Anchoring bias in product research happens when the first number, idea, example, or interpretation in a study pulls later answers towards it. A price mentioned before a willingness-to-pay question, a mockup shown before discovery, or a theme named after the first interview can all become anchors.
The practical fix is to control the order of evidence. Ask about the participant’s recent experience before showing your solution. Let them provide numbers before you name any. Use the same neutral opening across sessions. Decide how you will interpret the results before the most memorable answer arrives.
Anchoring is not proof that every early detail corrupts every interview. It is a well-established tendency, and its size depends on the task and context. The goal is to remove avoidable anchors, not to pretend that research can become perfectly neutral.
What is anchoring bias?
Anchoring bias is the tendency for an initial value or piece of information to influence a later judgement. Tversky and Kahneman originally described people starting from an initial value and adjusting away from it too little. Later work suggests that different kinds of anchors can operate through different processes, including selective attention to information that fits an externally supplied anchor.
Their best-known experiment makes the effect easy to see. Participants watched a rigged wheel stop on either 10 or 65. They then judged whether the percentage of African countries in the United Nations was above or below that number and estimated the percentage. The group shown 10 gave a median estimate of 25%. The group shown 65 gave a median of 45%. The number was visibly arbitrary, yet the estimates moved towards it.
Another experiment, summarised by Nielsen Norman Group, gave people five seconds to estimate one of two mathematically identical expressions:
- 1 × 2 × 3 × 4 × 5 × 6 × 7 × 8
- 8 × 7 × 6 × 5 × 4 × 3 × 2 × 1
The true answer is 40,320 in both cases. The median estimates were 512 and 2,250. The numbers at the beginning of the expression changed the estimate even though the underlying problem did not change.
Product research creates similar starting points. The anchor may be relevant, such as a current price, or arbitrary, such as the first solution a founder happens to mention. Either can narrow the participant’s frame before they have described their own.
Expertise does not make anchors harmless
Anchoring is easy to dismiss as a trivia-test problem. Northcraft and Neale tested it in a richer professional setting.
They gave students and working estate agents information about a real house in Tucson. Participants received a 10-page pack, reviewed comparable properties, and toured the house for up to 20 minutes. The house’s actual listing price and appraised value were both $74,900. The researchers changed the listing price shown in the pack.
For the estate agents included in the first experiment’s analysis, the low and high conditions produced this result:
| Listing-price anchor | Mean appraisal estimate |
|---|---|
| $65,900 | $67,811 |
| $83,900 | $75,190 |
The same house and supporting evidence produced appraisal estimates more than $7,000 apart. In a second experiment, listing price significantly influenced all four estimates made by both professionals and students. Yet only 19% of the professionals, compared with 37% of the students, named listing price as a factor they had considered.
This does not mean expertise never helps. It means expertise alone is not a research control, and self-report is a poor way to detect whether an anchor worked.
Warnings are also limited. Epley and Gilovich found that forewarning and incentives affected responses to some self-generated anchors but not externally provided anchors in the same way. In a customer interview, your price or mockup is externally provided. Telling someone to ignore it does not restore the conversation you would have had before showing it.
Where anchors enter product research
Anchoring is not one interview mistake. It can enter during study design, moderation, and analysis.
| Anchor source | What it can change | Safer approach |
|---|---|---|
| Product pitch before discovery | Which problems feel relevant enough to mention | Collect a recent story before describing the product |
| Price or percentage named first | The range of later estimates | Ask about current spend, cost, frequency, or value before testing a price |
| Mockup shown at the start | How the participant frames the problem | Separate discovery from concept evaluation |
| Fixed question order | The context carried into later answers | Put open questions first; randomise independent items when comparison matters |
| First participant or vivid quote | Which later evidence the team notices | Define themes and decision rules before synthesis |
The pitch becomes the problem frame
Suppose you open with: “We are building an AI interviewer that helps teams get more honest feedback.” You then ask what makes customer research difficult.
The participant now knows that honesty and interviewing matter to you. They may search their memory for examples that fit those topics, even if recruitment, analysis, or internal politics is the larger problem. You have not merely explained the study. You have supplied a theory of the problem.
A neutral opening gives the participant room to set the frame:
Tell me about the last time you needed to learn something from customers. What were you trying to decide?
Use the story to discover what mattered before you introduce your category or solution.
A number narrows the pricing conversation
“Would you pay €50 a month?” can tell you how someone reacts to €50. It cannot tell you what number they would have produced without that anchor.
Start with evidence that already exists:
- What do you use today?
- What does it cost?
- How much time does the workaround take?
- Who owns the budget?
- What happened during the last purchase or renewal?
If you need to test a specific price, do it after you have recorded the participant’s current spend, alternatives, decision process, and unaided expectations. Treat answers after the price appears as reactions to that price.
Maren’s pricing interview guide gives a full script that avoids “would you pay?” as the primary evidence.
A mockup replaces discovery with evaluation
A prototype is useful when your question is whether people understand or value a particular concept. It is harmful when your question is how they currently solve a problem.
