The 19 best feedback and survey tools for ChatGPT and Claude in 2026

Peter Kildegaard · 7 October 2026

Every feedback tool now says it works with ChatGPT and Claude. What matters is the next step: can your agent build a study, publish it and read every answer, or can it only read what you already collected? And when it reads, does it see the customer’s full words, or only a summary it cannot quote?

This guide is for founders, PMs, designers and marketers who want their AI agent to do the research. It shows, for 19 tools checked on 7 October 2026, how each connects to Claude and ChatGPT, what your agent can do with it, what it reads and what it costs, and ends with a one-week test for any connection.

Key takeaways

Get an expert researcher in your AI agent, free. Connect Feedbaq from Claude’s connector directory (or ChatGPT) and your agent plans and builds a study from what it knows about your product. Customers answer by voice, by showing their screen, or by text, and the AI follows up on what each person said. Collecting responses costs $49 a month, and the agent reads every answer with a link to the moment.
Choose a connection by what your agent can do next. Eight of these 19 tools work only on evidence you already have. Eleven let the agent build something, and only some of those let it publish the study and read every answer back.
Make the agent read full answers, not a summary. An agent that only sees a summary has nothing to quote, and you have nothing to check.
A directory listing connects in one click. A custom connector or ChatGPT’s developer mode takes a server address, and in a company workspace often an admin.

Start here

Find the line that describes your next job. Each names the tool we would connect first and the one to compare it with.

You need customers to explain a problem, and want your agent to do the work. Start with Feedbaq: it is free in your agent, which builds the study and reads every answer, and you pay only when you collect responses. Compare Perspective AI for open, adaptive interviews in Claude.
You need a form your agent can build and read. Start with Tally, free in both directories. Compare Jotform for operational forms.
Your company already runs surveys. Connect SurveyMonkey or Typeform where they live. Compare Survicate for surveys inside your product.
You need a quick vote from strangers. Use PickFu, and approve the cost before your agent launches.
Your research already lives in a research tool. Connect Maze or Dovetail. Compare Lyssna for design tests.
Feedback pours in from support, reviews and surveys. Start with Enterpret if you need it tied to revenue. Compare Chattermill.
The feedback is in your calls. Connect Granola.

The 19 tools and what your agent can do with each

The columns that matter most are “Builds and runs” and “Reads”: whether your agent can run the study, and what it reads when the answers come in. The Claude and ChatGPT columns tell you how much setup stands between you and that first step. “Directory” means you find the tool in the assistant’s own list and click connect. “Custom connector” and “Developer mode” mean you paste the vendor’s server address.

Let your agent run the study: Feedbaq, Perspective AI, Speak AI, PickFu
Let your agent build forms and surveys: Tally, Jotform, Typeform, SurveyMonkey, Survicate, Formbricks, Weavely
Let your agent read research in full: Feedbaq, Maze, Lyssna, Dovetail, Granola
Let your agent read feedback that keeps arriving: Enterpret, Chattermill, Zonka Feedback, Canny

Feedback and survey tools for ChatGPT and Claude compared by connection route, agent actions and price, checked 7 October 2026
Tool	Claude	ChatGPT	Builds and runs	Reads	Price from
Feedbaq	Directory	Developer mode	Builds and publishes studies	Every answer in full, linked to the moment	Free; $49/mo to collect
Perspective AI	Custom connector	Not documented	Builds and publishes interviews	Full transcripts	$99/mo
Speak AI	Custom connector	Developer mode	Creates recorders and audio surveys	Full transcripts	$20/user/mo, annual
PickFu	Custom connector	Developer mode	Builds and launches paid polls	Every written reason	$1/response
Tally	Directory	Plugin directory	Builds and edits forms	Every submission	Free
Jotform	Directory	Plugin directory	Builds and edits forms	Every submission	Free
Typeform	Directory	Plugin directory	Builds and edits forms	Responses	Free
SurveyMonkey	Directory	Plugin directory	Builds surveys and share links	Responses	Free
Survicate	Custom connector	Developer mode	Drafts; you launch	Answers, with citations	$114/mo, annual
Formbricks	Custom connector	Not documented	Builds link surveys	Feedback records, on Enterprise	Free
Weavely	Custom connector	Plugin directory	Builds and publishes forms	No responses	Free
Maze	Directory	Plugin directory	No	Raw transcripts and quotes	Free; paid on request
Lyssna	Custom connector	Developer mode	No	Study responses	$166/mo, annual
Dovetail	Directory	Plugin directory	Writes docs and highlights	Transcripts and highlights	Free; Enterprise contract
Granola	Directory	Plugin directory	No	Notes; transcripts on paid plans	Free; $14/user/mo
Enterpret	Directory	Developer mode	No	Themes, linked to source records	Enterprise contract
Chattermill	Custom connector	Developer mode	No	Themes with customer quotes	Not published
Zonka Feedback	Directory	Developer mode	No	Scores, themes and verbatims	Not published
Canny	Custom connector	Plugin directory	Creates ideas	Ideas and their sources	$79/mo, annual

