Skip to content

Human feedback for AI agents

Your agent made it. Real people tell you if it works.

Show US adults an ad, a landing page, a product screen or a recorded flow your agent made. They tell you what works, what confuses them and what looks fake. You order it like any study, with no sales call.

From a research panel of 137,863+ adults who signed up to take part in paid studies.

What each person sees: your agent’s work above your question. This one is an example; you write your own.

User testing for AI agents

What people can test for you

Anything your agent made that a person can look at, listen to or read. Show it as an image, a video or audio above your questions.

  • Ads

    A video ad your agent made. People react on camera, then watch again and stop where it looks fake or loses them. Or they rate and rank 2 to 5 video ads and say what looks fake, with no camera. An image ad goes above your questions.

    Order it as: Video ad test, Ad ranking or Survey

  • Landing pages and screens

    Screenshots of a page or an app. Is it clear, would they sign up, and what looks off?

    Order it as: Quick poll or Survey

  • Product flows

    A flow on your live website, or a screen recording of it. People try it themselves while their screen and voice are recorded, or watch the recording and say where they’d get stuck.

    Order it as: Usability test, Survey or Video reaction

  • Copy, names and prices

    Headlines, product names, prices and emails. Which version wins, and does it read like a person wrote it?

    Order it as: Quick poll

  • AI output

    Text, images and voices your agent made. Is it good, and does it look AI-made?

    Order it as: Survey or Quick poll

  • Interviews about it

    One-on-one voice interviews with our AI interviewer about what your agent made, each with its full transcript.

    Order it as: AI interview

How it works

From what your agent made to what people think

You set the study up yourself, or your agent does, on the same pages. Every study is reviewed before it goes to the panel.

  1. 1

    Sign in with an emailed code

    Enter a work email, then the 6-digit code we send to it (it expires in 10 minutes). No password. A new account names its company or team and accepts the Business Terms.

    Sign in
  2. 2

    Build the study

    Press New study, pick a format and write your questions (for an AI interview, its goal and interview guide). Any question can show an image, audio or video above it, from a link. For a video ad test, start at /business/start?start=ad-test: its draft has the moment-by-moment test on, and for the ad you upload the video or give a link to it. From New study, pick Recording, add the video and switch the test on. For an ad ranking, start at /business/start?start=ad-ranking, or pick Ad ranking in a new study, and add 2 to 5 video links. Pick who should take part: an age range, a gender, and US regions or states; the builder shows about how many people match.

    Your studies
  3. 3

    Add funds by card

    Top up $100 to $10,000 at a time, in whole dollars, on Stripe’s secure checkout. Submitting holds the study’s full price from your balance, and what isn’t used comes back when the study ends.

    Balance and holds
  4. 4

    Submit for review

    A person at StudyScout reviews every study before it goes to the panel. We email you when it’s live, or tell you what to change.

    How review works
  5. 5

    Read the results

    Your results page fills in as people take part, and Export CSV downloads every answer. To show your team, make a private, read-only link to the results that needs no login; each link expires. Not for usability tests.

    See a sample

What you get

Every answer, and who it came from

Your results page fills in as people take part. Here is what each response carries, and samples of the rest.

  • Who answered

    Every response comes with a random response ID, the person’s age, age band, gender, US state and region, and the time they took. Responses whose member agreed to the widened words also carry details such as the person’s closest city, when they took part (UTC), the country, region and time zone of their connection, their device class, browser and operating system, and for an interview each answer’s time, speaking seconds and guide question, and a respondent ID that is the same for the same person in each of your studies (not their StudyScout account ID). Never a name, email address, phone number, street address, IP address or account ID.

    Read the terms
  • Sample results page

    A sample study’s results, with made-up data: answers by age, gender and region, text answers, recordings, an AI interview and the CSV.

    See the sample
  • Sample CSV

    The sample study’s export, in the same columns as a real study’s CSV.

    Download it
  • Sample ad test

    A sample ad test’s results: each reaction video with how much of the ad was watched and a transcript, answers by age, gender and region, and the CSV. The second watch is in it too, with made-up stops and comments.

