Ethics in Marketing Research

Teaching Note

Introduction

Imagine you have just finished a customer research project for a client–a really good client who pays you well. You know the answer your client was hoping to find through your research, but your data says something different. The deadline is tomorrow. Your boss, who has been in this field a lot longer than you have, tells you the finding is “probably noise” and suggests you re-cut the data by a different segment to see if it looks cleaner–more aligned with what the client was hoping to find. The new cut, it turns out, does look cleaner and does align much more with the client’s hopes and expectations.

What did you just do and was it ok?

These are not questions the law will usually answer clearly for you. Even many published professional codes might not answer these questions for you, at least not in enough detail to matter in the moment. These are ethical questions, and if you spend a career in marketing research you will face dozens of versions of them.

Ethics in marketing research are often dismissed as compliance topics. You know. The type of paragraphs you scroll past in a privacy training video, the checkbox on a survey platform about GDPR. That framing misses where most of the real decisions live. The important ethical questions in marketing research are not rare and dramatic. They are constant and small. Which question to ask. Who to ask it of. What to do when the answer doesn’t fit. How to describe a finding to a boardroom that has already decided what it believes.

The purpose of this note is to frame those questions using three lenses:

1. Obligations to the people who give you data
2. Integrity in the research act itself
3. Honesty in how findings are reported and used

Every single one of these lenses is a place where thoughtful practitioners often agree to disagree. The goal of this brief is to help you see the questions clearly enough to reflect and answer as best you can when they inevitably arrive.

When and Why

Academic researchers who collect data from human beings work inside an institutional review process (known as an IRB for “Institutional Review Board”). The IRB is imperfect, but it forces a conversation about what you are researching, who may be at risk as a result of your investigation, how it is explained to participants, and how to ensure participants understand their rights in the process.

Commercial marketing research typically has no equivalent. A researcher who builds a survey in SurveyMonkey and posts it to a paid panel does not walk through consent language with an ethics board. Professional codes do exist. For example, the Insights Association and ESOMAR both publish detailed standards. But adherence is voluntary, and the pressures that push against commercial researchers are real and not trivial. Deadlines are short. Panel economics reward speed. Clients pay for answers, and some clients prefer their answers confirm hunches.

There is also a structural asymmetry that makes marketing research particularly prone to ethical drift. You frequently come to know things about respondents that they did not knowingly share. A customer who fills out a satisfaction survey has consented to one narrow disclosure. The company that merges that survey with loyalty-card purchase history, clickstream data, and a third-party demographic data-append comes to know something much richer. The ethical question is no longer just “what did the respondent agree to tell us?” It is “what are we now in a position to know, and what are we entitled to do with that knowledge?”

Then, there are the algorithms. When findings are translated into automated decisions ethical failures that would be modest at human scale become systemic at machine scale. The topics that follow all become harder to police when the decisions happen in milliseconds, inside systems few people fully understand.

The Three Ethical Lenses

Obligations to the people who give you data

Begin where the research begins. Everything starts with a human being who has agreed to give you some of their time and some of their truth. Here’s what you should consider every time.

Consent should be informed. In practice, this means respondents understand what they are answering, how the data will be stored, whether it will be linked to other data, and whether it will be sold or shared. “By clicking this survey you agree to our privacy policy” is a legal gesture, not an ethical one. If the actual uses of the data would surprise a reasonable respondent, consent has not been obtained in any meaningful sense.

Privacy is not anonymity. Data that has been “de-identified” often isn’t. In 2006, Netflix released one hundred million film ratings as an anonymous training set for a prediction contest. Researchers cross-referenced the data with public reviews on IMDb and re-identified users within weeks. In 2012, Target’s internal statisticians built a pregnancy-prediction score from purchase patterns — unscented lotion, calcium supplements, large handbags that could double as diaper bags — and began mailing relevant coupons to women who had not told Target, or anyone else, that they were pregnant. In both cases, the ethical problem rests on an unsupervised infrastructure of research more than a case of one individual breaking a promise of confidentiality.