Choose by what your agent can do next

A connection is a list of actions. The question is whether the action you need next is on the list. Every tool here says it works with ChatGPT and Claude, and every one of them does. But eight of the 19 work only on evidence you already have: they search past studies, notes or feedback, and none of them can ask a customer anything new. Of the eleven that build, Survicate drafts surveys and leaves the launch to you, and Weavely publishes forms but cannot read what comes back.

That matters because most questions start with evidence you do not have yet. If your agent can only read, it can tell you what customers said last quarter, and then you leave the chat to build the study that answers today’s question. If it can build, publish and read, it carries the question from the decision to the answer, with the context it already has about your product.

The second test is what the agent reads: the full answers, or a summary of them. An agent working from a summary has nothing to quote, so its brief is a paraphrase of a paraphrase. 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. In a 2026 study in Qualitative Health Research, a language model sometimes returned quotes that nobody in the interviews had said. Vendors are moving toward the source: since May 2026, every number Enterpret’s server returns links to the records behind it.

Feedbaq is built for both tests, and it is free in your agent. Your agent becomes your expert researcher: it plans the study, sets up adaptive follow-up questions and builds it, and you preview it as often as you like. Publishing it to collect responses costs $49 a month. When answers arrive, the agent reads each one in full, with the transcript timed word by word and a link to the moment in the recording, so every quote in its brief opens on the second it was said.

And there is more to read. In a 2024 experiment by Lenzner, Höhne and Gavras, spoken answers averaged 43.8 words and typed answers 14.2. A customer who talks, or shows you the problem on screen, gives your agent three times as much to quote as one who fills in a box. Perspective AI comes closest, but without screen recordings, and its documentation covers Claude only.

How we chose

This guide is desk research, not a hands-on test. On 7 October 2026 we read each vendor’s own MCP or integration page for the Claude and ChatGPT route and the actions its server offers, and each vendor’s pricing page. The sources are linked at the end.

A documented connection. Each tool connects to Claude, ChatGPT or both through its own MCP server, documented by the vendor.
Feedback or survey work. Each tool collects, stores or analyzes what customers say.
A price or a named customer. Every tool publishes a price or a free plan, or names customers in its own stories.
A clear answer to two questions. For every tool we say whether the agent can build and run the study, and what it reads.

We make Feedbaq and recommend it first for new answers from your own customers. Its founder worked as a staff product manager at Maze, one of the tools in this list. Where a vendor documents a connection by server address only, we list it under the custom connector in Claude and developer mode in ChatGPT, which is how both assistants take an address.

Let your agent run the study

The most useful connection lets your agent ask customers something new and read what they said. These four carry a study from the draft to the answers. Feedbaq runs studies by voice, screen and text, with adaptive follow-ups about what each person said, in Claude and ChatGPT. Perspective AI runs open, adaptive interviews by voice or text from Claude. Speak AI collects spoken answers through recorders, and PickFu buys a quick vote from a consumer panel. For how to write the follow-ups, see our guide to AI follow-up questions.

1. Feedbaq

feedbaq.to

Made by us

Feedbaq gets you the depth of a customer call without the calendar. People answer a study link in their browser by voice, by narrating their screen, or by text and choice, and adaptive AI follow-ups ask about what each person said. It is free in your own AI agent, which becomes your expert researcher: it plans and builds the study, and collecting responses costs $49 a month. Feedbaq is in Claude’s connector directory and connects through ChatGPT’s developer mode.

Your agent becomes your expert researcher, for free. Connect Feedbaq and your agent plans the study from your product context and the decision you are making, and sets up to three adaptive follow-up questions per question with your guidance. Building and previewing are free for as long as you like; publishing to collect responses is $49 a month. Then the agent reads every answer when it arrives. Each answer comes back with its transcript timed word by word and a link to the moment, so the agent can quote a customer and you can play the quote.

It is one click away in Claude. Feedbaq is listed in Claude’s connector directory. In ChatGPT it connects through ChatGPT’s developer mode, and it also works with Codex, Cursor and Grok; setup is on the MCP help page.

