The 24 best AI user research tools for product managers in 2026

Peter Kildegaard · 7 October 2026

Every AI research tool now writes a summary. The question that separates them is what happens when you click on a finding: do you land on the second a customer said it, on a cited transcript, or on nothing but the AI's own words? A product decision built on a summary you cannot check is a guess with good grammar.

This guide is for a PM, product lead or founder who needs customer evidence for a decision. It sorts 24 tools by the research job they do, shows for each what a finding opens on, what it costs and when to pick the nearest alternative instead, and ends with two workflows that carry a question through to a spec.

Key takeaways

Before you act on an AI research finding, open the customer’s own words behind it. Choose tools that put those words one click away from the finding.
For new answers from your own customers, start with Feedbaq. People show you the problem on screen or explain it by voice, the AI follows up on what they said, and your agent links every quote to the moment it was said. It is free in your AI agent, which becomes your expert researcher; collecting responses costs $49 a month.
Pick by the evidence you are missing: new explanations, people you cannot reach, task behavior, calls you already had, research you already ran, feedback that keeps arriving, or product data.
Price the study, not the seat. Credits, interview minutes and panel recruits make the bill grow with every question you ask, and ten of these tools sell paid plans only through sales.

Start here

Find the line that describes what you are missing. Each names the tool we would try first and the one to compare it with.

You need customers to explain a problem. Start with Feedbaq: they show it on screen or tell you by voice, and the AI follows up on what they said. Compare Koji for open, adaptive interviews.
You cannot reach the right people yourself. Start with Listen Labs, which recruits, interviews and reports. Compare Conveo for video interviews across languages.
You need to see whether people can do the task. Start with Feedbaq if your customers can try it on their own screen and narrate. Choose Maze when you need success rates and paths.
You lose what customers said on calls. Start with Granola. Choose Otter or tl;dv if you want to replay the customer’s voice.
Interview rounds pile up unread. Start with Looppanel on your own, or Dovetail for a team.
Feedback pours in from every channel. Start with Enterpret if you need it tied to revenue. Compare Chattermill.
A metric moved and nobody knows why. Ask the analytics tool you have, or start with PostHog on a newer product. Then ask the affected users.
You have the evidence and need the story. Connect your research tools to Claude. For one fixed set of files, use Gemini Notebook.

The 24 tools by research job

The column that matters most is the third one: where a finding leads when you click it. Every tool here uses AI to summarize. They differ in what you can open behind the summary: the exact second of a recording, a cited transcript, a clip, or only the text the AI wrote.

Hear from your own customers: Feedbaq, Koji, Perspective AI
Reach people you cannot reach yourself: Listen Labs, Outset, Conveo, GetWhy
Watch people use the product: Feedbaq, TheySaid, Maze, Lyssna, Sprig
Keep what customers said on calls: Granola, Otter, tl;dv
Find patterns across research and feedback: Looppanel, Dovetail, Enterpret, Chattermill
Product analytics: PostHog, Amplitude, Mixpanel
Question the evidence with your own agent: Claude, ChatGPT, Gemini Notebook

AI user research tools for product managers compared, prices checked 1 October 2026 and 6 October 2026
Tool	Job	A finding opens on	Price from	Free plan
Feedbaq	Own customers	The moment in the voice or screen recording	Free; $49/mo to collect	Yes
Koji	Own customers	Verbatim quote and audio	€29/mo	10 credits
Perspective AI	Own customers	Full conversation transcript	$99/mo	100 credits
Listen Labs	Recruited	The interviews behind each claim	Not published	No
Outset	Recruited	Interview recordings	Enterprise contract	No
Conveo	Recruited	Verbatim quotes and video	Not published	No
GetWhy	Recruited	Clips embedded in the report	Not published	No
TheySaid	Task tests	Recordings and clips	Enterprise, custom	5 tests/mo
Maze	Task tests	Paths and clips	On request	1 study/mo
Lyssna	Task tests	Each participant’s answers	$166/mo, annual	15 responses
Sprig	Task tests	Individual responses	Enterprise contract	No
Granola	Calls	Cited transcript, no audio	$14/user/mo	30-day history
Otter	Calls	Transcript with audio	$8.33/user/mo, annual	300 min/mo
tl;dv	Calls	Timestamp in the video	€18/seat/mo, annual	Yes
Looppanel	Analysis	Video clip per finding	$45/mo, annual	15-day trial
Dovetail	Analysis	Source material and clips	Enterprise contract	1 project
Enterpret	Analysis	Cited feedback records	Enterprise contract	No
Chattermill	Analysis	Customer quotes per theme	Not published	No
PostHog	Analytics	Session replays	Usage-based	1M events/mo
Amplitude	Analytics	Replays and feedback	Usage-based	2M events/mo
Mixpanel	Analytics	Replays and cohorts	Usage-based	1M events/mo
Claude	Your agent	The sources you connect	$17/mo, annual	Yes
ChatGPT	Your agent	The files and apps you give it	$20/mo	Yes
Gemini Notebook	Your agent	The cited passage	Free	Yes

The one test: open the words behind the finding

An AI summary is a claim. A product decision needs the evidence behind it. Take one finding a tool gives you, the kind you would put in a spec, such as “new users give up at setup because the data connection is confusing”. Now try to reach the customer’s exact words in a click or two: who said it, when, and what they said around it.

