Make people choose
List 6 to 30 comparable items. Respondents see them 3 to 5 at a time, pick one as most and a different one as least, and move on to the next set ("Set 3 of 9"). Their progress saves after every set.
MaxDiff
Find out what matters most, not what people rate highly
Ask people to pick the best and the worst from a few items at a time, and get a ranked list of features, benefits or messages, each with a share of preference and a 95% interval. The design, the sets and the model are built in.
New accounts start with a 14-day trial of Growth, MaxDiff included.
Stop asking people to rate everything a 5
Ask customers to rate ten features from 1 to 5 and most come back as important. Ask them to rank all ten and the order falls apart after the first few, especially on a phone. MaxDiff asks for one easy decision at a time: of these four, which matters most and which matters least? Across a dozen small sets, those choices separate the items a rating scale lumps together, so you know what to build, say or test first.
No design software, no statistician on call
When you publish, SurveyRock builds 100 balanced versions of the design and gives each respondent one, so across your sample each item is shown about equally often, next to every other item and in every position. Recommended settings follow your list length, with the number of sets and the time each respondent needs shown as you go.
A ranking you can defend in the meeting
The results card ranks every item by its share of preference, adding to 100, with a 95% interval on each. The data table adds the model's utilities, best-minus-worst scores and the raw counts. Filter the dashboard and the scores are re-estimated for that segment.
No score when the data can't carry one
Below 30 respondents, when an item was shown fewer than 30 times, or when the data can't separate the items, the card shows raw counts and says why instead of printing shares. It also warns you when scores rest on fewer than about 100 respondents or when many people stopped part-way.
Take it further in R or your own tools
Download one row per respondent and set, with the items shown, the most and least picks and any weights, in the long format R's survival::clogit takes. Use it for individual-level estimation such as hierarchical Bayes, which SurveyRock doesn't calculate.
Every respondent answers the same study
Once a MaxDiff has responses, its items and design can't change, so every respondent answers the same design. You can still fix wording. To change the design, duplicate the survey and start fresh.
How it works
From a long list to a clear order
- 1
Add your list
Drag MaxDiff into the survey and paste your items, one per line: features, benefits, messages. Write the question as the thing the sets answer.
- 2
Keep the recommended design, or tune it
Items per set and times each item is shown start on Recommended. The line underneath tells you how many sets each respondent will answer and how long it takes.
- 3
Publish and collect
SurveyRock builds the balanced versions at publish. Respondents answer one set at a time on any device.
- 4
Read the ranking
The card ranks items by share of preference with intervals. Switch the dashboard filter to compare segments, or export the design and choices for deeper modelling.
Questions
Step by step: MaxDiff questions
How is the score calculated?
A multinomial logit on the most and least picks, with cluster-robust standard errors by respondent. Shares come from the zero-centred utilities and are rescaled to add to 100; the 95% intervals use the delta method on the log share, so they stay above zero. With a weighting scheme, each respondent's contribution is multiplied by their weight.
Is an item with twice the share twice as important?
It was preferred about twice as often. That isn't the same as twice as important: MaxDiff measures relative preference, and whether its scores behave as a ratio scale is disputed in the research literature.
Can I test whether two items differ, or two segments?
SurveyRock doesn't run significance tests between items or between groups, and overlapping intervals aren't a test. To compare segments, switch the dashboard filter and read each segment's scores.
Do I get each respondent's own scores?
No. The scores describe a group. For individual-level scores, export the design and choices and estimate them in your own tools.
Does it tell me whether anything matters at all?
No. MaxDiff ranks the items against each other; if none of them matter to someone, they're still ranked. That needs an anchored design, which SurveyRock doesn't offer.
Can I build one through the API or with Claude?
Yes. The API's createSurvey and addQuestions, and the MCP tools create_survey and add_questions, create a MaxDiff with the question type maxdiff, on Growth and up.
Which plans include MaxDiff?
Growth, Pro and Enterprise. New accounts start with a 14-day trial of Growth. If an account moves below Growth, a MaxDiff it already has keeps collecting and its results stay visible, but a new one can't be added.
How does this compare with Qualtrics?
Qualtrics includes MaxDiff among the "20+ pre-built, guided solutions" of Strategic Research, its online plan at $420 a month billed annually ($5,040 a year) (qualtrics.com/buy-online, read September 30, 2026). On SurveyRock, MaxDiff is part of Growth at $1,790 a year, or $179 billed monthly.