Customers answer by link, with no account. They talk, show their screen in a desktop browser, or type; voice questions let them type instead. You send the link to the customers you can reach, and the agent reads what comes back.

You bring: The decision you need to make, the customers you can reach, 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 Claude why trial users stall at setup. It drafts a short study with a screen-recording task from what it knows about the product, publishes it when you approve, and two days later returns the reasons with a link to each moment.
Shortlist it if: You want your agent to run new research with customers you can reach, not only read what you already have.
Choose it over: Tally, Jotform or Typeform when typed answers will not explain the problem; Maze, Lyssna or Dovetail when you need new answers, not a search of past studies; Perspective AI when you work in ChatGPT as well as Claude, or want people to show you their screen.
Think twice if: Your study depends on strangers who match a narrow profile: pair Feedbaq with a recruiting panel, or use PickFu for a quick consumer vote.

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. Perspective AI

getperspective.ai

Perspective AI runs adaptive research conversations, by voice or text, with screening and required questions, and probes where an answer is thin. Its MCP server lets Claude design a study, publish it and read the full transcripts. The documentation covers claude.ai, Claude Desktop, Claude Code, Cursor and VS Code, not ChatGPT.

You bring: A research brief, screening rules and the questions every conversation must cover
You get: Full conversation transcripts, counts and theme search across conversations, all readable by your agent
Example: In Claude, ask for a study of people who switched tools in the last six months, require “what made you leave?”, and come back to the transcripts once ten conversations are in.
Shortlist it if: You work in Claude and want open, adaptive interviews that your agent can design and read.
Choose it over: Feedbaq when each conversation should go wherever the participant takes it, with screening rules in front.
Think twice if: Your team works in ChatGPT: the MCP documentation covers Claude clients, Cursor and VS Code only.

Price: Pro $99 a month with 1,000 credits; Enterprise on request
Free plan: 100 starter credits, about ten conversations
Watch the extra costs: Each conversation uses 10 credits, so Pro covers about 100 a month.

3. Speak AI

speakai.co

Speak AI records, transcribes and analyzes audio and video, and collects spoken answers through recorders and audio surveys you embed. Its MCP server lets Claude or ChatGPT create recorders and surveys, upload media and read full transcripts.

You bring: Recordings you upload, or spoken answers collected through a recorder on your site
You get: Transcripts, AI insights and clips your agent can search and quote
Example: Put a recorder on your cancellation page, then ask ChatGPT to read last month’s transcripts and group the reasons people gave.
Shortlist it if: Your audio comes from many places, such as interviews, webinars and a recorder on your site, and you want one transcript library your agent can read.
Choose it over: Feedbaq when you need a media library and transcription by the hour more than a structured study.
Think twice if: You want one predictable bill: transcription and AI work draw on credits.

Price: Pro $25 a user a month, or $20 billed annually; pay as you go at $1 a credit
Free plan: 7-day trial, no card needed
Watch the extra costs: Standard transcription uses 2 credits an hour; Pro includes 25 credits a user each month.

4. PickFu

pickfu.com

PickFu runs paid polls: you post two or more options, a targeted panel picks one and writes why, and AI summarizes the reasons. Its free MCP server lets your agent build a poll, launch it to the panel and read every written reason.

You bring: Images, names, headlines or videos to compare, and the audience traits you want
You get: Votes, a written reason from every respondent and an AI summary of the reasons
Example: Ask Claude to compare two landing-page headlines with US small-business owners, approve the cost, and read why people rejected the losing one.
Shortlist it if: You need a quick read from strangers on a headline, icon or name.
Choose it over: Feedbaq when you need people you cannot reach yourself, within hours.
Think twice if: Your buyers hold specific B2B roles: a consumer panel will not represent them.

Price: From $1 a response; surveys from $15. PickFu+ from $999 a year, paid as credit
Free plan: Free account and free MCP; you pay for each poll you launch
Watch the extra costs: Targeting, question type and sample size raise the price per response.

Let your agent build forms and surveys

A form your agent can build and read is the right tool when typed answers settle the question. Tally is the free default, listed in both directories. Jotform suits forms that feed operations, Typeform suits forms that face customers, and SurveyMonkey suits a company that already runs its surveys there. Survicate reaches users inside your product, Formbricks runs on your own servers, and Weavely lets you try the form inside ChatGPT. None of them asks about what a person just wrote. Our Tally comparison covers when a form is enough.

5. Tally

tally.so

Tally builds forms the way you write a document, with logic, uploads and payments. It is in Claude’s connector directory and ChatGPT’s plugin directory, and its free connection lets your agent create and edit forms and read every submission.

