Connect: MTurk Alternative for Researchers
Platform closure notice · August 2026
MTurk Is Shutting Down.Migrate to Connect AI.
On September 30, 2026, Amazon Mechanical Turk permanently closes — and CloudResearch’s own MTurk Toolkit retires the same day. If you’re comparing alternatives to Mechanical Turk, this page explains what’s ending, what it means for studies already in the field, and how to move to Connect AI — from the team that spent a decade making MTurk work for researchers.
Free to join · No AWS account required · First study live today
“Following an assessment, we’ve made the decision to close Amazon Mechanical Turk, effective September 30, 2026. Amazon Mechanical Turk will permanently close on September 30, 2026.”
Site-wide banner, mturk.com · posted August 25, 2026
Open to everyoneAmazon Mechanical Turk is a trademark of Amazon.com, Inc. CloudResearch is not affiliated with or endorsed by Amazon.
Nature Human Behaviour places Connect AI first for response quality among nine opt-in online samples.
Read the StudyThe news, plainly
What’s Happening, and Why It Matters
What’s happening:
On August 25, 2026, Amazon posted a notice to mturk.com: following an assessment, it has decided to close Amazon Mechanical Turk, effective September 30, 2026. This is the full shutdown, not June’s narrower change — the marketplace ends after 21 years, and it takes the MTurk Worker type in SageMaker Ground Truth and Amazon Augmented AI with it.
What that means for work in flight:
September 30 is the last day HITs can be submitted, and anything still unsubmitted expires that day. Requesters keep the standard 30-day window afterward to approve or reject completed work and award bonuses, through October 30, 2026. After that the account is finished — so pull down your results files, worker IDs, and payment records before the console goes away.
Why it matters:
Every lab, class, and company still collecting on MTurk now needs a new source of participants, and not much time to find one. CloudResearch — which built the most widely used research toolkit on top of MTurk — retires that product the same day. The question is no longer whether to move; it’s where, and how quickly you can be running again.
Moving is the easy part
Up and Running the Same Day
No AWS account, no procurement queue, no rebuild. Three steps between a closing MTurk account and your next project collecting data.
- Minutes
Create your account
Sign up with your email. No AWS account, no approval queue, no invitation, nothing for procurement to review first.
- Copy and paste
Bring what you already built
Your CrowdHTML templates move across with minimal changes, and any survey with a URL still works — Qualtrics, SurveyMonkey, Gorilla, Typeform. Your instruments don't change.
- The same day
Launch your project
Set demographic filters or census-matched quotas, then launch. Projects start returning responses instantly.
The compliance work is already done
Find the security and compliance documentation your institution needs in our Trust Center. And if you need help with IRB, procurement, or onboarding requirements, reach out to our team—we’ll work with you to keep the process moving smoothly.
SOC 2
SOC 2 Type II
Microsoft SSPA
TX-RAMP
Also HIPAA and GDPR compliant, with ISO 27001-aligned controls.
The countdown20 days remaining
Every Date That Still Matters
The Toolkit sunset is announced
Our co-founder publishes “Sunsetting the MTurk Toolkit” — six weeks before Amazon's first closure notice. The research community, he writes, has already moved on.
MTurk closes to new customers
AWS moves Mechanical Turk into “maintenance.” New requester accounts can no longer be created. Amazon says existing users are not impacted.
Amazon announces the shutdown
A notice goes up on mturk.com: following an assessment, Amazon has decided to close Mechanical Turk outright.
MTurk closes — and the Toolkit with it
The last day HITs can be submitted. The marketplace goes dark after 21 years, and our MTurk Toolkit retires the same day.
In 20 daysMigrate to Connect AI
Everything you ran on MTurk has a home here. Most researchers launch their first Connect AI study the same day they sign up.
Get Started Free
The case
Considering MTurk Alternatives? Here’s What Changes With Connect AI
Connect AI isn’t MTurk with a coat of paint. It’s what we built after ten years of watching exactly where a general-purpose task marketplace falls short for research.
Vetting that never stops
On MTurk, an approval rating was the best quality signal you had. Every Connect AI participant is screened by Sentry — behavioral checks, device fingerprinting, AI-content detection — on every study, not just once at sign-up.