Vulnerable populations deserve more care, not less. Children, the elderly, and people in acute financial or emotional distress are groups for whom ordinary protections of informed consent may not be sufficient. Commercial pressure tends to push in the opposite direction, because vulnerable populations are often the most commercially interesting. While that tension isn’t always easy to resolve, it is always worth recognizing and considering.

Do not use research as a pretext. Selling Under the Guise of research (sugging) and FundRaising Under the Guise (frugging) are practices in which a pitch or a solicitation is disguised as a survey. They are prohibited by every professional code and illegal in many jurisdictions, but they are worth mentioning for a reason beyond their potential for direct harm. Every sugged call trains the public to distrust the next legitimate one. If you have wondered why survey response rates have collapsed over the past twenty years, the answer is not mysterious. People stopped picking up.

Integrity in the research act itself

The second lens turns from respondents to researchers. The question here is whether the research, as conducted, is capable of telling the truth.

A biased question cannot produce unbiased data. “How important is it for companies to act ethically?” is a question almost no one will rate as unimportant. “Do you agree that our product is the best on the market?” is a question that measures the respondent’s willingness to agree. A question that leads the respondent to an answer serves questionable research purposes. A useful self-check is to imagine a question returning the result you least want. Would you still defend the wording? If a bad result would make you say “well, the question wasn’t quite right,” then the question wasn’t measuring what you claimed. It was measuring how well you steered the respondent.

Researchers fool themselves more often than they defraud others. A few patterns worth watching for in your own work:

• HARKing, or Hypothesizing After the Results are Known. You ran twenty comparisons, one was significant, you write up the analysis as though that had been your hypothesis all along.
• p-hacking. You run the analysis under many specifications (i.e., different controls, different subgroups, different outlier treatments, etc.) and report the cut that “worked.”
• The forking paths problem. Even without explicit cheating, every analyst makes dozens of implicit choices about cleaning, coding, and modeling. Those “researcher degrees of freedom” aggregate into a surprisingly high rate of plausible-looking false positives.
• Confirmation bias. The most corrosive bias is often the one no one names out loud, especially when the sponsor has a favored answer or when you do. Confirmation bias affects us all. Just be aware of it and do your best to keep an open mind. Don’t be too eager to find evidence you were right. If anything, be contrarian.

Qualitative research is not exempt. When you have forty interview transcripts and choose three vivid quotes to illustrate a theme, the question to ask is whether they were chosen because they were representative, or because they sounded good. The softer the methodological constraints, the more honesty is required to stay inside them.

Represent the sample accurately. A convenience sample of forty undergraduates is not a window into “consumers.” Over-claiming scope is a form of dishonesty even when nothing is fabricated. If the sample cannot bear the weight of the conclusion, either strengthen the sample or soften the conclusion.

Honesty in reporting and use

The third lens turns outward, to the people who will act on what you have found.

Research exists to inform decisions. That means it must perform in the boardroom. But “performs well” ought to mean performs well because it is true and not because it is tidy. The most common soft fraud in marketing research is the slow smoothing of inconvenient findings until a clean narrative emerges. The axis gets truncated to make a modest effect look dramatic. The disconfirming subgroup quietly disappears. A correlation is described in the language of causation. While the research does not lie, it allows itself to be interpreted in a favorable way.

There is a simple test for this, and it is not original, but it is worth carrying with you. Would you put your name on the slide? On this finding, on this recommendation, on this headline number? If you would worry that you might be linked to this assertion, why are you sending it forward?

Research does not end when the deck is delivered. Insights are applied. Targeting decisions use them. Algorithms are trained on them. Products are designed around them. It is part of your job to understand how your findings will be used, and to ask whether the people who provided the data would recognize, and accept, what is being done with it. The archetypal failure here is Cambridge Analytica, which applied standard marketing-research techniques at scale, without consent, to exploit psychological vulnerabilities for political persuasion. The same structure shows up in less dramatic forms every week — fabricated scarcity messages, segments built around vulnerability, onboarding flows optimized for the firm rather than the customer.