You bring: A description of the form, or a Tally form you already run
You get: A shareable form, and submissions your agent can read, count and summarize
Example: Ask Claude to build a beta waitlist form with a role question that branches, then a week later ask which roles signed up and what they wrote.
Shortlist it if: You want a free form your agent can build and read, and typed answers are enough.
Choose it over: Jotform or Typeform when you want the connection and unlimited submissions at no cost.
Think twice if: You need the reasons behind short typed answers: a form cannot ask about what someone just said.

Price: Pro $29 a month or $290 a year; Business $89 a month or $890 a year
Free plan: Yes: unlimited forms and submissions, subject to fair use; the connection is free on every plan

6. Jotform

jotform.com

Jotform is a form platform with a free MCP connection, listed in Claude’s connector directory and ChatGPT’s plugin directory. Your agent can list, create and edit forms, add test submissions, and read and filter real ones.

You bring: A description of the form, or the forms you already run in Jotform
You get: Forms and filtered submissions in the conversation
Example: Ask ChatGPT for this week’s pending event registrations from companies over 50 people, and draft a reminder for each.
Shortlist it if: Your forms feed operations, such as registrations, applications or orders, and you want your agent to work through the submissions.
Choose it over: Tally when you need HIPAA features, which come with Gold.
Think twice if: You collect a lot on the free plan: Starter allows 5 forms and 100 submissions a month.

Price: Bronze $34 a month billed annually or $39 monthly; Silver $39 or $49; Gold $99 or $129
Free plan: Yes: 5 forms and 100 submissions a month; the connection costs nothing extra

7. Typeform

typeform.com

Typeform builds polished forms that show one question at a time. Its MCP server, free on every plan, is in Claude’s connector directory and ChatGPT’s plugin directory, and lets your agent build and edit forms, analyze responses, manage contacts and set up automations.

You bring: Your questions, branching logic and brand styling, described in the chat
You get: A branded form, and responses your agent can analyze and act on
Example: Ask Claude to flag this week’s negative answers to your onboarding survey and draft a reply to each.
Shortlist it if: The form faces customers and its design reflects on your brand.
Choose it over: Tally or Jotform when presentation and Typeform’s automations matter more than price.
Think twice if: You want video answers or Clarify with AI, its follow-up question: both come with Talent, at $119 a month billed annually.

Price: Basic $28 a month billed annually or $39 monthly, 100 responses a month; Plus $56 or $79
Free plan: Yes, with a limited number of responses a month

8. SurveyMonkey

surveymonkey.com

SurveyMonkey is a survey platform whose connector is in Claude’s connector directory and ChatGPT’s plugin directory. Your agent can find your surveys, create and edit one, make a shareable web link and pull in the responses for analysis.

You bring: A survey goal in plain words, or the surveys you already run
You get: A live survey link, and responses your agent analyzes in the chat
Example: Ask ChatGPT for a five-question pulse survey after a feature launch, take the link from the same reply, and come back for the trends when responses arrive.
Shortlist it if: Your company already runs its surveys in SurveyMonkey.
Choose it over: Tally or Jotform when your survey history and panel already live in SurveyMonkey.
Think twice if: Your account stores data outside the US: the connector serves accounts in SurveyMonkey’s US data center, and it cannot edit a live survey that already has responses.

Price: Standard €39 a month billed monthly; Advantage €36 a month billed annually; Team Advantage €30 a user a month, three users minimum
Free plan: Basic: free, 10 questions a survey and limited responses; the connector works from Basic up
Watch the extra costs: €0.10 for each response over your plan’s limit; SurveyMonkey Audience responses are sold separately.

9. Survicate

survicate.com

Survicate sends targeted surveys in your product, on your site, by email and by link, and keeps the answers in a Research Hub. Its MCP server, on every plan, lets Claude or ChatGPT draft surveys and query answers with citations. It never launches a survey.

You bring: Questions, a channel and targeting rules, such as users who activated this week
You get: Draft surveys for you to review, and answers with citations and respondent attributes
Example: Ask Claude what detractors said about onboarding this month, with the comments behind each point, then have it draft a follow-up survey for you to launch.
Shortlist it if: You run recurring surveys across product, site and email, and want your agent to read the answers.
Choose it over: Typeform or SurveyMonkey when you need to reach users at a moment inside your product.
Think twice if: You want AI follow-up questions on a budget: they start on Pro, at $349 a month.

Price: Growth from $114 a month billed annually: 250 responses and 1,000 data points a month, 10 seats
Free plan: No free plan: a 10-day trial with up to 25 responses
Watch the extra costs: AI follow-up questions start on Pro, from $349 a month.