If the words are there, you can check the finding in seconds and show the team the customer instead of your conclusion. If you cannot, you are trusting a paragraph that sounds right. That is a real risk. When Nielsen Norman Group tested AI research tools, some could not attribute a quote to a session or link it to the clip, and one could not tell the researcher’s notes from what participants said. In a 2026 study in Qualitative Health Research, a language model sometimes returned quotes that nobody in the interviews had said.

The strongest answer is a recording you can play from that moment. You hear the tone, and on a screen recording you see what the customer was looking at. A text quote is good evidence. A quote you can hear is better. A paragraph with no source is a guess.

Analytics tools pass the test in their own way. Behind a finding about behavior sit the cohort definition and the session replays, and they tell you where people struggle and whom to ask. The words that explain why still have to come from those customers.

You can run the same test on any agent. Ask it: “For each theme, give the number of different participants, their exact words and a link to each source. Include answers that contradict the theme.” Then open three of the links. Count people, not answers: five answers from one customer are one voice.

How we chose

This guide is desk research, not a hands-on test. We read each vendor’s current pricing page, product documentation and customer stories, on 6 October 2026 for tools new to our lists and on 1 October 2026 for tools in our broader PM guide. The main sources are linked at the end. Where a vendor publishes no price, the entry says so.

A research job, done with AI. Each tool had to collect, capture, organize or analyze customer evidence with an AI feature a PM can use today.
A price or a named customer. Every tool publishes a price or a free plan, or names customers in its own stories. Several AI interview tools with neither stayed out.
A path back to the source. We describe what a finding opens on for every tool, because that decides how far you can trust it.
A clear difference from the tool next to it. Each entry names the nearest alternative and when to pick it instead.

We make Feedbaq and recommend it first for hearing from your own customers. Its founder worked as a staff product manager at Maze, one of the tools in this list, so we know these platforms from the inside. Prototype builders and spec writers make the thing you test; they do not collect evidence, so they are in our broader guide to AI tools for product managers.

Hear from your own customers

The fastest evidence a PM can get comes from customers you can already email. Analytics shows that someone abandoned onboarding. A call explains why, but booking one with every customer takes weeks, and a short survey answer leaves you with more questions than you started with. These three tools get you the explanation without the calendar. Feedbaq lets customers show you the problem as well as tell you, and its AI follow-ups adapt to each answer, up to three on every question. Koji and Perspective AI hold an open interview that adapts as it goes. For what to ask, see our customer discovery questions.

1. Feedbaq

feedbaq.to

Made by us

Feedbaq gets you the depth of a customer call without the calendar. You share a study link, and people answer in their browser by voice, by narrating their screen while they use your product, or by text and choice. An AI follows up on what each person actually said, and your own AI agent can build the study, publish it and read every answer.

Feedbaq is built around the one test: a finding opens on the moment itself. Transcripts are timed word by word, so a click plays the recording from that word. You hear the hesitation, or watch what was on screen when they said it. Your agent reads the same timings, so every quote in its summary links to its second.

Your agent does more than read. It already knows your product and your earlier findings, so it can draft the study around the decision, set up the follow-ups, publish it and come back when the answers arrive. Feedbaq is in Claude’s connector directory and also connects to ChatGPT, Codex, Cursor and Grok; setup is on the MCP help page.

It is free until you collect. Connect Feedbaq to your AI agent and the agent becomes your expert researcher: it plans and builds studies with you, and you preview them, free. Publishing to collect responses costs $49 a month, for every study and every teammate, with no credits to ration.

More customers answer. They need no account or app, can type when they cannot talk, and can speak their own language.

You bring: The decision you need to make, the customers who can inform it, and your agent’s context on the product
You get: Recordings with transcripts timed word by word, follow-up answers, topic counts, and quotes your agent links to the moment they were said
Example: Ask trial users who stalled at setup to show you the step on screen and talk it through. When someone says “I wasn’t sure what it wanted”, the AI asks what they expected, and you watch the moment they hesitated.
Shortlist it if: You can reach the customers you want to hear from and need them to explain a problem in their own words, without booking calls.
Choose it over: Koji when you want people to show the problem on screen, not only describe it, and one plan price instead of credits; Listen Labs or Outset when you can invite your own customers and want to start today.
Think twice if: Your study depends on strangers who match a narrow profile: pair Feedbaq with a recruitment panel, or choose Listen Labs.