Targeting without qualification surveys
Stop paying to find out who's in your sample. Connect AI profiles participants on hundreds of demographic and psychographic attributes, with quotas and census-matched sampling built in.
No AWS billing maze
One account, transparent per-response pricing, and a receipt your grants office will understand. No AWS console, no surprise fees to reconcile at the end of the month.
Longitudinal research, built in
What took qualification gymnastics on MTurk is a first-class feature here: Waves re-contacts your exact participants for follow-up sessions — with an 80% return rate.
Messaging that isn't a workaround
Conversations lets you message participants anonymously, inside the platform — no bonus hacks, no email scripts, no anonymity risks.
Accountability in both directions
Participants rate studies, and researchers rate participants. That two-way reputation system is why Connect AI holds 4.7★ across 7,876 participant ratings.
Fast — without the bots
Studies fill in hours, not days. And because Sentry screens every session, speed never comes at the price of quality: our tracker studies show 100% AI bot detection.
Support from actual researchers
Our support team includes PhDs who run online studies themselves. Ask a methods question, get a methods answer — usually the same day.
Check our claims
Don’t Take Our Word for It
Independent · peer-reviewed“Representativeness versus response quality: Assessing nine opt-in online survey samples”
Also independently verified: Douglas, Ewell & Brauer (2023), PLOS ONE — in a five-platform comparison of MTurk, Prolific, CloudResearch, Qualtrics, and SONA, CloudResearch’s vetted participants delivered among the highest data quality tested.Read the Study
“Connect has the highest data quality I’ve ever gotten! I’m not sure how you do it, but it’s incredible.”
“I’ve used Connect since it first came around, and it has become my first stop for data collection. The quality of participants is on par with — if not better than — other online samples I use.”
“Connect combines best-in-class researcher tools with an outstanding panel of attentive and reliable survey respondents.”
Make the move
Your MTurk Task, Running on Connect AI
Bring the spreadsheet and CrowdHTML you already use. Change the data placeholders, preview the participant experience, and run it with modern project controls.
What you can run
Every Task Type You Ran on MTurk
Connect AI runs on regular HTML, with support for AWS Crowd HTML Elements — so the templates you wrote for MTurk move across with minimal changes. Each example below is rendering live against sample data.
Quick questionnaires and longer studies.
Typical uses: Opinion and attitude studies, concept and message testing, screeners, and follow-up questionnaires — any study where you are asking people written questions rather than labelling something.
Show someone an image and ask which category it belongs in.
Typical uses: Sorting images into categories — content moderation, product tagging, and building labelled image sets to train or test a model.
Draw boxes around the things you care about in a photo.
Typical uses: Marking where things are in a photo, for training object detection, checking shelf and storefront images, and quality inspection.
Mark the key fields on a scanned document.
Typical uses: Pulling specific fields off documents — receipts, invoices, forms and statements — including labelled data for document-reading models.
Check a record field by field, and correct anything that is wrong.
Typical uses: Checking records against a source and correcting what is wrong: business listings, catalogues, claims, and compliance reviews.
Typical uses: Anything the others do not cover — rating video or audio, reviewing transcripts, comparing two AI answers side by side, or an instrument your team designs from scratch.
This is the data the task produces — the shape it arrives in when you download the batch results.
Nothing was filled in — pick some answers in the task, then submit again.
Rendering against the selected row. The task is fully interactive.