A Note on AI

The 2025 revision of the ICC/ESOMAR Code was prompted, in large part, by a recognition that generative AI is rapidly reshaping every step of the research process. The big issues the industry is grappling with are synthetic respondents, AI-powered transcript coding, model-generated executive summaries, and automated targeting. These innovations don’t create entirely new ethical questions as much as they intensify old ones.

It is worth briefly walking the three lenses through this layer.

Obligations to respondents become harder to track. The most contested development is the synthetic respondent—an AI-generated persona prompted to “respond” as a member of a target audience. The technique can be useful for early concept screening, but problems begin the moment synthetic insights are presented as if they came from real people. Disclosure obligations attach not only to the simulated participants (who don’t exist) but to the real ones whose data trained the model, often without consent for that use. There is also a quieter problem. Synthetic respondents are typically trained on historical survey and polling data, which means they inherit the demographic skews, response biases, and blind spots of the samples that produced them. A synthetic panel is only as representative as the human one it was distilled from, and usually less so. The emerging standard is straightforward. Synthetic insights must be clearly labeled, must not be used to substantiate marketing claims, and must be validated against human research before driving consequential decisions. Add to this an extra obligation on the part of the researcher: know the source of the training data.

Integrity in the research act becomes harder to audit. When an LLM codes interview transcripts, generates themes, or summarizes open-ended responses, the garden of forking paths grows considerably. Every prompt is a researcher decision, every model version is a hidden parameter, and the polished prose of the output can mask interpretive choices a human analyst would have flagged. Researchers have started calling the resulting effect “authority laundering,” which is simply the use of a model’s confident language to justify findings that have not been independently validated. The practical implication is clear. Keep an audit trail of prompts, model versions, and the reasoning behind any acceptance or rejection of the model’s output. Hallucination and bias amplification are not occasional bugs. They are persistent features.

Reporting and downstream use become more dangerous because they become more pervasive. AI-enabled targeting magnifies the autonomy and manipulation concerns from the previous section. The same techniques that allow precise, helpful personalization also enable precise, exploitative nudging at scales no human marketer could achieve. The Cambridge Analytica case took a team of specialists; an equivalent operation today takes a competent prompt and an API key.

The fifth principle of the 2025 ICC/ESOMAR Code addresses this directly: researchers bear responsibility for the work they put their name on, regardless of which tools or techniques produced it. That principle is the bridge to the practical guidance below.

When you’re not sure…

You will encounter questions where the right answer isn’t obvious. When that happens, three habits will serve you better than instinct.

Lean on the codes. The ICC/ESOMAR International Code, revised in 2025, is the field’s accumulated consensus on how to do this work well. Its five principles distill to a working summary:

1. Be honest and transparent
2. Take due care to avoid harm
3. Communicate clearly with participants
4. Do nothing that erodes public trust in research itself
5. Accept responsibility for the work

The Insights Association code provides comparable guidance for US-specific contexts.

Ask before you act. Most ethical missteps in commercial research happen in private, under deadline. Before signing off on a borderline call, find one person whose judgment you trust and walk them through it. The act of explaining the situation out loud is often clarifying on its own.

Write it down. When you make a judgment call about how to phrase a question, how to handle a missing subgroup, or what to include or omit from a reported finding, note what you decided and why. Over a career, that record is what tells you whether you have been holding the line or quietly drifting.

The professionals who do this work well use the tools the field has built, lean on colleagues, and keep meticulous records. When the work is done, they can name what they did and why. That’s your north star.

Further Resources

• Insights Association, Code of Standards and Ethics for Marketing Research and Data Analytics. The leading US professional code, updated regularly and more specific than most readers expect.
• ESOMAR, Global Guideline on Market, Opinion, and Social Research. The international counterpart.