Disclaimer: This article provides general information and editorial analysis for research buyers and practitioners. It does not provide legal advice or endorse particular verification products. Examples are illustrative. Researchers should assess data quality, accessibility, privacy and professional obligations for each project.

Why trustworthy market research needs participant verification, clear AI policies and human judgement

A completed questionnaire does not automatically represent a genuine consumer’s experience.

Online research can encounter duplicate submissions, invented eligibility claims, inattentive answers and automated responses. Generative AI adds another complication: fluent, plausible text can be produced without reflecting the experience a study intended to capture.

For businesses buying research, this raises an essential question: what evidence supports the claim that the findings represent the intended audience?

Why the industry is paying attention

The Market Research Society’s Campaign for Better Data brings together research organisations to improve data quality through shared experience and professional practice. Its work includes guidance on AI and synthetic data.

ESOMAR also identifies AI-enabled research fraud as a data-quality threat, including generated responses and identities presented as if they came from real participants.

This does not mean online surveys have become unusable. It means buyers should examine how evidence is collected and checked, rather than judging a study solely by its sample size or presentation.

Different problems need different responses

It helps to distinguish between several situations:

Situation	Why the distinction matters
A bot completes a survey while posing as an eligible person	The response may contain no genuine participant experience.
A person invents their eligibility	Their answers may come from someone outside the target audience.
A genuine participant gives careless answers	This is a quality problem, but it does not necessarily establish deliberate deception.
A participant uses an accessibility or language tool	Assistance may help them communicate their own experience accurately.
A participant asks AI to invent their answer	The resulting text may no longer reflect their own views or experience.
Researchers deliberately generate synthetic responses	These are model outputs and should be identified and evaluated as such.

A responsible project should establish which forms of assistance are permitted and communicate those rules clearly.

Using technology to express an answer is different from using technology to fabricate an experience.

How unreliable answers could affect business decisions

Consider an illustrative study testing demand for a new subscription service.

If responses include invented enthusiasm from people outside the intended customer group, the business could overestimate demand. If genuine customers are removed by an overly aggressive screening system, the findings could also become distorted.

Possible consequences include unsuitable pricing, misplaced advertising expenditure or product features that fail to address customers’ needs.

Larger samples do not automatically solve this problem. Collecting more answers cannot compensate for uncertainty about who supplied them and what those answers represent.

What should researchers check?

The MRS Data Analysis and AI Toolkit discusses using technology to identify suspicious response patterns, duplication, implausibly fast completion and inconsistencies. It also recognises the need to consider bias in verification systems.

Useful quality checks can include:

• Comparing eligibility answers with relevant information elsewhere in the questionnaire.
• Looking for duplicate or unusually repetitive submissions.
• Checking whether open-ended answers address the question.
• Reviewing completion patterns alongside survey length and complexity.
• Examining whether problems cluster around particular sample sources.
• Referring uncertain cases for human assessment.

These should form a documented process. A single unusual answer should not automatically establish fraud.

Polished writing is not proof of AI misuse

An articulate response may have been written by a genuine participant. A short or awkward answer may also communicate an authentic view.

As a practical editorial recommendation, researchers should treat writing style as a reason to investigate where appropriate, rather than as conclusive evidence.

Participants may use speech-to-text, translation, spelling support or other tools to communicate. If these preserve the person’s own meaning, their use should be assessed against the study’s stated rules and objectives.

Researchers need to know whether they are measuring spontaneous language, personal experiences, purchasing behaviour or something else. That distinction helps determine which assistance is acceptable.

Verification should preserve accessibility

Quality controls can create their own exclusions.

A participant who takes longer to answer may need breaks or assistive technology. Someone sharing a device or internet connection may be entirely legitimate. An inaccessible verification task could prevent an eligible participant from taking part.

Researchers should therefore offer appropriate alternatives and review questionable cases in context.

This is also a sampling issue: excluding certain groups can reduce the study’s ability to represent its intended population. ESOMAR’s online sample-quality guidance addresses participant validation and preventing repeat participation as part of wider quality assurance.

What research buyers should ask suppliers

Before commissioning an online study, buyers should ask:

1. Where will participants come from? Establish whether recruitment uses panels, open links, advertising or multiple suppliers.
2. How is eligibility checked? Ask how participants are matched to the target audience.
3. What controls address duplication and automation? Seek a clear description of the process and its limitations.
4. What is the participant AI policy? Clarify permitted assistance and how it is communicated.
5. How are exclusions reviewed? Understand whether people can be removed solely by an automated score.
6. What will the final report disclose? Request recruitment methods, exclusions, limitations and any use of synthetic responses.

These questions help make data quality a procurement requirement.

AI can support quality control

AI can assist researchers in reviewing large datasets and identifying submissions that deserve closer inspection. However, its output still requires evaluation.

MRS guidance on AI and related technologies provides a professional framework for responsible use. Research teams should be able to explain how technology contributes to their methods and how they maintain accountability.

The objective is to use automation where it adds value while ensuring that people remain responsible for interpreting evidence and making consequential decisions.

Confidence comes from transparent methods

Trustworthy research requires more than a reassuring statement that data has been “cleaned”.

Buyers need to understand the intended audience, recruitment sources, verification process, exclusions and limitations. Researchers should make those methods sufficiently clear for clients to assess what the findings can support.

As AI-generated content becomes easier to produce, the value of market research rests increasingly on demonstrating the connection between a finding and the people it claims to represent.

A polished report is useful. Reliable, traceable evidence is what makes it worth acting on.

Further reading

• MRS: Campaign for Better Data
• MRS: Data Analysis and AI Toolkit
• MRS: Guidance on AI and related technologies
• ESOMAR/GRBN: Guideline on Online Sample Quality

#MarketResearch #ResearchIntegrity #DataQuality #SurveyFraud #ArtificialIntelligence #ConsumerInsights #ResponsibleAI #MarketResearchUK