Survey
2 of 23 lines change<script src="https://assets.crowd.aws/crowd-html-elements.js"></script>
<crowd-form>
<h2>${study_name}</h2>
<h3>1. How often do you use
${topic}?</h3>
<crowd-radio-group name="frequency">
<crowd-radio-button name="daily">
Daily</crowd-radio-button>
<!-- four more options -->
</crowd-radio-group>
<h3>2. Which matter to you?</h3>
<crowd-checkbox name="price">Price</crowd-checkbox>
<crowd-checkbox name="speed">Speed</crowd-checkbox>
<h3>3. How satisfied are you?</h3>
<crowd-slider name="satisfaction"
min="1" max="7" step="1" pin></crowd-slider>
<h3>4. Anything else?</h3>
<crowd-text-area name="open"></crowd-text-area>
</crowd-form>
<script src="https://assets.crowd.aws/crowd-html-elements.js"></script>
<crowd-form>
<h2>{{ task.row_data['study_name'] }}</h2>
<h3>1. How often do you use
{{ task.row_data['topic'] }}?</h3>
<crowd-radio-group name="frequency">
<crowd-radio-button name="daily">
Daily</crowd-radio-button>
<!-- four more options -->
</crowd-radio-group>
<h3>2. Which matter to you?</h3>
<crowd-checkbox name="price">Price</crowd-checkbox>
<crowd-checkbox name="speed">Speed</crowd-checkbox>
<h3>3. How satisfied are you?</h3>
<crowd-slider name="satisfaction"
min="1" max="7" step="1" pin></crowd-slider>
<h3>4. Anything else?</h3>
<crowd-text-area name="open"></crowd-text-area>
</crowd-form>
Image classification
1 of 17 lines change<script src="https://assets.crowd.aws/crowd-html-elements.js"></script>
<crowd-form>
<crowd-image-classifier
name="subject"
src="${subject_photo}"
header="What is the main subject?"
categories="['House', 'Car',
'Storefront', 'Unclear']">
<full-instructions header="Instructions">
<p>Pick the subject that fills most of
the frame.</p>
</full-instructions>
<short-instructions>
<p>Choose one category.</p>
</short-instructions>
</crowd-image-classifier>
</crowd-form>
<script src="https://assets.crowd.aws/crowd-html-elements.js"></script>
<crowd-form>
<crowd-image-classifier
name="subject"
src="{{ task.row_data['subject_photo'] }}"
header="What is the main subject?"
categories="['House', 'Car',
'Storefront', 'Unclear']">
<full-instructions header="Instructions">
<p>Pick the subject that fills most of
the frame.</p>
</full-instructions>
<short-instructions>
<p>Choose one category.</p>
</short-instructions>
</crowd-image-classifier>
</crowd-form>
Bounding boxes
1 of 15 lines change<script src="https://assets.crowd.aws/crowd-html-elements.js"></script>
<crowd-form>
<crowd-bounding-box
name="storefront_boxes"
src="${storefront_photo}"
header="Box the sign, door and windows"
labels="['Sign', 'Door', 'Window']">
<full-instructions header="Instructions">
<p>Draw one box per item, as tight as you can.</p>
</full-instructions>
<short-instructions>
<p>Box every sign, door and window.</p>
</short-instructions>
</crowd-bounding-box>
</crowd-form>
<script src="https://assets.crowd.aws/crowd-html-elements.js"></script>
<crowd-form>
<crowd-bounding-box
name="storefront_boxes"
src="{{ task.row_data['storefront_photo'] }}"
header="Box the sign, door and windows"
labels="['Sign', 'Door', 'Window']">
<full-instructions header="Instructions">
<p>Draw one box per item, as tight as you can.</p>
</full-instructions>
<short-instructions>
<p>Box every sign, door and window.</p>
</short-instructions>
</crowd-bounding-box>
</crowd-form>
Receipt fields
1 of 16 lines change<script src="https://assets.crowd.aws/crowd-html-elements.js"></script>
<crowd-form>
<crowd-bounding-box
name="receipt_fields"
src="${receipt_scan}"
header="Box the merchant, date and total"
labels="['Merchant', 'Date', 'Total',
'Line item']">
<full-instructions header="Instructions">
<p>One box per field, as tight as you can.</p>
</full-instructions>
<short-instructions>
<p>Box each field once.</p>
</short-instructions>
</crowd-bounding-box>
</crowd-form>
<script src="https://assets.crowd.aws/crowd-html-elements.js"></script>
<crowd-form>
<crowd-bounding-box
name="receipt_fields"
src="{{ task.row_data['receipt_image'] }}"
header="Box the merchant, date and total"
labels="['Merchant', 'Date', 'Total',
'Line item']">
<full-instructions header="Instructions">
