sales-skills/sales-cloudresearch

CloudResearch Platform Help

CloudResearch (cloudresearch.com) is a data-quality-first online research-participant platform — it recruits, screens, and pays vetted participants for surveys, online tasks, and behavioral / AI-training studies. Its edge is data quality: Connect is CloudResearch's own panel, screened continuously by Sentry (behavioral analysis, device fingerprinting, AI-content detection, bot/duplicate/IP checks), with census-matched sampling and longitudinal Waves re-contact. It's the bring-your-own-instrument panel — you host the survey/task in your own tool (Qualtrics, SurveyMonkey, a custom app) and Connect supplies the humans.

It's the data-quality-first peer to Prolific (/sales-prolific): the same recruit class, but where Prolific's edge is an instant self-serve API token, CloudResearch's edge is Sentry screening and its API key is support-provisioned. CloudResearch also runs a legacy MTurk Toolkit over Amazon Mechanical Turk — but MTurk closes to new customers 2026-07-30, so steer new work to Connect.

Two things to say almost every time:

• Vetted people answering ≠ demand. A clean, high-quality sample is strong signal for problem/behavior research — but a participant saying "I'd pay" is still not a purchase. Keep the why; take a go/no-go to a real behavior test (/sales-idea-validation, /sales-funnel).
• The reward is not the cost. A platform fee rides on top (best-effort 25% academic / 40% commercial) — budget reward + fee and confirm at cloudresearch.com/pricing before launching.

Step 1 — Gather context

If references/learnings.md exists, read it first for accumulated platform knowledge.

Ask only what you can't infer:

1. What do you need?
   • A) Recruit / target — demographic filters, census-matched samples, quotas, feasibility checks
   • B) Study lifecycle — create → wire the survey redirect → launch → monitor → review/approve
   • C) Data quality — attention checks, screening out bots / AI-pasted answers, when to reject
   • D) Pay — approve, reject, bonus, and the platform fee + minimum pay
   • E) Automate / integrate — the Connect REST API, survey-tool integration, MTurk Toolkit vs Connect
2. Which product? Connect (its own panel — the default), Prime Panels (aggregated third-party panels), or the legacy MTurk Toolkit. Where does the survey/task live? (Connect supplies participants; the instrument is external.)

Skip-ahead: if the user wants to compare research/recruit tools across the market or the validate-before-building method, that's /sales-idea-validation — route in Step 2.

Step 2 — Route or answer directly

| If the user's question is about… | Route to | |---|---| | Comparing research/recruitment/idea tools across the market, or the validate-before-building method | /sales-idea-validation {question} | | Prolific — the self-serve-API peer panel (instant token, HMAC webhooks, MCP, CLI) | /sales-prolific {question} | | Recruiting B2B / hard-to-reach professionals for moderated interviews (a recruit-only marketplace) | /sales-respondent {question} | | An all-in-one recruit + run + repository research suite | /sales-great-question {question} | | Running a usability / IA test (card sort, tree test, prototype tasks) on a built artifact | /sales-uxtweak or /sales-lyssna {question} | | Running a real behavior demand test (smoke-test page, waitlist, pre-sale) after research | /sales-idea-validation or /sales-funnel {question} |

When routing, give the exact command: "This is a {domain} question — run: /sales-idea-validation {original question}"

Otherwise, answer CloudResearch-specific questions using Step 3.

Step 3 — CloudResearch reference

Read references/platform-guide.md for the full reference — the module/automation-surface table (what's API-accessible vs UI-only), pricing/fee + minimum-pay math, the Account → Project → Assignment data model, survey integration (redirect + participantId), and quick-start recipes. Read references/cloudresearch-api-reference.md for the API surface (support-provisioned X-API-KEY auth, the project/assignments/demographics functions, idempotency tokens, and the poll-not-webhook reality). Answer using only the relevant section — don't dump the full reference.

