What Is Maze?: And How Does It Compare to Sprig

Example H2

Introduction

Maze is a user research platform that combines participant recruiting, unmoderated testing, AI-moderated interviews, and analysis in one product. Product designers and researchers use Maze to test Figma prototypes, live websites, and native mobile apps, run card sorts and tree tests, and interview users with an AI Moderator. Maze answers whether people can use a design and why they struggle. Maze's own blog says it is not built for large quantitative market studies or long-running tracking. Sprig is a leading choice for teams that need that quantitative side and in-product reach together, pairing Maximum Difference Scaling (MaxDiff), conjoint analysis, and response quotas on its Enterprise plan with in-product surveys on web and native mobile apps.

Key takeaways about Maze

Maze is a testing and interview platform first, with surveys as one method among several:

Maze tests Figma prototypes, live websites, and native mobile apps with real participants.
Maze runs card sorting and tree testing for information architecture decisions.
Maze's AI Moderator conducts interviews without a human moderator and is sold as an add-on to Enterprise plans.
Maze's MCP server, launched June 24, 2026, lets AI tools query existing research data.
Maze's own blog says it is strongest for fast feedback, not for large quantitative market studies or long-running tracking programs.
Maze documents no MaxDiff, conjoint analysis, response quotas, or significance testing in its survey feature.

What is Maze?

Maze describes itself as "a user research platform that brings recruiting, testing, and analysis together." Participant sourcing, study building, and reporting sit in one workspace, so a designer can go from a prototype link to a shareable report without leaving Maze.

In plain terms, Maze is where product teams check whether a design works before engineering builds it. Rather than waiting for a research team to schedule sessions, a designer typically sets up a prototype test, sends it to the Maze panel, and reads an automated report.

Maze's mission statement, from its 2022 Series B announcement, is to "empower product teams to run research in the same agile way they build products."

Maze's research methods, as listed on its platform page

Maze lists eight methods on its user testing platform page:

Prototype testing
Live website testing
Mobile testing
Card sorting
Tree testing
Surveys
Moderated interviews
AI moderator

The next section turns those eight methods into the jobs teams actually hire Maze for.

What is Maze used for?

Maze is used for evaluative research: testing a specific design, flow, or piece of content with real people before or after launch. The common jobs cluster into five groups.

Validating a prototype before it reaches engineering

Prototype validation is the job Maze is best known for. A product designer at a banking app might send a Figma prototype of a new transfer flow to a small panel sample, 30 participants for illustration, and see where they misread the confirmation screen.

Reviewers on G2 frequently praise the Figma integration for prototype testing. Rather than exporting screens and rebuilding them in a testing tool, designers test the prototype they already have.

Fixing navigation and information architecture

Card sorting and tree testing typically show how people group content and where they expect to find it. A software company restructuring its settings menu can run a tree test to learn whether users look for "Billing" under "Account" or under "Workspace."

Interviewing users at a volume a researcher could not moderate alone

Maze's AI Moderator runs interviews without a human on the call. Maze's April 2026 blog post reports that one unnamed team "ran 200 interviews in just 10 days."

Testing concepts, copy, and messaging

Maze's April 2026 update added visual stimulus testing for concepts, designs, and copy. Maze positions the feature for testing reactions to a headline, a landing page, or a new product name before it ships.

Collecting feedback on a live website

Maze's In-Product Prompts place surveys and screeners on a live website or web app, and they are often the quickest way to reach current visitors. A team can ask a short satisfaction question on the checkout page, or invite qualified visitors to book an interview.

How does Maze work?

Maze works by pairing a study method with a participant source and an automated report. A researcher picks the method, chooses who takes part, and reads results that Maze assembles into clips, themes, and charts.

The table below maps each Maze method to the kind of evidence it produces, which decides the questions each method can answer.

Every method in the table except surveys produces behavioral or qualitative evidence, and even Maze surveys are directional rather than population estimates.

Behavioral evidence: task tests on prototypes, sites, and apps

Behavioral tests record what participants do rather than what they say. Maze reports whether participants complete a task and where they deviate, and pairs that with recordings and clips.

For native apps, Maze's help center recommends the App Test block inside the Maze Participate mobile app. Screen recording is available there, though not when participants test through a mobile browser.

Structural evidence: card sorting and tree testing

Card sorts and tree tests generally answer information architecture questions rather than preference questions. The output is a map of how people think content is organized, which is the input a navigation redesign needs.