10. Formbricks

formbricks.com

Formbricks is an open-source survey platform for link, website and in-app surveys, in its cloud or on your own servers. Its MCP server connects to the Claude apps as a custom connector, and to Claude Code and Codex, to list, create, update and delete link surveys.

You bring: A survey described in the chat; for in-product surveys, its SDK in your app
You get: Link surveys in your workspace; on Enterprise, feedback records your agent can read and add
Example: Ask Claude Code to create a link survey for new signups while you ship the feature it asks about.
Shortlist it if: You want survey data on your own servers and an agent that works from your coding tools.
Choose it over: Survicate or Typeform when open source or self-hosting matters.
Think twice if: Your team works in ChatGPT: Formbricks documents Claude and Codex, not ChatGPT.

Price: Pro $74 a month for 2,000 responses; Scale $325 a month for 5,000; two months free on annual billing
Free plan: Yes: Hobby, 250 responses a month, one workspace
Watch the extra costs: The MCP tools for feedback records need Enterprise Edition.

11. Weavely

weavely.ai

Weavely is a free AI form builder. In ChatGPT it is a listed plugin that shows a working form you can fill in inside the chat; in Claude you add it as a custom connector. Your agent can build, style and publish forms, and the responses arrive in Weavely.

You bring: A description of the form, typed, spoken or uploaded as a document
You get: A working preview in the chat, then a published form whose responses collect in Weavely
Example: Describe a workshop sign-up form in ChatGPT, submit the preview yourself to try the logic, then publish it.
Shortlist it if: You want to see and try a form in the conversation before it goes live.
Choose it over: Tally when trying the form inside ChatGPT before you publish matters most.
Think twice if: You want your agent to analyze the responses: they stay in Weavely, out of the chat.

Price: Free; Pro adds custom domains
Free plan: Yes: unlimited forms and responses

Let your agent read research in full

The fullest evidence an agent can read is a Feedbaq study. With Feedbaq connected, your agent reads every response in full: transcripts timed word by word, topic labels, and a link from each quote to the moment in the recording.

For research you already ran elsewhere, connect the tool where it lives. Maze returns raw transcripts and quotes from your studies, and Lyssna gives read-only access to your tests. Dovetail reads and writes across a shared repository, and Granola brings in your calls. None of these four creates a study, so use them to learn what you already know before you ask anyone again.

12. Maze

maze.co

Maze is a product research platform for prototype tests, surveys and AI-moderated interviews. Its MCP server is listed in Claude’s and ChatGPT’s directories, works in Gemini, Copilot and Cursor, and lets your agent search your studies and read raw transcripts, responses and quotes. It does not create studies.

You bring: Studies your team already ran in Maze
You get: Transcripts, participant quotes and findings across studies, in your agent
Example: Before planning a new onboarding test, ask Claude what earlier Maze studies found about setup, with the quotes behind each finding.
Shortlist it if: Your team already runs research in Maze and wants it in the agent where decisions get written.
Choose it over: Lyssna or Dovetail when your studies already live in Maze.
Think twice if: You want your agent to set up the next study: the connection reads past research, and studies are built in Maze.

Price: Paid plans on request
Free plan: Yes: 1 study a month, 5 seats
Watch the extra costs: Panel credits cost $5 each; one participant in an AI-moderated study uses five credits.

13. Lyssna

lyssna.com

Lyssna runs preference, five-second, prototype and live-site tests, surveys and interviews. Its MCP server, in beta on paid plans, connects by URL to Claude, ChatGPT, Copilot, Gemini and Cursor, with read-only access to the studies you already ran.

You bring: Studies you ran in Lyssna
You get: Answers grounded in participant responses across studies
Example: Ask ChatGPT what participants said about onboarding across your last four studies, before you run a fifth.
Shortlist it if: You test designs and messages in Lyssna and keep asking what earlier studies found.
Choose it over: Maze when you want a published self-serve price.
Think twice if: You are on the free plan: the MCP server comes with paid plans, from Growth.

Price: Growth $166 a month billed annually, or $199 monthly
Free plan: Yes: 3 seats and 15 self-recruited responses, without the MCP server
Watch the extra costs: Responses from Lyssna’s panel are priced separately on every plan.

14. Dovetail

dovetail.com

Dovetail is a customer research repository. It is in Claude’s connector directory and ChatGPT’s plugin directory, and your agent can search the workspace, read transcripts, highlights and docs, and create projects, docs and highlights.