Price: Free in your AI agent; $49 a month to publish and collect responses, with every feature and unlimited studies and team members
Free plan: Yes: connect your AI agent, plan and build studies, and preview them, free

2. Koji

koji.so

Koji runs AI-moderated interviews by voice, chat or phone and groups the answers into themes. Every theme opens down to the verbatim quotes and the audio behind it.

You bring: A research goal and a link you send to your participants
You get: Adaptive interviews with transcripts, and themes you can drill into down to the exact quote
Example: Interview 30 recent sign-ups by voice about what they used before, then open the quotes behind the strongest theme.
Shortlist it if: You want each participant in an open, adaptive conversation rather than a fixed set of questions.
Choose it over: Perspective AI when you want voice interviews, MCP and API on one plan for €79 a month.
Think twice if: You want voice on the cheaper plan: Insights covers text studies, and voice interviews start on Interviews at €79.

Price: Insights €29 a month; Interviews €79 a month, with voice interviews, MCP and API; Enterprise from €500
Free plan: Yes: 10 credits, no card needed

3. Perspective AI

getperspective.ai

Perspective AI runs adaptive research conversations with screening and questions every participant must answer, and probes where an answer is thin. Its MCP server works with Claude Desktop, Claude Code and Cursor.

You bring: A research brief, screening rules and the questions every conversation must cover
You get: Conversation transcripts and an analysis you can question
Example: Screen for people who switched tools in the last six months, require “what made you leave?”, and let the AI probe each reason.
Shortlist it if: Every conversation must cover the same required questions and still follow the unexpected answer.
Choose it over: Koji when screening and required questions matter more than voice.
Think twice if: You run many conversations: each uses 10 credits, so Pro covers about 100 a month.

Price: Pro $99 a month with 1,000 credits; Enterprise on request
Free plan: 100 starter credits, about ten conversations
Watch the extra costs: Research as a Service, where Perspective designs and runs the study and recruits participants, is quoted separately.

Reach people you cannot reach yourself

Pay for recruitment when the people you need are not on your customer list. These four platforms find participants, run AI-moderated interviews and deliver the report. Listen Labs is the most complete for a PM team. Outset adds usability and concept tests at enterprise scale. Conveo and GetWhy lead with video, for studies across markets and for findings that must persuade a room. None publishes a platform price, so ask each for a quote on the same study.

4. Listen Labs

listenlabs.ai

Listen Labs designs the study, recruits from a network of over 50 million participants or from your own list, runs AI-moderated interviews and writes the report. Its customers include Microsoft, Anthropic and Sweetgreen. It announced that it is joining Salesforce, with the same team building the product.

You bring: A research question and the audience you need, or your own contact list
You get: Completed interviews and a report in which each claim points to the interviews behind it
Example: Recruit IT buyers at mid-size companies and learn why they turned down a concept after seeing it.
Shortlist it if: Finding the right participants is the hardest part of your study.
Choose it over: Feedbaq or Koji when you cannot reach the people yourself and want recruitment, interviews and the report from one vendor.
Think twice if: You want to start today with a card: it sells through a demo, with no published price.

Price: Not published; sold through a demo
Free plan: None published

5. Outset

outset.ai

An enterprise research platform for AI-moderated interviews, usability sessions and concept tests in text, voice or video. Its AI interviewer adapts to each answer, and the platform synthesizes across hundreds of interviews.

You bring: A discussion guide, the concepts or flows to show, and a target audience
You get: Interview recordings and transcripts with a synthesis across them
Example: Show three onboarding concepts to target users and compare how each group explains what it understood.
Shortlist it if: A research team needs interview depth at survey volume, study after study.
Choose it over: Listen Labs when usability and concept tests belong in the same platform as the interviews.
Think twice if: You run one study a quarter: pricing is custom, and recruitment and services are scoped separately.

Price: Enterprise contract
Free plan: None published
Watch the extra costs: Recruitment and research services are scoped with the plan.

6. Conveo

conveo.ai

Conveo runs AI-led video interviews with follow-up questions, in many languages. Its charts click through from a number to the verbatim quotes behind it, and it stitches tagged quotes into highlight reels.

You bring: A study guide and participants from your list, a link, a QR code, WhatsApp or a panel
You get: Video interviews, transcripts and charts that open on the quotes behind them
Example: Interview buyers in Germany and Spain about the same concept, then cut a reel of how each market describes it.
Shortlist it if: You need video interviews across languages and markets, and stakeholders who want to see faces.
Choose it over: Listen Labs when video and several languages are the core of the study.
Think twice if: You need a subscription price before you start: the platform is quoted per study.

Price: Not published
Free plan: None published
Watch the extra costs: Recruiting through its Respondent integration costs $6 a participant plus incentives, usually $1 an interview minute: about $21 for a 15-minute interview.

7. GetWhy

getwhy.io

GetWhy designs the study, recruits participants, runs AI-moderated video interviews and delivers a narrative report with the video evidence embedded, exported to PowerPoint in one click. Its customer stories include Unilever, eBay, Heineken and Maersk.