Survey evidence: opinions, ratings, and open text

Maze surveys support Likert-style opinion scales, multiple choice, multi-select, ranking, yes or no, number inputs, open text, and demographic fields. Maze's AI can generate up to three follow-up questions from an open-text answer.

Maze's survey page does not document significance testing, weighting, response quotas, MaxDiff, or conjoint analysis. Maze surveys are typically used to add a rating or an open question to a test rather than to size a population.

The next section looks at the AI layer that runs across these methods.

What AI capabilities does Maze offer?

Maze offers AI in four places: an AI Moderator for interviews, an AI Study Builder for setup, Maze AI for analysis, and an MCP server for connecting research data to AI assistants. Most of it sits on Enterprise plans.

Maze AI Moderator

Maze's AI Moderator runs interviews from a stated research goal. Maze says it "probes deeper when it matters, moves on when the goal is reached, and runs any time in any language."

Outputs include traceable quotes, synthesized themes, and auto-generated reports. The April 2, 2026 expansion added visual stimulus testing and a choice between freeform and structured discussion styles. Maze lists screen sharing for prototype walkthroughs as coming soon.

Maze's AI Moderator page states the AI Moderator "is available as an add-on with our Enterprise plans." Interview studies also sit on Enterprise, so teams on other plans typically rely on the AI follow-up questions in Maze surveys instead.

Researchers remain responsible for the interview goal, the discussion guide, and reading the themes against the transcripts.

Maze AI Study Builder and Maze AI analysis

The AI Study Builder drafts a study from a goal, and Maze AI handles analysis and theme generation. Maze's pricing page lists both as Enterprise features, and Maze's survey page says its AI can flag bias in question wording and run sentiment analysis on responses.

Maze surveys also include an AI Conversation block that asks a voice-based follow-up after a participant answers. As with any AI-drafted study, researchers remain responsible for reviewing question wording and interpreting the themes the model returns.

Maze MCP server

Maze's Model Context Protocol (MCP) server, announced June 24, 2026, lets Claude, ChatGPT, Cursor, and other MCP-compatible tools search studies, read transcripts, and analyze past findings. Existing workspace permissions apply.

Maze's help center describes the access as querying existing research, and Maze does not document study creation through its MCP server. Researchers remain responsible for checking any AI assistant's summary against the underlying studies.

What the AI layer does not change

Maze's AI speeds up qualitative and behavioral research rather than turning it into population measurement. An AI Moderator that runs 200 interviews produces 200 accounts, not an estimate for the full user base.

That distinction frequently gets lost when teams read theme counts as percentages.

How does Maze recruit participants?

Maze recruits through four sources: its own panel, a team's own participant database, prompts on a live website, and shareable links.

Maze Panel

Maze's homepage claims a panel of more than 6 million participants. Maze's pricing page lists panel recruitment on all plans, with limits.

Maze does not state on those pages how much of the panel is owned rather than sourced from partners. Buyers with niche business-to-business targeting needs should generally ask before committing to a timeline.

Maze Reach

Maze Reach typically stores a team's own participants, and Maze's survey page describes sending studies to them by email campaign. Reach typically suits teams that research their existing customers repeatedly and want one place to manage who has been contacted.

Maze In-Product Prompts

In-Product Prompts appear as popovers on a website or browser-based product. Teams can trigger them by URL, target Amplitude cohorts, and use them for Net Promoter Score (NPS), customer satisfaction (CSAT), and product-market fit questions.

Prompts can also carry a screener linked to a scheduler such as Calendly, so qualified visitors can book a session without leaving the site. That makes In-Product Prompts as much a recruiting tool as a feedback tool.

Maze's In-Product Prompts page describes web and browser-based products only. Maze does not document an SDK for prompting users inside a company's own native mobile app, and mobile testing happens inside the Maze Participate app instead.

Shareable links

Any Maze study can go out as a link through a team's own channels, such as a newsletter, a community, or a support email.

Who makes Maze?

Maze was founded in 2018 by Jonathan Widawski, who is CEO, and Thomas Mary, who was CTO at the time of the Series B. In June 2022, Maze raised a 40 million dollar Series B led by Felicis, bringing total funding to 60 million dollars at the time, according to TechCrunch and Maze's own announcement.