You bring: Interviews, recordings, documents and feedback channels already in Dovetail
You get: Verbatim customer quotes and docs in the conversation, and new notes written back to Dovetail
Example: Ask Claude which interviews describe a workaround for reporting, read the quotes, and save a doc with the findings to the project.
Shortlist it if: Your team keeps its research in one shared place and wants the agent to read and write there.
Choose it over: Maze or Lyssna when evidence comes from calls, documents and feedback channels, not only tests.
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 project and one channel

15. Granola

granola.ai

Granola is a meeting notepad that captures your computer’s audio, so no bot joins the call. It is in Claude’s connector directory and ChatGPT’s plugin directory, and your agent can search your notes and, on paid plans, full transcripts and team notes.

You bring: Customer calls on your computer, plus the notes you type
You get: Notes, decisions and quotes from your meetings, in your agent
Example: Ask Claude which objections came up in this month’s discovery calls, and open the notes behind each one.
Shortlist it if: Much of your customer feedback arrives in calls you are already on.
Choose it over: Dovetail when the feedback is in calls you had this week, not in planned research.
Think twice if: You stay on Basic: the connection reaches your last 30 days of notes, and transcripts need a paid plan.

Price: Business $14 a user a month; Enterprise $35
Free plan: Yes: notes from the last 30 days

Let your agent read feedback that keeps arriving

When feedback arrives on its own every day, connect the tool that already sorts it. Enterpret ties feedback to accounts and revenue, and links every number it returns to the records behind it. Chattermill handles consumer feedback at volume. Zonka Feedback reads your own NPS, CSAT and CES programs, and Canny holds feature requests with the accounts that asked. All four read; to ask the customers behind a theme why, run a study. Our guide to customer feedback tools covers this layer in more depth.

16. Enterpret

enterpret.com

Enterpret connects feedback from support, sales calls, reviews and surveys to account and revenue data. Its MCP server is in Claude’s connector directory and connects to ChatGPT by URL. Since May 2026, every number it returns links to the records behind it. Its customer stories include Canva, Notion and ElevenLabs.

You bring: Support tickets, call transcripts, reviews and account attributes, already in Enterpret
You get: Themes, sentiment and verbatims, with a link from each number to the source records
Example: Ask Claude which onboarding complaints come from accounts renewing this quarter, then open the records behind the count.
Shortlist it if: Feedback arrives through many channels and you need to tie it to revenue.
Choose it over: Chattermill when B2B account and revenue context matters most.
Think twice if: You have one survey and a few hundred comments: it is a company platform sold through a demo.

Price: Enterprise contract
Free plan: None published

17. Chattermill

chattermill.com

Chattermill analyzes surveys, reviews, tickets and calls from more than 65 integrations, and backs each theme with counts and customer quotes. You add it to Claude as a custom connector and to ChatGPT in developer mode. Its customers include Uber, HelloFresh and E.ON Next.

You bring: Feedback streams already flowing into Chattermill
You get: Themes with counts and the customer quotes behind them
Example: Ask ChatGPT which app complaints rose this month, with three customer quotes for each.
Shortlist it if: A CX or product team handles a high volume of consumer feedback and reports on it every month.
Choose it over: Enterpret when reviews and support conversations matter more than account revenue.
Think twice if: You also need to collect the feedback: it analyzes, so pair it with a survey tool.

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

18. Zonka Feedback

zonkafeedback.com

Zonka Feedback runs NPS, CSAT and CES surveys across channels and analyzes the comments with AI. Its MCP server, on every plan, is in Claude’s connector directory and connects to ChatGPT by URL, and reads scores, themes, sentiment and verbatim comments. Its customer stories include Leroy Merlin and Larsen & Toubro.

You bring: Survey scores, comments and reviews already in Zonka
You get: Score breakdowns, themes and the verbatim comments behind them
Example: Ask Claude which comments explain this month’s CSAT drop at one location, with the verbatims.
Shortlist it if: You run a CX survey program across locations or channels and want to question it from Claude.
Choose it over: Chattermill when the feedback starts as your own NPS, CSAT and CES surveys.
Think twice if: You want your agent to create surveys: the connection only reads, and Zonka lists write actions as coming soon.

Price: Not published; quoted by response volume and data credits
Free plan: No free plan: a 14-day trial
Watch the extra costs: AI Feedback Intelligence is an add-on, priced by data credits.

19. Canny

canny.io

Canny collects feature requests from a portal, support tickets and sales calls, and links each to the accounts that asked. Its MCP server is in ChatGPT’s plugin directory and connects to Claude as a custom connector. Your agent can read and create ideas, insights and companies, with revenue context.

