
A developer can demonstrate how they test AI-generated code before it reaches production. A designer can show how they turn generated concepts into a coherent visual system. A researcher can explain how sources are verified. A writer can compete on reporting, subject knowledge and editorial judgment rather than word count. A consultant can use AI to accelerate analysis while making clear that the final recommendation remains their responsibility.
In each case, AI is not hidden from the client and it is not the product being sold. It is part of the professional process.
That fits a wider change already taking place across independent work. Businesses increasingly use freelancers for specialized expertise rather than simply as additional labor, a trend visible in Flexable’s analysis of the global rise of freelancing.
If trust continues to favor professionals who use AI over AI operating alone, expertise may become more valuable precisely because basic production becomes easier.
What the 66% and 26% figures actually tell us
The headline finding comes from a client survey conducted in October 2024.
It measures reported trust, not objective quality, and the sample should not be treated as representative of every buyer of professional services worldwide.
The result therefore does not show that human and AI collaboration always produces better work than AI acting alone. Experimental research makes clear that performance depends on the task, the quality of the AI system, the expertise of the person involved and how responsibility is divided between them.
What the survey does show is narrower and more useful.
Clients in the sample were far more comfortable trusting work when a professional remained involved.
Independent research on AI trust, disclosure, oversight and performance helps explain why. People are increasingly willing to use AI and accept AI-assisted work while remaining cautious about situations where the technology operates without meaningful human control.
For freelancers, agencies and businesses buying professional services, that points to a practical shift in where professional value may sit.
The advantage is unlikely to come from pretending AI is not part of the work.
It may come from making the human contribution clear enough that clients know exactly what they are trusting.