You bring: A business question and the audiences or markets to reach
You get: A report with the participant clips embedded next to each finding
Example: Test a new pricing message with buyers in three markets and play their reactions in the quarterly review.
Shortlist it if: The findings have to persuade people who only believe a customer they can see.
Choose it over: Conveo when you want the report and the presentation made for you, with research services on hand.
Think twice if: You want self-serve research on a small budget: there is no published price.

Price: Not published; self-service and full-service options
Free plan: None published

Watch people use the product

When the question is whether people can do something, watch them try. What people say about a flow and what they do in it often differ. The quickest way to watch is to ask your own customers to share their screen. In a Feedbaq study, they use your product on their own computer and talk you through it, and the AI asks about the moment they hesitate. Choose Maze or Lyssna when you need task success rates and click paths from a test plan. TheySaid puts an AI moderator in a user test for free. Maze pairs task data with AI interviews on Enterprise, Lyssna is the quick self-serve option for design tests, and Sprig asks inside your product at the moment that matters.

8. TheySaid

theysaid.io

TheySaid runs AI-moderated user tests and surveys in which participants talk with the AI by voice, in over 70 languages. Its reports link to the recordings, transcripts, quotes and highlight clips behind each insight.

You bring: A user test or survey, and people you invite or recruit from its panel
You get: Recordings, transcripts and clips behind each insight
Example: Ask five customers to try a new checkout while the AI asks what they expected at each step, then share the clip where two of them hesitated.
Shortlist it if: You want to try an AI moderator on a real study for free.
Choose it over: Maze or Lyssna when you want an AI moderator talking with participants on a free plan.
Think twice if: You want open AI interviews: they come with Enterprise, and Free runs one active test at a time.

Price: Enterprise, custom and billed annually; monthly billing costs 33% more
Free plan: Yes: 5 AI user tests and 50 AI surveys a month, one active test
Watch the extra costs: Panel recruiting costs extra.

9. Maze

maze.co

Maze combines prototype and usability testing with AI-moderated interviews and an AI study builder on Enterprise.

You bring: A prototype or live site, tasks and questions; for interviews, the goals you want to learn about
You get: Task results and each participant’s path through the prototype, clips, and on Enterprise, interview transcripts with themes
Example: Test a new onboarding flow for where people get stuck, then run AI-moderated interviews about what they expected instead.
Shortlist it if: You want task data and qualitative interviews in one research platform.
Choose it over: Lyssna when you need AI-moderated interviews next to prototype tests; Outset when prototype testing is the core job.
Think twice if: You are on a self-serve plan: the AI moderator and most AI features are Enterprise.

Price: On request
Free plan: Yes: 1 study a month and 5 seats, with prototype testing and pay-per-use panel recruitment
Watch the extra costs: Panel credits cost $5 each; one participant in an AI-moderated study uses five credits, or nine for extended B2B targeting.

10. Lyssna

lyssna.com

Lyssna is a self-serve tool for quick design tests, such as preference, five-second, prototype and live-site tests, plus surveys and interviews. Growth adds AI follow-up questions, summaries and an MCP server.

You bring: Two or more designs, a prototype or a live page, and a few questions
You get: Preference splits, first impressions and each participant’s reasons
Example: Show two pricing-page layouts for five seconds each and compare what people remember and why they prefer one.
Shortlist it if: You need fast reactions to designs or copy before the team commits.
Choose it over: Maze when you want a published self-serve price; Feedbaq when you are comparing designs and want a panel to recruit from.
Think twice if: You need AI follow-ups on a budget: they come with Growth, and panel responses cost extra.

Price: Growth $166 a month billed annually, or $199 monthly
Free plan: Yes: three seats and 15 self-recruited responses
Watch the extra costs: Responses from Lyssna’s panel are priced separately on every plan.

11. Sprig

sprig.com

Sprig is an enterprise research platform whose AI agents help design, field and analyze studies in your product, on the web and mobile, by email and by link. Its customer stories include Thumbtack and Notion.

You bring: A research question, the product moments or audiences to target, and the response volume you need
You get: Targeted in-product studies and AI analysis of the responses
Example: Ask users who just finished onboarding which of two concepts fits how they work, then read their reasons.
Shortlist it if: Your company runs recurring in-product research and wants one program behind it.
Choose it over: Outset when the question should reach users at a specific moment inside your product.
Think twice if: You have one occasional question: there is no public price or free plan.

Price: Enterprise contract, priced by response volume, capabilities and deployment
Free plan: None published

Keep what customers said on calls

Your best evidence may already be in calls you were on. These three capture it, and they differ in what you can go back to. Granola keeps a cited transcript and your notes, but not the audio. Otter keeps the audio with the transcript. tl;dv keeps the video and lets colleagues jump to the moment. If you will ever play a customer’s words to the team, pick one that keeps the recording.

12. Granola

granola.ai

Granola is a meeting notepad that listens to your computer’s audio, so no bot joins the call, and fills out the notes you type from the transcript. Ask a question across your meetings and the answer cites the meeting behind each point.