You bring: Requests from your portal, support tickets and sales calls, already in Canny
You get: Ideas with the accounts, revenue and conversations behind them
Example: Ask ChatGPT which accounts asked for audit logs this quarter, ranked by revenue, and draft the idea for review.
Shortlist it if: A B2B team hears many requests through sales and support and decides by who asked.
Choose it over: Enterpret when your feedback is mostly feature requests.
Think twice if: You are on the free plan: the MCP server needs Pro or Business, Canny Ideas turned on, and an Owner to connect first.

Price: Pro from $79 a month billed annually, for 100+ tracked users; Business on request
Free plan: Yes: 25 tracked users, without the MCP server

The close calls

Feedbaq or a form your agent can fill in?

Choose by whether a typed answer will explain the problem. Tally, Jotform and Typeform give your agent a form to build and submissions to read, and Tally does it for free, but a typed answer averages about a third of the words of a spoken one. A form is enough for a sign-up, a waitlist or a rating. Feedbaq gives your agent a study: customers talk or show their screen, the AI follows up on what they said, and the agent reads every answer with a link to the moment. Pick a form when you know what to ask and short answers will do. Pick Feedbaq when you need to hear why.

Feedbaq, or Maze and Dovetail?

Feedbaq runs new research and lets your agent read it in full; Maze and Dovetail let your agent read research you already have. Maze returns raw transcripts and quotes from past studies, and Dovetail searches a shared repository; neither connection creates a study. Feedbaq builds the study, publishes it and reads every answer with its moment. If your team already runs research in Maze or Dovetail, connect both: ask what you already know, then let your agent run a Feedbaq study on the gap.

Is Weavely’s preview in ChatGPT worth it?

For drafting a form, yes; for learning from it, no. Weavely shows a working form in the chat, so you can submit it yourself and test the logic before you publish. Respondents still answer on a published link, and their responses stay in Weavely, where your agent cannot read them. If you want the agent to analyze what comes back, build the form in Tally instead.

Feedbaq or Perspective AI?

Choose by where you work and what you need to see. Perspective AI holds open voice or text interviews that adapt as they go, bills by credit, and documents Claude, Cursor and VS Code. Feedbaq asks everyone the same main questions and adapts its follow-ups to each answer, so you can compare people side by side, lets them show the problem on screen, and connects to Claude and ChatGPT. Perspective AI suits open conversations in Claude, but Feedbaq is the pick when you want customers to show you the problem on screen, every quote to open on its moment, and a start that is free in your agent.

A directory listing or a custom connector?

Prefer the directory when a team will use the connection. A listing takes one click and a sign-in, which anyone on the team can do. A custom connector or developer mode works just as well once it is set up, but it starts with a server address, and in a company workspace an admin often has to allow it first. For a tool only you will use, the setup is a few minutes either way.

Test a connection with one small study

Run this before you trust a connection with a real decision. It takes a week and five customers.

Pick one decision, such as whether to simplify the first setup step, and five customers you can email today.
Connect from the directory where you can. Ask your agent to draft the study from what it knows about the product, and read the preview yourself before it publishes.
Answer the study yourself first, then send the link to the five customers.
When the answers are in, ask: “For each reason, give the number of different people, their exact words and a link to each answer.” Open three of the links. If the agent cannot give you a quote and a link, it is reading a summary, and the connection fails the test.
Ask one question the study cannot answer. A good agent says so instead of inventing a motive.

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

Which feedback and survey tools work inside ChatGPT and Claude?

In Claude’s connector directory: Feedbaq, Tally, Jotform, Typeform, SurveyMonkey, Maze, Dovetail, Granola, Enterpret and Zonka Feedback. In ChatGPT’s plugin directory: Tally, Jotform, Typeform, SurveyMonkey, Weavely, Maze, Dovetail, Granola and Canny. Feedbaq, Survicate, Speak AI, PickFu, Lyssna, Chattermill, Enterpret and Zonka connect to ChatGPT in developer mode with a server address. Perspective AI and Formbricks document Claude but not ChatGPT.

Can my AI agent build and run a survey or study for me?

Yes, with some tools. Feedbaq lets your agent plan, build and publish a study and read every answer with a link to the moment in the recording. Perspective AI does the same for interviews from Claude, PickFu launches paid polls, and Tally, Jotform, Typeform and SurveyMonkey build forms and surveys and read the submissions. Survicate drafts surveys but you launch them, Weavely publishes forms but cannot read the responses, and Maze, Lyssna, Dovetail, Granola, Enterpret, Chattermill, Zonka Feedback and Canny only read what you already have.