Once a participant sees your dashboard, the interview is about the dashboard. Their spreadsheet, email chain, or manual workaround becomes harder to recover because you have supplied a more polished answer.
Run the conversation in phases:
- Unaided discovery: Reconstruct a recent event and current workaround.
- Unaided expectations: Ask what the participant would look for or try next.
- Stimulus: Show the same concept in the same way to each participant.
- First reaction: Capture what they notice before explaining it.
- Evaluation: Probe understanding, relevance, concerns, and fit.
This preserves evidence from before and after the stimulus. Use the separate concept testing interview script when the concept itself is the object of study.
Question order creates context
Pew Research Center documents an October 2003 poll in which 45% favoured legal agreements giving same-sex couples the same rights as married couples when the question followed one about marriage. Support was 37% without that immediately preceding context.
That is an eight-point difference associated with question order. Pew explains that earlier questions can produce contrast or assimilation effects and randomises many response options to distribute order effects rather than letting one order dominate.
An interview is not a representative opinion poll, so do not transfer the eight-point result to your study. Transfer the design lesson: earlier questions can supply context for later ones.
For semi-structured interviews, keep the opening and core topics stable enough to compare sessions. Put broad, unaided questions before detailed prompts. If two independent concepts may contaminate each other and you need a fair comparison, vary their order deliberately and record the condition.
The first interview anchors the analysis
The first participant says onboarding is confusing. By interview four, “onboarding confusion” has become a theme, and every pause or support story gets filed underneath it.
This overlaps with confirmation bias, but the timing matters. The early theme supplies the anchor; confirmation bias helps it survive.
Reduce the risk before the first session:
- write the research questions and expected counter-evidence;
- define what counts as a repeated pattern;
- keep raw excerpts linked to participant and context;
- code at least an initial batch together rather than finalising themes one interview at a time;
- actively record evidence that does not fit the current interpretation.
A striking first quote can be important. It should not silently become the taxonomy for everything that follows.
How to avoid anchoring bias in customer interviews
Use this six-step process:
- Write a neutral objective. “Understand how customers prepare monthly reporting” leaves more room than “validate demand for automated reporting”.
- Standardise the opening. Give every participant the same necessary context, consent information, and first question.
- Collect unaided evidence first. Ask for a specific recent event before naming features, prices, problems, or solutions.
- Let the participant provide numbers. Record current cost, time, frequency, and thresholds before introducing yours.
- Separate discovery from stimulus. Put mockups, demos, and price tests in a clearly marked later phase.
- Pre-commit to analysis rules. Decide what would strengthen, weaken, or complicate the hypothesis before reading the most persuasive transcript.
This process also helps with leading questions, but the two problems are different. A leading question suggests a preferred answer through wording. An anchor can be neutral in tone and still establish the reference point.
A ten-minute anchoring audit
Open the notes or transcript from your last five customer conversations.
First, find the earliest mention of your product, proposed solution, or category. Did participants describe their own situation before they heard your framing?
Second, highlight every number the interviewer supplied: price, time saving, frequency, percentage, team size, or benchmark. For each one, mark which later answers may have used it as a reference point.
Third, compare the opening and core question order across sessions. Variation is not automatically bad in a semi-structured interview, but undocumented variation makes comparisons harder to defend.
Fourth, write down the theme you believed after the first interview. Look for later excerpts that were forced into it, as well as evidence that challenged it.
Finally, label every result as either before stimulus or after stimulus. If you cannot tell, the next study needs a clearer boundary.
The audit will not calculate how much bias entered the study. It will show where you gave it the opportunity.
What this means for interviews run with Maren
Maren lets a researcher define the objective and discussion guide, choose an interview style and depth, and run adaptive follow-up questions. That structure can make openings and research instructions more consistent across participants.
It does not make a study anchor-free. A researcher can still write a loaded objective, put a target price in the guide, or ask for concept feedback before collecting unaided experience. Adaptive follow-ups also mean conversations will not be identical, which is appropriate for qualitative research but makes a neutral starting point even more important.
There is a second reason for caution. A 2025 Findings of EMNLP paper found anchoring effects in controlled LLM price-negotiation simulations. That is not a study of AI-led user interviews, so it cannot tell us how much anchoring occurs in this setting. It is enough to reject any claim that an AI interviewer is inherently immune.
Use the same controls you would expect from a careful human moderator:
- keep the objective descriptive rather than confirmatory;
- place past-behaviour questions before solution or price questions;
- keep essential context consistent;
- review the generated guide for numbers and assumptions;
- separate discovery interviews from concept or pricing studies when possible.
Maren can apply a research structure consistently. The researcher still owns the structure.
The simple rule
The first information in a product-research conversation deserves more scrutiny than the most enthusiastic answer at the end.
Ask for the participant’s story before supplying yours. Let their numbers enter the room before your price. Show the mockup only after you understand the current behaviour. Decide how you will evaluate the evidence before the first memorable quote becomes the frame.
Anchoring cannot be removed by asking everyone to be objective. It can be reduced by designing the study so fewer avoidable anchors appear before the evidence you care about.