You bring: Any call on your computer, plus the rough notes you type
You get: Your notes filled out from the transcript, and answers with citations to the meetings and transcripts behind them
Example: Before a pricing review, ask which customers raised seat costs this month, then open each cited transcript to read the exchange.
Shortlist it if: You spend hours a week on calls and lose what was said before you write the spec.
Choose it over: Dovetail or Looppanel for everyday calls you never planned to analyze; bot recorders because nothing joins the call.
Think twice if: You want to hear the customer again: Granola keeps the transcript and deletes the audio after transcribing.

Price: Business $14 a user a month; Enterprise $35
Free plan: Yes: 30 days of meeting history

13. Otter

otter.ai

Otter transcribes live meetings and uploaded recordings, writes AI summaries, and keeps the audio with the transcript so you can replay any sentence.

You bring: Live calls, or recordings of interviews you import
You get: A searchable transcript tied to the audio, with summaries
Example: Import last month’s five churn calls and replay the exact sentence before you quote it in a brief.
Shortlist it if: You want a cheap, replayable record of every customer call.
Choose it over: Granola when you want to hear the customer again, not only read the notes.
Think twice if: You import a lot of past recordings: Basic allows three imports in total, and Pro ten a month.

Price: Pro $16.99 a user a month, or $8.33 billed annually; Business $30, or $19.99 billed annually
Free plan: Yes: 300 minutes a month, up to 30 minutes a conversation, three lifetime imports

14. tl;dv

tldv.io

tl;dv records and transcribes Zoom, Google Meet and Teams calls, writes AI notes, and turns the notes you type into timestamps that jump to that moment in the video.

You bring: Video calls, recorded with a bot or without one
You get: Video recordings with timestamped notes and transcripts you can search across meetings
Example: Search last quarter’s sales calls for “export” and jump to each moment a prospect asked for it.
Shortlist it if: You want video of customer calls that colleagues can open at the right moment.
Choose it over: Otter when you want video, not only audio, and moments you can share with a link.
Think twice if: You keep your research history on the free plan: AI notes and retention are limited there.

Price: Pro €29 a seat a month, or €18 billed annually; Business €39, or €29 billed annually
Free plan: Yes: unlimited recordings and transcripts

Find patterns across research and feedback

Analysis tools differ in their input: research you ran, or feedback that arrives on its own. Looppanel and Dovetail work on interviews and studies, Looppanel for one person and Dovetail for a team. Enterpret and Chattermill work on the daily stream from support, reviews and surveys. All four link a theme back to the material behind it; open it before the theme goes into a roadmap.

15. Looppanel

looppanel.com

Looppanel records and transcribes interviews, writes AI notes and groups the evidence by question or theme on an analysis board, with a video clip for each finding.

You bring: Recordings from a round of discovery or usability interviews
You get: Answers grouped by question, an insights summary and video clips for each finding
Example: After ten churn interviews, read every answer to “what did you switch to?” side by side and pull clips for the readout.
Shortlist it if: You run interview rounds yourself and want analysis without an enterprise contract.
Choose it over: Dovetail when one person needs a self-serve price; Granola when you need analysis by question across a study.
Think twice if: Several people need the repository: Solo has one editor, and Pro costs $395 a month with a five-editor minimum.

Price: Solo $49 a month, or $540 a year; Pro $395 a month, or $4,200 a year
Free plan: No free plan; Solo has a 15-day trial
Watch the extra costs: Solo includes 5 transcription hours and 5 uploads a month.

16. Dovetail

dovetail.com

Dovetail is a customer research repository. It turns calls, recordings, documents and survey answers into searchable evidence, with AI chat, summaries and agents that watch incoming feedback channels.

You bring: Interview recordings, support conversations, survey exports and documents
You get: Summaries, themes and answers grounded in the stored material, linked to the source
Example: Ask what onboarding problems came up across last quarter’s interviews and support tickets, then open the clips behind each theme.
Shortlist it if: Your team already runs research and needs one place where past evidence stays findable.
Choose it over: Looppanel when evidence must last across studies and be shared across a team.
Think twice if: You are buying for yourself: beyond the small free plan, Dovetail sells Enterprise contracts.

Price: Enterprise contract for the paid plan
Free plan: Yes: one channel and one project, with chat and summaries inside that project

17. Enterpret

enterpret.com

Enterpret connects feedback from support, sales calls, reviews and surveys to account and revenue data, and keeps it in a taxonomy that adapts as new themes appear. Its customer stories include Canva, Notion and ElevenLabs.

You bring: Support tickets, call transcripts, reviews and customer attributes such as plan and revenue
You get: Themes and cited answers about which problems affect high-value or at-risk accounts
Example: Ask which product problems come up most in tickets from accounts that renew this quarter.
Shortlist it if: Feedback arrives through many channels and you need to connect it to revenue or retention.
Choose it over: Dovetail when feedback arrives every day from many channels and you need it tied to accounts and revenue.
Think twice if: You work alone or at low volume: it is a company-level platform sold through a demo.