Is Feedbaq free in Claude and ChatGPT?

Yes. Connect Feedbaq to your AI agent for free and the agent becomes your expert researcher: it plans and builds studies, and you preview them, free, for as long as you like. Publishing a study to collect responses costs $49 a month, with every feature and unlimited studies and team members. Feedbaq is listed in Claude’s connector directory.

Can people answer a survey inside ChatGPT or Claude?

Not in these tools. Your agent builds the study or form in the chat, and people answer on a link in their browser. Weavely shows a working preview inside ChatGPT so you can try the form before you publish, and Feedbaq participants answer by voice, screen recording or text with no account.

Which tools let my agent read full answers instead of a summary?

Feedbaq returns every answer in full with a transcript timed word by word and a link to the moment. Perspective AI and Speak AI return full transcripts, Tally and Jotform return every submission, Maze returns raw transcripts and quotes, and Enterpret links every number to the source records. On Granola’s free plan, the connection reaches only your last 30 days of notes.

Which connected tools can I use to analyze feedback I already have?

For past research, Maze, Lyssna and Dovetail; for calls, Granola; for feedback that keeps arriving from support, reviews and surveys, Enterpret, Chattermill and Zonka Feedback; for feature requests, Canny. None of these creates a new study, so pair them with Feedbaq when you need to ask customers why.

What is the difference between a directory listing and a custom connector?

A directory listing appears in the assistant’s own list, so you search for the tool and click connect. A custom connector in Claude, or developer mode in ChatGPT, takes the vendor’s server address, and in a company workspace an admin often has to allow it first. Both give the agent the same tools once connected.

Keep reading

Want the full picture of AI research tools beyond ChatGPT and Claude? Read our guide to AI user research tools for product managers.

An AI research agent: user research without a research team

Connect Feedbaq to Claude, ChatGPT or another AI agent and it runs the whole research loop: writes the study, publishes it and analyzes every answer.

Read the guide

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

24 AI user research tools for PMs, sorted by research job, with the one test that matters: can you open the customer's words behind a finding? Prices checked.

Read the guide

Sources

Lenzner, Höhne and Gavras, spoken versus written answers to open-ended probes Journal of Survey Statistics and Methodology, 2024.
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 7 October 2026.
Feedbaq in Claude
Feedbaq, MCP setup and tools
Perspective AI, MCP documentation Checked 7 October 2026.
Perspective AI, pricing Checked 7 October 2026.
Speak AI, MCP server Checked 7 October 2026.
Speak AI, pricing Checked 7 October 2026.
PickFu, using Claude and PickFu as your research team 11 May 2026. Checked 7 October 2026.
PickFu for agents
PickFu, pricing Checked 7 October 2026.
Tally, Claude connector Checked 7 October 2026.
Tally, MCP server Checked 7 October 2026.
Tally, Pro and Business prices Checked 7 October 2026.
Jotform, MCP Checked 7 October 2026.
Jotform, pricing Checked 7 October 2026.
Typeform, MCP server Checked 7 October 2026.
Typeform, pricing Checked 7 October 2026.
SurveyMonkey, Claude connector Checked 7 October 2026.
SurveyMonkey, plugin in ChatGPT Checked 7 October 2026.
SurveyMonkey, pricing Prices shown in euros. Checked 7 October 2026.
Survicate, MCP and how to connect it Checked 7 October 2026.
Survicate, pricing Checked 7 October 2026.
Formbricks, connect AI agents (MCP) Checked 7 October 2026.
Formbricks, pricing Checked 7 October 2026.
Weavely, ChatGPT plugin Checked 7 October 2026.
Weavely, Claude connector Checked 7 October 2026.
Maze, MCP Checked 7 October 2026.
Maze, MCP demo and directory listings
Maze, pricing Checked 1 October 2026.
Lyssna, making your research reusable with AI Checked 7 October 2026.
Lyssna, pricing Checked 7 October 2026.
Dovetail, Claude connector Checked 7 October 2026.
Dovetail, MCP server Checked 7 October 2026.
Dovetail, pricing Checked 7 October 2026.
Granola, MCP Checked 7 October 2026.
Granola, pricing Checked 7 October 2026.
Enterpret, May 2026 product highlights Checked 7 October 2026.
Enterpret, MCP server setup
Chattermill, MCP server setup Checked 7 October 2026.
Chattermill, product and customers
Zonka Feedback, MCP server Checked 7 October 2026.
Zonka Feedback, pricing Checked 7 October 2026.
Canny, MCP server Checked 7 October 2026.
Canny, pricing Checked 7 October 2026.