<p>One box per field, as tight as you can.</p>
</full-instructions>
<short-instructions>
<p>Box each field once.</p>
</short-instructions>
</crowd-bounding-box>
</crowd-form>
Full verification form
6 of 28 lines change<script src="https://assets.crowd.aws/crowd-html-elements.js"></script>
<style> /* ~90 lines of layout CSS */ </style>
<crowd-form answer-format="flatten-objects">
<div class="vrf-header">
<h1>Verify Business Listing Details</h1>
<span>Record ${record_id}</span>
</div>
<crowd-instructions link-text="View full instructions">
<short-summary>...</short-summary>
<detailed-instructions>
<div>${batch_instructions}</div>
</detailed-instructions>
<positive-example>...</positive-example>
<negative-example>...</negative-example>
</crowd-instructions>
<img src="${image_url}">
<td>${business_name}</td>
<td>${street_address}</td>
<td>${phone}</td>
<fieldset> /* 5 questions, 3 with
conditional correction fields */ </fieldset>
</crowd-form>
<script> /* reveal-on-select logic */ </script>
<script src="https://assets.crowd.aws/crowd-html-elements.js"></script>
<style> /* ~90 lines of layout CSS */ </style>
<crowd-form answer-format="flatten-objects">
<div class="vrf-header">
<h1>Verify Business Listing Details</h1>
<span>Record {{ task.row_data['record_id'] }}</span>
</div>
<crowd-instructions link-text="View full instructions">
<short-summary>...</short-summary>
<detailed-instructions>
<div>{{ task.instructions }}</div>
</detailed-instructions>
<positive-example>...</positive-example>
<negative-example>...</negative-example>
</crowd-instructions>
<img src="{{ task.row_data['image_url'] }}">
<td>{{ task.row_data['business_name'] }}</td>
<td>{{ task.row_data['street_address'] }}</td>
<td>{{ task.row_data['phone'] }}</td>
<fieldset> /* 5 questions, 3 with
conditional correction fields */ </fieldset>
</crowd-form>
<script> /* reveal-on-select logic */ </script>
Custom HTML
5 of 17 lines change<!-- An HTMLQuestion HIT: your own markup,
submitted to MTurk's external endpoint. -->
<form>
<h2>Review clip ${clip_id}</h2>
<iframe src="https://www.youtube.com/
embed/${video_id}"></iframe>
<label>Would you watch this again?
<input type="range" name="appeal"
min="1" max="5">
</label>
<textarea name="notes"></textarea>
<input type="hidden" name="assignmentId"
value="${assignmentId}">
<button type="submit">Submit</button>
</form>
<!-- A Custom HTML task: the same markup, with
Connect serializing the form on submit. -->
<form>
<h2>Review clip {{ task.row_data['clip_id'] }}</h2>
<iframe src="https://www.youtube.com/
embed/{{ task.row_data['video_id'] }}">
</iframe>
<label>Would you watch this again?
<input type="range" name="appeal"
min="1" max="5">
</label>
<textarea name="notes"></textarea>
<!-- no hidden id needed -->
<button type="submit">Submit</button>
</form>
How a project runs
Three stages, start to finish
Upload your batch
A CSV where every row is one task item. Column headers become the variables your template reads.
Your template renders per row
The same template runs once for every row, with that row’s values substituted in — exactly what you tried above.
Collect and resolve
Request several independent reviews per row, keep the ones that agree, and send disagreements back out for another round.
Connect AI runs and manages the participant assignments. Agreement grouping and exception rounds are an example of the automation teams build around those results.
The same ideas, renamed
Your MTurk Vocabulary, Translated
Works with the survey tools you already use: Qualtrics · SurveyMonkey · Gorilla · Typeform · any survey URL
Straight answers
The Questions Researchers Are Asking
Yes. On August 25, 2026, Amazon posted a notice to mturk.com stating that, following an assessment, it has “made the decision to close Amazon Mechanical Turk, effective September 30, 2026,” and that the service “will permanently close” on that date. This is the end of the marketplace itself, not the narrower change announced in June. MTurk had already been closed to new customers since July 30, 2026; now existing requester and worker accounts are ending too.