At that point Maze reported more than 130 employees across 35 countries. Maze's homepage displays customer logos including Adobe, SAP, Toyota, and Fidelity, retrieved September 2026.

Funding after 2022 is reported inconsistently by data aggregators, so this guide relies only on the primary announcement.

What do buyers say about Maze?

Buyer reviews generally rate Maze highly for ease of use and the Figma integration, with complaints centered on mobile testing limits, prototype loading, and a lower customer service score on Capterra's small sample. Two review platforms carry enough detail to cite.

On G2, Maze holds 4.5 out of 5 across 111 reviews. Reviewers commonly praise quick test setup and prototype testing, and some report limits on mobile testing, prototype loading delays, and conditional logic for branching tests.

On Capterra, Maze also scores 4.5 out of 5, but across only 10 reviews, so that score carries less weight than the G2 figure. Capterra reviewers rate customer service at 3.8, against 4.3 for ease of use.

Where is Maze the strongest fit?

Maze is the strongest fit for design and product teams that need behavioral and qualitative evidence quickly, without a dedicated research operations function. Five situations stand out:

Maze fits design teams that validate Figma prototypes every sprint before handing work to engineering.
Maze fits navigation and information architecture projects that need card sorts and tree tests together.
Maze fits teams that want interview volume from an AI Moderator rather than a researcher's calendar.
Maze fits organizations spreading research across designers and product managers who are not trained researchers.
Maze fits early concept and messaging reactions where the reasoning behind a response is the deliverable.

A buying committee that only needs evaluative design research may find Maze covers the whole job.

Maze platform summary

Strengths. Breadth of evaluative methods, Figma-native prototype testing, card sorting and tree testing, an AI Moderator for interviews at volume, a large published panel, and 4.5 out of 5 on G2 across 111 reviews.

Limitations. No documented MaxDiff, conjoint analysis, response quotas, or significance testing. In-product prompts cover the web only. Several core capabilities, including the AI Moderator, mobile testing, and card sorting, sit on Enterprise plans.

Maze is best for product design and research teams that need to know whether people can use a design and why they struggle, from prototype through launch.

What is Maze not built for?

Maze is not built for questions that need a number to hold across a whole population, repeated over time. Maze's March 2026 market research post describes Maze as strongest for fast feedback on concepts and messaging, "not when you're running large, complex quantitative market studies or long-running tracking programs."

Each limit below names the better instrument.

Large quantitative studies and tracking over time

Tracking satisfaction or brand health quarter over quarter generally needs repeated runs of an identical instrument to a defined audience. A survey platform built for longitudinal measurement, such as Sprig, is the better instrument, and the brand tracking guide covers the design.

Ranking features and pricing trade-offs

Maze documents ranking questions but not Maximum Difference Scaling, commonly called MaxDiff, or conjoint analysis. A platform that ships MaxDiff and conjoint as question types, such as Sprig, is the better instrument for prioritization and pricing decisions.

Surveying users inside a native mobile app

Maze does not document an in-app SDK for native mobile apps. A product team that wants to ask users a question at the moment they finish a flow in its own iOS or Android app needs an in-product survey tool with native mobile SDKs, such as Sprig.

Statistically defended estimates

Maze does not document weighting or significance testing. Teams that must defend a difference between segments to a finance or executive audience typically export data to a statistical package such as R or SPSS.

Behavioral analytics across every user

Maze tests recruited participants rather than instrumenting every session in production, so its samples are typically small and chosen. A product analytics tool such as Amplitude or Mixpanel is the better instrument for funnel metrics across the full user base.

A research repository across many tools

Maze stores Maze studies, and research programs increasingly span more sources than one testing tool. Teams consolidating interviews, surveys, and support tickets from many sources generally use a dedicated research repository such as Dovetail rather than a testing platform.

How does Maze compare to Sprig?

Maze is built for evaluative research and interviews. Sprig is an enterprise research platform built around AI agents for study design, fielding, and synthesis, and it is designed for quantitative measurement across in-product, email, link, and panel channels.

The table scores ten dimensions from each vendor's published documentation, retrieved September 2026.

Where Maze wins

Maze wins four rows outright: testing breadth, card sorting and tree testing, AI-moderated interviews, and panel size. Maze's method set covers Figma prototypes, live websites, and native apps, where Sprig offers a Prototype Testing study type and Recorded Task.

