Artificial intelligence does not need to change the facts to change how persuasive a recommendation feels.
New research suggests that simply framing the same basic idea in language that fits a person’s existing values can make an AI chatbot more convincing, increase support for its recommendation and even raise how much users say they would be willing to pay for the service.
The study, published in Scientific Reports on 18 September 2026, examined how large language models influence workplace-style decisions when their recommendations are tailored to the moral values associated with a user’s political identity.
Across one exploratory study and two pre-registered experiments, the researchers found that people were more likely to endorse an idea when the AI presented it using values that aligned with their own worldview. The same value-matched framing also increased commercial engagement, including willingness to pay.
The effect was strongest among participants with firmer political views.
The findings do not show that an AI system can simply rewrite a person’s beliefs at will. They do show something more subtle and commercially important: a recommendation can become more persuasive when the system makes users feel that the argument fits the way they already think about right and wrong.
The AI did not need a different recommendation
The researchers were interested in a question that is becoming increasingly important as generative AI enters offices, consulting work, strategy teams and everyday decision-making.
If two people receive broadly the same recommendation from an AI system, will they respond differently depending on how that recommendation is framed?
To test this, Giles Hirst, Wayne Johnson, April J. Li and Andreas W. Richter drew on moral foundations theory, which examines how different moral concerns shape people’s judgments and decisions.
The central manipulation was not whether the AI recommended one option or another. Instead, the researchers changed the moral language used to justify the recommendation.
For some users, the AI framed an idea using values that were more consistent with their political identity. For others, the framing was less aligned.
When the framing matched the user more closely, the recommendation became more effective.
People were more likely to endorse the idea
The first clear result was greater idea endorsement.
Participants were more willing to support the AI’s recommendation when its reasoning was framed in values that felt more congruent with their own.
This matters because many workplace uses of AI involve exactly this kind of judgment.
An employee might ask an AI system to evaluate a proposal, recommend a strategy, suggest a policy or help decide between competing ideas. The answer may contain the same underlying evidence but still feel more or less convincing depending on the language used to present it.
The study suggests that persuasion can therefore enter AI-assisted decision-making through framing rather than through factual differences.
Value matching also increased willingness to pay
The effect did not stop at agreement.
Participants exposed to value-congruent AI recommendations also reported greater willingness to pay for the service.
That result gives the study a direct commercial dimension.
If an AI system makes users feel more understood, they may not only rate its recommendations more highly but also place greater economic value on access to it.
For businesses building AI products, this creates an obvious incentive to personalise how systems communicate.
But it also creates an important governance question.
At what point does useful personalisation become persuasion designed to exploit a user’s existing values?
Two mechanisms appeared to drive the effect
The researchers identified two separate pathways that helped explain why value-matched framing worked.
The first was what they describe as issue selling.
When the AI used value-congruent language, participants found the recommendation itself more compelling. The framing made the argument easier to accept.
The second mechanism was a stronger sense of being understood.
Users were more commercially engaged when the AI’s language made them feel that the system understood them personally.
These two processes are related but not identical.
One changes how convincing the recommendation feels. The other changes the relationship the user feels they have with the system.
Together, they show why personalised AI can become more influential even when the underlying recommendation remains similar.
Feeling understood can become a commercial asset
This is one of the most important findings for companies building customer-facing or employee-facing AI systems.
Traditional software is usually valued for what it does.
Generative AI can also be valued for how the interaction feels.
A chatbot that uses language, examples and reasoning that resonate with a user may appear more intelligent, more useful and more personally relevant than one that gives the same recommendation in neutral language.
That feeling of fit can become part of the product itself.
This may partly explain why AI companies are investing heavily in personalisation, memory and context-aware assistants.
The more a system learns about the user, the easier it becomes to present information in a way that feels naturally aligned with that person’s preferences.
The effect was stronger for people with firmer political views
The study also found that value-congruent framing was particularly influential among users whose political views were more firmly held.
That result is important because strong prior beliefs are often assumed to make people harder to persuade.
The new findings suggest that strong beliefs may instead create a clearer pathway for persuasion when an argument is framed in language that fits those beliefs.
A person with a well-defined worldview gives the system more predictable values to speak to.
That does not mean politically committed people are uniquely vulnerable to AI. It means the persuasive value of personalisation may increase when users have more clearly structured preferences.
This is not simply a political story
The experiments used political identity as a practical way to identify different value orientations, but the broader mechanism can extend well beyond politics.
People differ in what they value when making decisions about money, careers, products, health, sustainability and risk.
An AI financial assistant could emphasise security for one user and opportunity for another.
A workplace assistant could frame a proposal around efficiency, fairness, innovation or stability depending on the employee.
A shopping assistant could emphasise sustainability to one customer and durability or value for money to another.
The basic principle is the same: the recommendation may become more persuasive when the reasoning is translated into the user’s preferred value system.
AI personalisation could improve communication
There is an entirely legitimate side to this.
Communication works better when information is presented in a way that audiences understand.
A manager explaining a change to employees may naturally emphasise different aspects depending on what matters to different teams.
A doctor may explain the same treatment differently to patients with different levels of medical knowledge.
An AI system capable of adapting its explanations could make complex information easier to understand and more relevant to the person receiving it.