Price: Enterprise contract
Free plan: None published

18. Chattermill

chattermill.com

Chattermill brings in surveys, reviews, support conversations and calls through more than 65 integrations, finds the themes, and backs each one with counts and customer quotes. You can query it from Claude, ChatGPT or another agent through its MCP server. Its customers include Uber, HelloFresh and E.ON Next.

You bring: Survey answers, reviews, tickets and call transcripts from the tools you already use
You get: Themes with counts and the customer quotes behind them, in Chattermill or in your own agent
Example: Ask from Claude which complaints about the mobile app rose this month, with three customer quotes for each.
Shortlist it if: Feedback arrives in volume across many channels and someone reports on it every month.
Choose it over: Enterpret when reviews, surveys and support conversations matter more than account and revenue data.
Think twice if: You have one channel and low volume: it is a company platform sold through a demo.

Price: Not published
Free plan: None published

Product analytics

Analytics shows where people struggle; it cannot tell you why. Use these three to find the step and the cohort, watch a few replays, then ask those users. If your company already has one of them, use its AI before you switch.

19. PostHog

posthog.com

PostHog combines product analytics, session replay, surveys, feature flags and experiments with PostHog AI, which answers questions about your data in plain language.

You bring: A question in plain words, asked of your product’s events, replays and survey answers
You get: A chart or query built from your question, and the session replays of the people behind it
Example: Ask which signup source has the worst week-one retention, then open replays of that cohort before you write the brief.
Shortlist it if: You are choosing analytics for a newer product and want behavior, replays and surveys in one place.
Choose it over: Amplitude or Mixpanel when you also want replays, surveys, flags and experiments from one vendor with large free allowances.
Think twice if: Nobody owns instrumentation: the AI is only as good as the events engineers send.

Price: Usage-based beyond the free allowances, priced per product
Free plan: Every month: 1M events, 5,000 replays, 1,500 survey responses and 500 AI credits; unlimited team members

20. Amplitude

amplitude.com

Amplitude is a product analytics platform with replay, experiments, surveys, AI agents and an MCP server on every plan, plus AI Feedback, which analyzes customer comments next to behavior.

You bring: Tracked events and customer feedback sources
You get: Retention, funnel and cohort analysis, agent answers, and feedback themes you can read next to the behavior
Example: Ask its agent which first-week actions separate accounts still active at 90 days, then read the AI Feedback themes from the accounts that left.
Shortlist it if: Your company already runs on Amplitude, or measurement drives most product decisions.
Choose it over: PostHog when you want analytics-first depth with feedback analysis built in.
Think twice if: You need many experiments or saved charts on a small budget: Free limits both, and Growth is priced on request.

Price: Plus starts at $0 and is priced by event volume beyond 2M; Growth and Enterprise on request
Free plan: 2M events a month, 10,000 replays, 2,000 AI feedback records, unlimited seats

21. Mixpanel

mixpanel.com

Mixpanel is event analytics for funnels, retention and cohorts, with Mixpanel Agent and an MCP server for asking questions in plain language.

You bring: Consistently named product events and a question, asked in Mixpanel or from your own agent over MCP
You get: Funnels, retention curves and cohort comparisons built from the question, with replays of the sessions
Example: While drafting a spec in Claude, ask through Mixpanel’s MCP server how week-two retention changed after the new import flow.
Shortlist it if: Your team already tracks events in Mixpanel, or you want a focused analytics tool you can query from your agent.
Choose it over: Amplitude or PostHog when you want focused event analysis and no broader suite to adopt.
Think twice if: Nobody maintains a tracking plan: messy event names make every answer suspect.

Price: Growth: first 1M events a month free, then volume pricing; Enterprise on request
Free plan: Up to 1M events a month, 10,000 replays, unlimited seats

Question the evidence with your own agent

A general agent is only as trustworthy as what it can read. Pasted summaries give it nothing to cite. Connected tools give it the full record: with Feedbaq connected, Claude reads every transcript and can link each quote to its moment. Gemini Notebook takes the opposite approach and answers only from the files you load, with a citation for every answer.

22. Claude

claude.com

Anthropic’s assistant for long documents, research synthesis and code. Connectors from its directory, including Feedbaq, let it read your research tools directly, so it can quote from the full record instead of a summary you pasted.

You bring: Transcripts, notes and data you upload, or research tools you connect
You get: Themes, briefs and critiques that quote and link the sources it can read
Example: Connect Feedbaq and Mixpanel, then ask why trial users drop at step three, with participant counts and a link to each quote.
Shortlist it if: You want one agent that works across your research tools and writes the brief at the end.
Choose it over: ChatGPT is a close equal: pick the one your company supports and whose connectors cover your tools.
Think twice if: Your company has standardized on another assistant: data policy usually decides this one.