September 30, 2026 is the last day workers can submit HITs, and any HIT still unsubmitted at that point automatically expires. After that, Amazon applies its standard 30-day approval policy: requesters can approve or reject completed HITs and award bonuses until October 30, 2026. No new data can be collected after September 30 — so if a study has to finish on MTurk, it has to be in the field now.
Four things, in this order. Stop scheduling new MTurk data collection and move those studies to another platform. Get anything already in the field submitted before September 30. Download everything you will need later — results files, the worker IDs you rely on for exclusions or longitudinal matching, and payment records for your grants office — because the console will not be there afterward. Then settle up: approve or reject outstanding work and pay any bonuses before October 30, 2026.
Yes. Amazon's notice states that the closure also applies to SageMaker Ground Truth and Amazon Augmented AI: the Amazon Mechanical Turk Worker type is no longer available when creating labeling jobs or human review workflows as of September 30, 2026. If you route data labeling, model evaluation, or human-in-the-loop review through MTurk workers, that pipeline needs a new source of humans.
CloudResearch is retiring the MTurk Toolkit on September 30, 2026 — the same day MTurk itself closes. We announced our decision in May 2026, three months before Amazon announced the shutdown. It was our first product, and we've written openly about why it was time: the research community had largely moved on from MTurk, and our newer tools — Connect AI, Engage, and Sentry — were built from the ground up with the data-quality controls that MTurk-era research had to bolt on. The Toolkit was always a layer on top of MTurk; when MTurk goes, there is nothing left for it to sit on.
No platform can port MTurk worker identities — those accounts belong to Amazon's ecosystem, and they end with it. What you can do is rebuild the panel capability itself, usually faster than expected: Connect AI's targeting filters recruit to your criteria without qualification studies, and Waves lets you re-contact the same participants across sessions with an 80% return rate. For complex longitudinal programs, our team can help you design the transition.
Independent research puts Connect AI at the top of the market. A July 2026 study in Nature Human Behaviour assessed nine opt-in online survey samples and placed Connect AI first on its response-quality leaderboard, with participants consistently among the highest in attentiveness, effort, and honesty. It echoes a 2023 PLOS ONE comparison of MTurk, Prolific, CloudResearch, Qualtrics, and SONA, which found CloudResearch's vetted participants delivered among the highest data quality of any platform tested. Every Connect AI participant also passes Sentry screening on every study, which is how our tracker studies sustain 100% AI bot detection.
Stagnaro et al., 2026, Nature Human Behaviour → · Douglas et al., 2023, PLOS ONE →
You set participant pay (with fair-pay minimums), and Connect AI adds a transparent per-response fee — there's no AWS billing layer and there are no platform subscription fees to get started. See our pricing page for current rates, or talk to our team about volume programs.
Usually, and faster than researchers expect — this is the step that decides how quickly a lab can actually move, so it is worth starting early. CloudResearch is SOC 2 Type II certified, HIPAA and GDPR compliant, and runs ISO 27001-aligned controls; the documentation your institution will ask for is published in our Trust Center rather than requested by email. Two things tend to shorten the review compared with MTurk: Connect AI is a direct account, so there is no separate cloud vendor for procurement to assess alongside it, and your study-level protocol carries over unchanged — participants are consented inside your own instrument exactly as they were before. If your institution needs something specific, our team can work through it with you.
For a typical survey study: about as long as your coffee takes to cool. Create a free account, paste your survey link, choose your audience, and launch — researchers routinely go from sign-up to live study the same day. That speed matters more than usual right now: with MTurk closing on September 30, 2026, most labs have weeks rather than quarters to make the switch. Larger programs (longitudinal panels, API integrations, AI-training pipelines) get support from our team.
Go deeper
The Record Is Public
We’ve been documenting MTurk’s trajectory for years. The receipts are all still up.
MTurk’s Chapter Ends September 30.
Your Next Study Doesn’t Have to Wait.
Join the platform built by the people who know exactly what MTurk did well — and exactly where it fell short.
This page is kept current as Amazon’s timeline develops. Last updated August 31, 2026.