That is one reason personalised AI has significant potential.
The problem appears when adaptation moves from improving comprehension to strategically increasing compliance.
The line between assistance and influence may become harder to see
Generative AI is unusual because the same system can advise, explain, persuade and sell within one conversation.
A user might begin by asking for information and end by receiving a recommendation tailored to their psychological profile.
Unlike a traditional advertisement, the persuasive component may not always be obvious.
There may be no slogan, banner or sales pitch.
The influence can emerge through the wording of an apparently helpful answer.
This creates a difficult question for platform designers and regulators: when should users be told that an AI recommendation has been personalised specifically to make it more persuasive?
The workplace may be especially important
The authors specifically highlight implications for workplace decision-making.
Employees are increasingly using generative AI for brainstorming, writing, analysis, planning and strategic recommendations.
In many organisations, an AI-generated suggestion can become part of the evidence used to justify a decision.
If the system learns how an employee or manager prefers to think, its influence may increase even if the quality of the recommendation does not.
This could become particularly important when multiple employees are evaluating the same proposal.
Two people could receive differently framed versions of effectively the same recommendation and come away with different levels of confidence in it.
That makes transparency around AI personalisation more than a technical issue.
It becomes part of organisational governance.
South African businesses are likely to face the same question
South African banks, insurers, retailers, telecommunications companies and professional-services firms are already integrating generative AI into employee and customer interactions.
As these systems become more personalised, companies will need to decide what kinds of adaptation are acceptable.
There is a meaningful difference between remembering that a customer prefers simple explanations and learning which value-based arguments are most likely to make that customer accept a product recommendation.
The first improves usability.
The second begins to look more like behavioural targeting.
That distinction will become increasingly important as AI systems gain access to richer customer histories, previous conversations and behavioural data.
Personalised persuasion could become difficult to audit
Traditional marketing campaigns are relatively visible.
A company can review the advertisement, email or promotional message that customers received.
Generative AI can create a different explanation for every user.
That makes oversight more complicated.
A company may approve the general behaviour of an AI assistant without seeing every persuasive message it generates.
If value-based framing improves conversion, systems may also gradually be optimised toward increasingly effective forms of persuasion.
That creates a need for monitoring not only what AI recommends but also how recommendations are framed.
The research does not show that AI controls users
The findings should not be overstated.
The study shows measurable shifts in endorsement and willingness to pay under controlled experimental conditions.
It does not show that value-congruent AI can force users to accept recommendations they strongly oppose.
Nor does it prove that the same effect will be equally strong across every culture, industry or type of decision.
The experiments focused on specific forms of value framing and used political identity as a key source of value congruence.
Real-world AI interactions are considerably more complex.
Users may also become less responsive to personalisation if they recognise that a system is deliberately mirroring their values.
There is also a difference between persuasion and manipulation
Persuasion is part of ordinary communication.
People choose examples, arguments and language that they think will resonate with an audience.
Manipulation usually implies that influence is hidden, deceptive or designed to bypass meaningful choice.
Personalised AI can sit uncomfortably between those categories.
A system may present completely accurate information while selectively framing it in the way most likely to move a particular user.
Nothing in the recommendation needs to be false for the interaction to become highly persuasive.
That is what makes the issue more difficult than ordinary misinformation.
The commercial incentive is obvious
For technology companies, the study identifies a potentially powerful advantage.
If users are more willing to pay for AI that feels aligned with them, personalisation can directly support revenue.
Companies therefore have a financial reason to make AI assistants feel more understanding and personally relevant.
In many cases that may improve the product.
But commercial optimisation can also push systems toward the kinds of framing that maximise engagement rather than the kinds that produce the most balanced decisions.
The same system that understands a customer better can potentially influence that customer more effectively.
The bigger question is what users know about the process
The study ultimately points toward a transparency problem.
Most users understand that recommendation engines use personal data.
It is less obvious that a conversational AI could alter the moral framing of an argument based on what it knows about the person receiving it.
As AI assistants become more persistent and personalised, the distinction may matter more.
A system that remembers a user’s preferences, previous decisions and strongly held values could become unusually effective at presenting recommendations that feel intuitively correct.
That may be useful when the user explicitly wants tailored advice.
It is more concerning when the system is also trying to sell something, influence a workplace decision or steer behaviour toward a commercial objective.
AI may become persuasive because it learns how to speak our language
The most striking implication of the research is that AI influence does not necessarily require better arguments.
It may require better alignment.
The recommendation becomes more compelling because the system presents it through values the user already recognises as important.
That is a powerful capability for education, coaching and communication.
It is also a capability that deserves careful oversight.
The more AI systems know about their users, the easier it becomes to move from answering a question to shaping the way the answer is received.
Source Information
Study Title: Workers shift their views and pay more when AI chatbots pander to their values
Authors: Giles Hirst, Wayne Johnson, April J. Li and Andreas W. Richter
Journal: Scientific Reports
Published: 18 September 2026
Method: One exploratory study and two pre-registered experiments grounded in moral foundations theory
Main finding: AI recommendations framed to align with users’ values increased idea endorsement and willingness to pay, with stronger effects among users with firmer political views
DOI: 10.1038/s41598-026-71409-1