Price: Pro $20 a month, or $17 billed annually; Team $25 a seat monthly, $20 billed annually
Free plan: Yes, with connectors and up to five projects; Research and Claude Code need Pro

23. ChatGPT

chatgpt.com

OpenAI’s assistant for research, drafting, data analysis and files, with deep research, connected apps and agents.

You bring: Transcripts, spreadsheets and documents you upload, or apps you connect
You get: Analyses of exported data and interview files, drafts and sourced research reports
Example: Upload a funnel export and twelve interview transcripts, and ask which explanations the interviews support for the drop.
Shortlist it if: Your company already shares work in ChatGPT, or you analyze exported data often.
Choose it over: Claude when your company already licenses ChatGPT; for analysis the two are close, so the shared workspace decides.
Think twice if: You plan to stay on Free: uploads and deep research are limited there.

Price: Plus $20 a month; Business $25 a seat monthly or $20 billed annually, two-seat minimum
Free plan: Yes, with limited uploads and deep research

24. Gemini Notebook

notebook.google

Google’s research notebook, called NotebookLM until its rename, answers questions only from the sources you load into it and cites the passage behind each answer.

You bring: Interview transcripts, survey exports and documents you choose
You get: Answers and comparisons with a citation to the passage in your source
Example: Load eight interview transcripts and ask which of three explanations for churn they support, passage by passage.
Shortlist it if: Your research sits in scattered files and you want a free way to question one set of them.
Choose it over: Claude or ChatGPT when every answer must cite your own files and nothing else.
Think twice if: You need a lasting team archive: each notebook knows only what someone uploaded to it.

Price: Free; higher limits come with Google AI plans and qualifying Workspace or Cloud plans
Free plan: Yes; since 2 September 2026, limits depend on how much compute you use and refresh every five hours

The close calls

Feedbaq or Koji?

Choose by what you need to see. Feedbaq asks everyone the same questions and follows up on each answer, so you can compare people side by side, and it lets them show the problem on screen while they explain it. Your agent can build the study as well as read it, and it is free until you collect responses, then one plan covers every study, with no credits per conversation. Koji holds an open interview that goes wherever the participant takes it, and bills by credit. Pick Feedbaq when you want answers you can compare and watch, and Koji when each conversation should find its own way.

Your own customers, or a recruited panel?

Ask your own customers whenever they can answer the question. They know your product, they answer faster and they cost nothing to reach. Pay for Listen Labs, Outset, Conveo or GetWhy when you need people who have never heard of you, such as buyers in a new segment.

Feedbaq or Maze?

Pick Feedbaq to learn why, and Maze to measure how many succeed. Feedbaq shows you real customers in their own setup, explaining as they go. Maze measures a defined task across many testers, with success rates and paths.

Granola, Otter or tl;dv?

Choose by what you want to go back to. Granola is the lightest: better notes, cited answers, no bot in the call. Otter adds the audio, so you can hear the tone. tl;dv adds video and shareable moments, which help when a colleague needs to see the customer react.

Looppanel or Dovetail?

Looppanel is for one researcher; Dovetail is for a team’s memory. Looppanel compares answers question by question within a study at a self-serve price. Dovetail keeps evidence findable across studies and channels for everyone, and its paid plan is an enterprise contract.

A specialist, or your own agent?

Use both: a specialist to hold the evidence, your agent to question it. A specialist collects, records or organizes what customers said. Claude or ChatGPT can then work across several of them in one conversation, as long as each connection gives the agent the full record and not only summaries.

Two PM workflows

Each step uses a separate tool, and at each step you decide whether the evidence is enough to go on.

Investigate an activation drop

Find the step where new users stall in PostHog, Amplitude or Mixpanel, and watch five replays of it.
Ask your agent to draft a short Feedbaq study for the users who stalled, from what it knows about the product. Ask them to show the step on screen and say what they expected.
Ask Claude, with Feedbaq connected, for the reasons, with the number of different participants behind each and a link to every quote.
Open three of the links and listen. Write the spec with the customers’ words in it, and keep the questions the study could not answer.

Run a discovery round

Talk to eight customers about the last time the problem happened, and capture the calls in Granola, Otter or tl;dv.
Compare the answers question by question in Looppanel, or question the transcripts in Gemini Notebook.
Take the strongest pattern to thirty more customers in a Feedbaq study, so you learn how widely it holds without thirty more calls.
Bring the decision to the team with the clips and quotes attached, so they can judge the evidence themselves.

Free template

The customer discovery template

Ask about the last time the problem happened, what it cost and what people do about it today, before you build anything.

Use this template, free

1What is your role, and what are you responsible for?Text
2Tell us about the last time you [did the task]. What happened?Voice
3What was the hardest part of that?Voice
4What do you do about it today?Voice
5Show us how you handle it today.Screen share
and 2 more questions

Questions people ask

What are the best AI user research tools for product managers?