    See the sample ad test

Pricing

From $1 per response

Pay per response, from a balance you top up by card. You pay only for responses that pass our quality checks.

  • Quick poll

    1–3 questions. At least 100 per study.

    You get each answer, with choice answers charted by age, gender and region and short text answers listed.

    $1per response

  • Survey

    Up to 7 minutes. At least 100 per study.

    You get every answer, with choice and scale answers charted by age, gender and region and text answers listed.

    $1.75per complete

  • Video ad test

    Reaction and comments, with transcripts. For the ad, upload the video or give a link to it (a .mp4, .webm or .mov file). A plain video reaction, with no second watch, costs the same. At least 20 per study.

    You get each person’s reaction and each stop on video, to play or download, with transcripts, each stop tagged “Looks fake or AI-made”, “I’d scroll away here” or “Other” and tied to its second of the ad, the share who said anything looked fake or AI-made, and what they think the ad is selling.

    $25per video

  • Recording session

    Script or own words, in 2-minute parts, each with a transcript. At least 20 per study.

    You get each part’s recording, video or audio, with a transcript.

    Per person

    Up to 10 minutes
    $10
    Up to 30 minutes
    $25
    Up to an hour
    $45
  • AI interview

    5–20 min voice interview, with transcript. At least 10 per study.

    You get each interview’s full transcript.

    $50per interview

  • AI video interview

    5–20 min video interview, with transcript. At least 10 per study.

    You get each interview’s transcript, with the video of each answer.

    $75per interview

  • Ad ranking

    2–5 ads rated and ranked, with what looks fake. At least 10 per study.

    You get how often each ad came first, its average place and rating, the share who said it looked fake or AI-made, and each person’s reasons with their age band, gender and state and, for members who agreed, their city, when they took part, and their device and connection.

    $20per person

  • Usability test

    Screen and voice on your website, with a transcript and each task’s result. At least 5 per study.

    You get each person’s screen recording with their voice and a transcript, each task’s time and whether they finished or got stuck, their answers, and a summary of each task, in a CSV too.

    Per person

    Up to 10 minutes
    $25
    Up to 20 minutes
    $40
    Up to 30 minutes
    $60
  • Your own survey link

    We send matching members to your survey in your own tool (Qualtrics, SurveyMonkey, Alchemer and the like) and count each complete. At least 100 per study.

    You get each complete with the person’s age, gender and state and, for members who agreed, their city, when they took part, and their device and connection, in a CSV that joins your survey tool’s own export by respondent ID.

    Per complete

    Up to 7 minutes
    $1.75
    Up to 15 minutes
    $2.50
    Up to 20 minutes
    $3.50

Reject up to 10% of responses within 7 days of a study closing; rejected responses aren’t charged.

1:1 interviews, focus groups, AI-output ratings or something else? Ask us for a quote. support@joinstudyscout.com

Studies go to US adults 18 and over. Prices are in US dollars.

For agents

Your agent can order a study today

Through the API or our MCP server, with a key: it signs in with a code sent to an email address it can read, makes a key, drafts the study and submits it. A person pays the card link, unless your agent can.

A plain-text summary for agents

Every format with its price and minimum, the steps to order, what results include and the rules a study follows, in one file an agent can read.

API

Order a study through the API

A key, a few JSON requests and the results. A person pays the card link, unless your agent can, and a person at StudyScout reviews every study before it goes to the panel.

  1. 1

    Get a key

    An agent with its own mailbox can do this alone: POST its address to /api/auth/send-otp, POST the 6-digit code it gets (it expires in 10 minutes) to /api/auth/verify-otp and keep the session cookie, POST a company name and acceptTerms: true to /api/business/account, which accepts the Business Terms for that company, then POST to /api/business/api-keys. Or a person makes a key on the dashboard. The key is shown once.

    Sign in
  2. 2

    Read the reference

    Every endpoint under /api/v1, with a whole example for every format, in OpenAPI 3.1. Send the key on each request as Authorization: Bearer.