Step 4 — Actionable guidance

• Say the caveat: a clean sample is not demand. CloudResearch gets real, well-screened people to answer (strong signal — genuine behavior/language) but a stated "I'd pay" is not observed demand. Keep the why; take the go/no-go to a real behavior test (pre-sale, smoke test) → /sales-idea-validation.
• Budget reward + platform fee, and present it as best-effort. Total ≈ reward + platform fee (best-effort 25% academic / 40% commercial, changed 2025-01-01), with an enforced minimum pay (best-effort ~$7.50/hr, recommended ~$10/hr — low pay also drags quality and speed). State all figures as best-effort and point to cloudresearch.com/pricing.
• Treat data quality as a design problem, not just a panel feature. Sentry stops bots, duplicates, and server-farm/foreign-IP fraud — it does not stop a real human pasting ChatGPT into a free-text box. Add your own attention checks, timing thresholds, and copy-paste detection in the instrument, then reject genuine failures with a clear reason and reserve bonuses for real effort.
• Fix "participants don't fit" with targeting + a feasibility check, not the panel. Apply demographic filters, run a feasibility estimate before launch so a quota isn't impossible, add custom screening questions with attention checks, and use quotas / census-matched sampling to balance the sample.
• Wire the survey redirect correctly — it's the completion signal. Capture participantId as a URL param / embedded field so responses tie back to the panel, and set the End-of-Survey redirect (or a fixed completion code) to fire only after the last question. A misconfigured redirect = completes that never register.
• Automate via the API — but expect poll, not push. The key is support-provisioned (email support), sent as X-API-KEY. Flow: feasibility → create project (send an idempotency token, 24h TTL) → update_status LIVE → poll assignments.list_all + project.retrieve_statistics (there are no webhooks) → approve valid submissions (approval releases payment), reject failures, bonus standouts. Reuse the same idempotency token on retries so you don't double-create or double-charge.
• Steer new work to Connect, not the MTurk Toolkit. MTurk closes to new customers 2026-07-30; the Toolkit is legacy. Connect (CloudResearch's own vetted panel) is the pitch and is unaffected.

If you discover a gotcha or tip not in references/learnings.md, append it there with today's date.

Gotchas

Best-effort from research (2026-07) — pricing (fee 25% academic / 40% commercial, min pay ~$7.50/hr, rec. ~$10/hr), the MTurk sunset date, and API specifics all move; verify at cloudresearch.com/pricing and the live Connect Swagger UI.

• The reward is not the total. A platform fee sits on top (best-effort 25% academic / 40% commercial, raised 2025-01-01) with an enforced minimum pay floor (~$7.50/hr, rec. $10/hr — low pay also drags quality and speed). Budget reward + fee before launching.
• Sentry stops bots, not a human pasting ChatGPT. The current data-quality threat is AI-written open-ended answers from real, verified people — add attention/timing checks and copy-paste detection in your own instrument, then reject failures and reserve bonuses for genuine effort.
• The API key is support-provisioned, not self-serve. Auth is an X-API-KEY header and you must email support to get a key — unlike Prolific's instant token. Treat the key like a password.
• No webhooks — you must poll. No outbound event system / MCP / native Zapier is documented; reconcile by polling assignments.list_all + project.retrieve_statistics, and use idempotency tokens (24h) on create/bonus so retries are safe.
• The MTurk Toolkit is sunsetting. Amazon Mechanical Turk closes to new customers 2026-07-30 — new adopters can't onboard the Toolkit; use Connect (unaffected).
• The survey redirect / participantId is the completion signal. Capture participantId and set the End-of-Survey redirect (or completion code) to fire after the last question, or completes never register.
• A clean sample ≠ demand. Hundreds of well-screened people answering "I'd buy this" is stated preference at scale, not a purchase — take the WTP go/no-go to a real behavior test.
• Rejections affect participants. Reject only genuine failures, with a specific reason — over-rejecting gets flagged and is unfair; that's why attention-check design matters more than blanket rejection.