It depends on the evidence you are missing. For new answers from your own customers, start with Feedbaq, or Koji for open interviews. To reach people you cannot reach yourself, Listen Labs. To watch people use a product, Maze or Lyssna. For calls, Granola, Otter or tl;dv. For patterns across interviews, Looppanel or Dovetail, and across daily feedback, Enterpret or Chattermill. For product data, PostHog, Amplitude or Mixpanel.

Which AI research tools let you check the source behind a finding?

Feedbaq links each quote to the moment in the voice or screen recording, so you can hear it or watch it. Koji drills from a theme to the verbatim quote and audio, Conveo clicks through from a chart to the quotes, Looppanel attaches a video clip to each finding, tl;dv jumps to the timestamp in the video, and Gemini Notebook cites the passage in your file. Granola cites the transcript but does not keep the audio.

Which AI tools collect new answers rather than summarize existing ones?

Feedbaq, Koji and Perspective AI collect answers from customers you invite. Listen Labs, Outset, Conveo and GetWhy recruit participants and run AI-moderated interviews. TheySaid, Maze, Lyssna and Sprig run tests and surveys. Granola, Otter, tl;dv, Looppanel, Dovetail, Enterpret and Chattermill work on calls, research and feedback you already have.

Can Claude or ChatGPT analyze customer interviews?

Yes. Upload the transcripts, or connect the research tool so the agent reads the full record instead of a pasted summary. Feedbaq is in Claude's connector directory, and with it connected Claude can link each quote to its moment in the recording. Ask for the number of different participants behind each theme, their exact words, a link to each source and the answers that contradict it.

Which AI user research tools are free to start?

Feedbaq is free in your AI agent, which becomes your expert researcher and plans and builds studies with you; collecting responses costs $49 a month, with unlimited studies. TheySaid has 5 AI user tests a month, Maze 1 study a month and Lyssna 15 self-recruited responses. Koji starts with 10 credits and Perspective AI with 100. Granola, Otter and tl;dv have free plans for calls, Dovetail has one free project, and PostHog, Amplitude and Mixpanel have free event allowances.

Do I need a recruitment panel for user research?

Not when your own customers can answer the question. They know the product and cost nothing to reach. Pay for a panel, such as Listen Labs, Outset, Conveo or GetWhy, when you need people who have never used your product, such as buyers in a new segment.

How is this different from a list of AI tools for product managers?

This guide covers one PM job: getting customer evidence for a decision. Our broader guide to AI tools for product managers also covers writing specs, building prototypes and planning, which help you make the thing you test but do not collect evidence about it.

Keep reading

Need AI for specs, prototypes and planning too? Read the broader guide to AI tools for product managers.

The 33 best AI tools for product managers in 2026, by task

33 AI tools for product managers, sorted by task: call notes, research, specs, prototypes, analytics and planning. What each does and what it costs.

Read the guide

The 25 best customer feedback tools in 2026, by the job you need done

25 customer feedback tools grouped by job: in-app surveys, rich answers, feedback analysis, request management, behavior and bug reports. Prices checked.

Read the guide

Sources

Nielsen Norman Group, AI-powered tools for UX research: issues and limitations Feifei Liu and Kate Moran, 2 July 2023.
Vikan et al., Reflecting on LLM support in reflexive thematic analysis, Qualitative Health Research 2026.
Feedbaq
Feedbaq, pricing Checked 6 October 2026.
Feedbaq, MCP setup and tools
Feedbaq in Claude
Koji, pricing Checked 6 October 2026.
Koji, product
Perspective AI, pricing Checked 6 October 2026.
Perspective AI, MCP documentation
Listen Labs, product and customers Checked 6 October 2026.
Listen Labs, joining Salesforce
Outset, pricing Checked 1 October 2026.
Outset, platform
Conveo, product and recruitment costs Checked 6 October 2026.
GetWhy, product and customer stories Checked 6 October 2026.
TheySaid, pricing Checked 6 October 2026.
TheySaid, features
Maze, pricing Checked 1 October 2026.
Maze help center, panel credit costs
Lyssna, pricing Checked 1 October 2026.
Sprig, pricing Checked 1 October 2026.
Granola, pricing Checked 1 October 2026.
Granola, meeting recall and transcript citations
Otter, pricing Checked 6 October 2026.
tl;dv, pricing Checked 6 October 2026.
tl;dv, timestamped meeting moments
tl;dv, plan limits
Looppanel, pricing Checked 6 October 2026.
Dovetail, pricing Checked 1 October 2026.
Enterpret, ElevenLabs customer story
Chattermill, product, MCP and customers Checked 6 October 2026.
PostHog, pricing Checked 1 October 2026.
Amplitude, pricing Checked 1 October 2026.
Mixpanel, pricing Checked 1 October 2026.
Claude, pricing Checked 1 October 2026.
OpenAI, ChatGPT Business pricing Checked 1 October 2026.
Google, Gemini Notebook Checked 1 October 2026.
Google, new flexible usage limits in Gemini Notebook September 2026.