    Open it
  3. 3

    Draft the study

    GET /api/v1/account says what you pay per format, GET /api/v1/audience how many people match, and POST /api/v1/studies makes the draft from JSON. Its answer's issues say what still blocks launch, and its price what it will cost. An uncompressed audio recording is a recording question with recording.kind audio and recording.rawAudio true: each comes back as a WAV made with the browser’s echo cancellation, noise suppression and automatic volume off. A video ad test is a recording question with recording.moments true and the ad as a video file link; its own questions are added for you. An ad ranking is format ad_ranking with 2 to 5 ads in adRanking.ads. A usability test is format usability: usability.url is an https link to your website, usability.minutes is 10, 20 or 30, and usability.tasks holds 1 to 5 tasks in plain words. Its recordingUse says how you’ll use the recordings.

  4. 4

    Add funds

    POST /api/v1/top-up returns a link to Stripe’s secure checkout for $100 to $10,000, in whole dollars. A person who holds the card pays it, or an agent that can: no card details pass through the API.

  5. 5

    Launch, then wait for review

    POST /api/v1/studies/{id}/launch holds the study’s full price from your balance. A person at StudyScout reviews every study before it goes to the panel: read GET /api/v1/studies/{id}/status until it says live, or rejected with a reason.

  6. 6

    Read the results

    GET /api/v1/studies/{id}/results as people take part, or /results/summary for a compact read, with its links to the CSVs and to recordings or transcripts. Each response comes with the person’s age, gender and US state, never their contact details.

MCP

Connect it as an MCP server

The same key and the same studies as the API, as tools your agent can pick up by itself. It needs no code of ours and no package to install.

  1. 1

    Connect your agent

    Add https://www.joinstudyscout.com/api/mcp as a remote MCP server (Streamable HTTP), with the same key as the API sent as the header Authorization: Bearer. An agent gets a key as the API section says. In Claude Code:

    claude mcp add --transport http studyscout https://www.joinstudyscout.com/api/mcp --header "Authorization: Bearer YOUR_KEY"
  2. 2

    Order with its tools

    One tool for each thing the API does: getAccount for prices and balance, createStudy for a draft, createUpload for a video from your own files, launchStudy to submit it, getStudyStatus to follow it, and getStudyResultsSummary to read the results.

  3. 3

    Know what still takes a person

    Whoever holds the card pays the link createTopUp returns, unless your agent can. And a person at StudyScout reviews every study before it goes to the panel: no key can approve one.

FAQ

Questions about testing what an agent made

Can people use my live product or site?
Yes, in a usability test. On a computer, people open your website in another tab, do the tasks you write and say out loud what they think as they go; you get each person’s screen recording with their voice and a transcript. A study can’t send people anywhere else, or ask them to install anything (Business Terms, section 6).
Who are the people?
US adults 18 and over, from a research panel of 137,863+ adults who signed up to take part in paid studies. You pick the age range, gender and US regions or states, and see about how many people match before you launch.
Can people tell me if an ad looks AI-made?
Yes. In a video ad test, each person says whether anything looked fake or AI-made, then stops on the moment and says why on camera, tied to that second of the ad. An ad ranking asks the same about each ad, and where, with no camera. They’re telling you what looks off to them. They can’t know how it was made.
How do I know the answers are real?
We screen out responses that come in faster than a real person could answer, and ones that fail an attention check when a study has one; those aren’t charged. You can also reject up to 10% of responses within 7 days of a study closing; rejected responses aren’t charged.
Is there a usability testing API, or an MCP server?
Both. There is an API, whose OpenAPI document lists every endpoint, and an MCP server with a tool for each. Each business makes its own keys, and the same key works for both.
How fast do results come in?
Every study is reviewed before it goes to the panel. Once it’s live, each response shows up on your results page as it comes in.

Put your agent’s work in front of real people

Show it, ask what you need to know, and read every answer with the person’s age, gender and state and, for members who agreed, their city, when they took part, and their device and connection.

From $1 per response.