Sprig does not document card sorting or tree testing, and Sprig runs no moderated or AI-moderated interviews. Maze's published panel is also larger than Sprig's.

Where Sprig wins

Sprig wins the four measurement rows and the MCP row. Sprig ships MaxDiff and conjoint as question types on the Enterprise plan, with conjoint limited to link surveys, and Sprig's SDKs place surveys inside native mobile apps.

Sprig documents response-based quotas on the Enterprise plan and supports multiple runs of one study for longitudinal work. Sprig's MCP server can create a study in draft, and a human approves it in the Sprig app before it reaches anyone.

Where neither wins

Published statistical methodology is a tie. Sprig, like Maze, does not publish its weighting approach, sample-size guidance, or significance-testing method, so teams with strict inference requirements typically finish the analysis in a statistical package.

Where Maze and Sprig overlap

The two platforms overlap in three places. Both run surveys on websites and web apps. Both test prototypes, since Sprig can embed a Figma or AI-generated prototype in a study and capture screen, voice, and video through Recorded Task.

Both also capture behavior around a live product, with Maze through live website tests and Sprig through heatmaps and session replay clips tied to in-product responses. Rather than treating the overlap as redundancy, assign each overlapping job to whichever tool already holds the relevant audience.

Sprig platform summary

Strengths. MaxDiff and conjoint analysis, in-product surveys on web and native mobile, email with a custom sending domain, response quotas, repeated runs, and three AI agents for design, fielding, and synthesis.

Limitations. No moderated or AI-moderated interviews, no documented card sorting or tree testing, a smaller published panel, and no published statistical methodology.

Sprig is best for product, research, and marketing teams that need repeatable numbers on how many users feel something, which option wins, and whether a metric moved.

Choosing between Maze and Sprig

Choose Maze if:

Your core question is whether people can complete a task in a design
Your team runs card sorts and tree tests for navigation work
You want AI-moderated interviews at volume
Your research runs mostly on Figma prototypes before launch

Choose Sprig if:

Your core question is how many users share a problem or preference
Your team ranks features or pricing options with MaxDiff or conjoint
You need to survey users inside a native mobile app
You track the same metric across releases or quarters

How can teams use Maze and Sprig together?

Teams use Maze and Sprig together by routing each question to the instrument that produces the right evidence. Maze finds the problem and explains it, and Sprig measures how big it is and whether a fix moved it.

Rather than forcing one tool to do both jobs, the router below assigns each common question to a tool.

A question-to-instrument router for product teams

Questions starting with "can," "where," and "why" typically belong to Maze, while questions starting with "how many," "which," and "did it move" belong to Sprig.

Questions that need both a reason and a size generally need both tools, with the Maze study defining what the Sprig study measures.

Worked example: from a prototype test to a ranked fix list

The figures in this example are illustrative, not benchmarks.

A fintech product team runs a Maze prototype test of a redesigned onboarding flow with 40 panel participants.
The Maze report surfaces six friction points, from an unclear identity check to a confusing account-type choice.
The team can fix two this quarter, so it needs to know which friction points matter most to real users.
The team runs a Sprig MaxDiff study as an in-product survey to recently onboarded users on web and in the native app.
Response quotas hold the sample balanced across free and paid plans, so one segment does not dominate.
The MaxDiff results rank the six friction points, and the team fixes the top two.
After release, the team runs an in-product question on the fixed steps, with identical wording across runs, to check whether reported friction fell.

Rather than choosing two fixes from the six loudest clips, the team chooses from a ranking across its users. Rather than relying on the next usability test to confirm the fix, the team measures it.

Researchers remain responsible for the MaxDiff design and for reading the results in context.

Four mistakes teams make when pairing Maze and Sprig

Pairing the two tools fails in predictable ways, and each failure comes from asking one kind of evidence to do another kind's job:

Reading a Maze survey rating from a small panel sample as the satisfaction of the whole user base
Treating AI Moderator theme counts as the share of users who hold a view
Running a MaxDiff study on friction points before qualitative work has defined them clearly
Changing the survey wording between runs, which breaks comparison with the earlier results

Rather than trusting a chart because it shows a percentage, name the population that produced it before reporting the number.

Frequently asked questions about Maze

What is Maze used for?