Related skills

• /sales-prolific — The self-serve-API peer panel: same recruit class, but its edge is an instant Token key + HMAC webhooks + MCP + a CLI, where CloudResearch's edge is Sentry data-quality screening and a support-provisioned key. Pick Prolific when you need a self-serve pipeline; CloudResearch when data-quality screening is the priority. Install: npx skills add sales-skills/sales --skill sales-prolific -a claude-code
• /sales-idea-validation — The validate-before-building method + the full research/recruit/validator tool landscape (a research "yes" is not demand). Install: npx skills add sales-skills/sales --skill sales-idea-validation -a claude-code
• /sales-respondent — The recruit-only B2B/professional interview marketplace (CloudResearch is broad online samples + data-quality screening; Respondent is hard-to-reach professionals for moderated interviews). Install: npx skills add sales-skills/sales --skill sales-respondent -a claude-code
• /sales-great-question — The all-in-one recruit + run + repository suite (CloudResearch is recruit + collect only; you bring your own instrument and repository). Install: npx skills add sales-skills/sales --skill sales-great-question -a claude-code
• /sales-funnel — Build the smoke-test / fake-door landing page that turns a research hypothesis into a real demand test. Install: npx skills add sales-skills/sales --skill sales-funnel -a claude-code
• /sales-do — Not sure which skill to use? The router matches any sales objective to the right skill. Install: npx skills add sales-skills/sales --skill sales-do -a claude-code

Examples

Example 1: "How do I launch a Connect study and approve submissions from my own backend?" (developer/automation)

User says: "I want to create studies and pay people through CloudResearch's API without clicking around the dashboard." Skill does: Explains the API pipeline (Step 4) — email support for an X-API-KEY, then feasibility → create project (with an idempotency token) → update_status LIVE → poll assignments.list_all + project.retrieve_statistics (there are no webhooks) → approve valid submissions (approval releases payment), reject failures, bonus standouts. Warns that the key is support-provisioned (not instant like Prolific) and to reuse the same idempotency token on retries. Result: The user runs the whole study lifecycle from their backend with the payment gate correct.

Example 2: "Half my CloudResearch open-ended answers look ChatGPT-written"

User says: "I'm getting free-text responses that read like AI — is the panel bad?" Skill does: Reframes it as a data-quality/instrument issue, not a bad panel — Sentry screens bots, duplicates, and server-farm fraud but not a real human pasting ChatGPT. Prescribes behavior-specific prompts, attention + timing checks, and copy-paste detection in your own survey, tighter demographic filters with a feasibility check, and rejecting genuine failures with a clear reason while reserving bonuses for quality. Result: The user cleans the sample by fixing screening + attention-check design instead of switching tools.

Example 3: "What will an n=300 study actually cost on CloudResearch?"

User says: "I want 300 participants for a 10-minute survey — what's the real bill?" Skill does: Explains total ≈ reward + platform fee (best-effort 25% academic / 40% commercial) with an enforced minimum pay (~$7.50/hr, rec. $10/hr): e.g. $1.50 × 300 = $450 participant pay, +40% commercial fee $180 → ≈ $630 (academic 25% → ≈ $562.50). Presents all figures as best-effort — confirm at cloudresearch.com/pricing — and suggests paying nearer $10/hr for faster, higher-quality data. Result: The user budgets reward + fee up front and sizes the study to their spend.

Troubleshooting

"My API calls 401 / where's my API key?"

Symptom: Requests fail auth, and there's no self-serve key in the dashboard. Cause: CloudResearch Connect uses an X-API-KEY header, and the key is provisioned by support — it isn't an instant self-serve token like Prolific's. Solution: Email [email protected] (or your account contact) to have a key issued, then send X-API-KEY: <key> on every request. The official OpenAPI spec is auth-gated, so verify the exact base path and reward unit on your first authenticated call.

"Completes aren't registering / participants say the survey didn't mark them done"

Symptom: Participants finish the survey but Connect shows no completion. Cause: The End-of-Survey redirect (or completion code) isn't wired, or participantId isn't being captured, so Connect never gets the completion signal. Solution: In the survey tool, capture participantId from the URL as an embedded/hidden field, and set the wizard's Redirect URL as the End-of-Survey action so it fires only after the last question. If your tool can't redirect, use a fixed completion code shown on the final page instead.

"I'm getting low-quality or seemingly-fraudulent responses"

Symptom: Some responses look bot-like, duplicated, or from suspicious locations. Cause: Sentry screens hard, but no panel is perfect — and AI-pasted free-text from real humans slips past any bot check. Blanket-rejecting is unfair and gets flagged. Solution: Add attention checks, timing thresholds, and copy-paste detection in your instrument; reject only genuine failures with a specific reason; use feasibility + demographic filters + quotas to target better up front; and report clearly fraudulent submissions for data-quality review. Pay nearer the recommended rate — higher pay correlates with higher quality.