Maze is used for evaluative research: testing Figma prototypes, live websites, and native mobile apps with real participants, running card sorts and tree tests, and interviewing users with an AI Moderator. Product designers and researchers typically use Maze to find where people struggle with a design before engineering builds it, and to hear why.

Is Maze a survey tool?

Maze is a user research and testing platform with a survey feature, not a survey platform. Maze surveys support opinion scales, multiple choice, ranking, and open text. Maze does not document MaxDiff, conjoint analysis, response quotas, or significance testing, and Maze's own blog says it is strongest for fast feedback rather than large quantitative studies and long-running tracking.

Does Maze have an AI moderator?

Maze has an AI Moderator that runs interviews from a stated research goal, probes on answers, and returns quotes, themes, and reports. An April 2, 2026 update added visual stimulus testing, with screen sharing for prototype walkthroughs listed as coming soon. Maze sells the AI Moderator as an add-on to Enterprise plans.

Does Maze have a participant panel?

Maze has a participant panel, which its homepage describes as more than 6 million people. Panel recruitment is listed on all Maze plans, with limits. Maze also offers Reach for a team's own participants and In-Product Prompts for recruiting visitors on a live website.

How is Maze priced?

Maze publishes plan tiers on its pricing page, and this guide does not restate prices because Maze's page is the current source. Maze places several capabilities on Enterprise plans, including the AI Moderator as an add-on, the AI Study Builder, Maze AI analysis, mobile testing, interview studies, card sorting, and single sign-on (SSO). Panel recruitment is listed on all plans with limits. Confirm current packaging with Maze directly before budgeting.

What is the Maze MCP server?

The Maze MCP server, announced June 24, 2026, connects Maze research data to Claude, ChatGPT, Cursor, and other Model Context Protocol tools. Users can search studies, read transcripts, and analyze findings from inside an AI assistant. Maze documents query access to existing research, not study creation.

Can Maze survey users inside a native mobile app?

Maze does not document an SDK for surveying users inside a company's own native mobile app. Maze In-Product Prompts cover websites and browser-based products, and Maze tests native apps through the Maze Participate app instead. Teams that need in-app surveys on iOS or Android typically use a survey platform with native mobile SDKs.

Does Maze support MaxDiff or conjoint analysis for quantitative research?

Maze does not document MaxDiff or conjoint analysis on its survey page. Maze surveys include ranking questions, which order a short list but do not estimate relative preference the way MaxDiff does. Sprig ships MaxDiff and conjoint as first-party question types on its Enterprise plan, with conjoint limited to link surveys.

Who founded Maze?

Jonathan Widawski and Thomas Mary founded Maze in 2018, and Widawski is CEO. In June 2022, Maze raised a 40 million dollar Series B led by Felicis, bringing total funding to 60 million dollars at the time and supporting a team of more than 130 people across 35 countries.

What is the difference between Maze and Sprig?

Maze is a user research platform for testing designs and interviewing users. Sprig is an enterprise research platform built around AI agents for measuring how many users share a problem, which option wins, and whether a metric moved. Maze wins on prototype testing, card sorting, tree testing, and interviews, while Sprig wins on MaxDiff, conjoint analysis, native mobile surveys, quotas, and tracking.

Can Maze and Sprig be used together?

Maze and Sprig are commonly used together, because each answers questions the other does not. A typical pairing uses a Maze prototype test to surface friction points, then a Sprig in-product MaxDiff survey to rank those friction points across the user base, then a repeat Sprig survey after release to check whether reported friction fell.

Is Maze good for quantitative research?

Maze is good for directional quantitative signals such as task completion and simple survey ratings, but Maze's own blog says it is strongest for fast feedback rather than large, complex quantitative market studies or long-running tracking. Teams that need MaxDiff, conjoint analysis, quotas, or repeated measurement generally pair Maze with a survey platform built for measurement.

Bottom Line

Maze is a well-reviewed user research platform for testing designs and interviewing users, with eight research methods listed on its platform page, from prototype testing to an AI Moderator. Maze answers whether people can use something and why they struggle.

If your team's open questions start with "can" and "why," Maze may cover the whole job. If your questions start with "how many," "which one," and "did it move," you need a survey platform built for measurement.

For that measurement work, Sprig ships MaxDiff and conjoint analysis on its Enterprise plan, response quotas, repeated survey runs, and in-product surveys on web and native mobile apps. Teams that ask both kinds of questions can use the two tools together, rather than stretching either one past its method set.