One of the best-known data collection services shut down last month. Amazon Mechanical Turk, the crowd marketplace that researchers and AI teams used for surveys, labeling and small online tasks, closed permanently on 30 September 2026, and its homepage now carries a single notice saying so. Anyone who built a data pipeline on it is now shopping for a replacement, and the choice depends on a question that is easy to skip: what kind of data are you actually collecting?
Quite different businesses sell under the same label. The AI-focused data collection companies record speech, images and video from real people to train models, while research platforms recruit verified people for surveys and interviews, and a smaller group gathers public data from the web at scale. This guide covers 11 providers across all three, each checked on its official site in October 2026, with what it collects, who it suits and what to check before you sign.
Data collection services compared
| # | Provider | Type | What it collects | Best for |
|---|---|---|---|---|
| 1 | Zilo AI | Managed AI data service | Data collection, transcription and annotation for AI | Teams that want collection and labeling handled together |
| 2 | Appen | Managed AI data service | Speech corpora, multimodal and physical AI data | Large multilingual speech programs |
| 3 | LXT | Managed AI data service | Text, audio, image and video, remote or on-site | Speech and language data in many locales |
| 4 | Defined.ai | AI data service and dataset marketplace | Custom collection plus ready-made datasets | Buying existing data before commissioning new data |
| 5 | Shaip | Managed AI data service | Healthcare, speech and vision data | Regulated and medical data |
| 6 | TELUS Digital | Managed AI data service | Automotive and robotics field collection | In-vehicle and physical AI data |
| 7 | Prolific | Participant platform | Survey, experiment and AI feedback responses | Academic research and human AI evaluation |
| 8 | Respondent | Participant recruitment | Interviews and surveys with verified professionals | B2B research |
| 9 | Cint | Survey sample marketplace | Survey responses in 130 countries | Research agencies and global studies |
| 10 | SurveyMonkey Audience | Self-serve survey panel | Consumer survey responses | Quick consumer surveys |
| 11 | Bright Data | Web data platform and managed service | Public web data and datasets | Pricing, listings and market monitoring |
1. Zilo AI
Zilo AI is a staffing and data annotation company in Bangalore that runs data collection, transcription, annotation, RLHF and AI evaluation work. It sits first because for most AI teams collection is only half the job. Raw recordings and images are not training data until someone has transcribed, segmented and labeled them, and buying those steps from two vendors means two sets of guidelines, two handovers and two places for errors to creep in. Zilo can take the work from collection through to labeled files.
Its strongest published capability is speech. Zilo’s audio annotation service covers verbatim and clean-verbatim transcription, speaker diarization, word-level timestamps, emotion and sound event tags, and dialect and code-mixed transcription, including major Indian languages. The wider data annotation services cover image, video and text. Zilo says it has over 1,600 trained annotation and ASR experts, has annotated more than 10 million data points for industries from retail to BFSI to healthcare, and works in about 26 languages.
Who it suits: AI teams, particularly those building for Indian and multilingual markets, who want collection and annotation from one partner. What to check: agree the collection scope in writing, including who records the data, how consent is captured and what metadata comes with each file.
2. Appen
Appen is an Australian-listed company that says it has 30 years of AI data experience and covers more than 500 locales. Its data products include speech corpora and voice agent evaluation, vision-language annotation, LiDAR and robotics data for physical AI, and alignment data for frontier models. It is SOC 2 and ISO 27001 certified and also sells off-the-shelf datasets.
Who it suits: large companies that need speech or multilingual data at volume. What to check: project minimums, and how Appen recruits and verifies contributors for your specific languages.
3. LXT
LXT collects text, audio, image and video data, and says it draws on more than 10 million vetted contributors in over 150 countries, covering more than 1,000 language locales. Collection can be remote, mobile on-site or in person, and LXT runs five ISO 27001-certified secure facilities for sensitive work. It also offers annotation, transcription, model evaluation and surveys through clickworker.
Who it suits: speech recognition and conversational AI teams that need many languages and accents. What to check: whether your project will use LXT’s managed workforce or self-managed crowd access, since the quality controls differ.
4. Defined.ai
Defined.ai combines custom data collection with a marketplace of ready-made datasets you can buy and license. It says its crowd includes more than 1.6 million experts across 150 countries and 500 languages, dialects and locales, and it recently gained ISO 42001 certification for AI management. Services include annotation, machine translation, conversational AI data and model evaluation.
Who it suits: teams that would rather license an existing speech or text dataset than wait for a new one. What to check: the license terms of any marketplace dataset, including commercial use and how the original speakers consented.
5. Shaip
Shaip says it sources data from more than 60 countries in audio, video, image and text formats, and it grew out of healthcare data work. Its catalog includes physician dictation, transcribed medical records and speech datasets in dozens of languages. Its site lists GDPR, HIPAA, SOC 2 Type II and ISO 27001 compliance, which matters if your data includes patients or personal details.
Who it suits: healthcare AI and conversational AI teams with strict privacy requirements. What to check: how de-identification is done and documented for medical data.
6. TELUS Digital
TELUS Digital, the digital services arm of the Canadian telecoms group TELUS, runs field data collection that most vendors cannot: in-cabin and open-road collection for automotive in more than 50 countries, and egocentric, point-of-view collection for robotics and physical AI. It says its AI community has more than a million contributors working in over 500 languages and dialects, and it also annotates the data it collects.
Who it suits: automotive and robotics companies that need data captured in the physical world. What to check: the hardware, sensor setup and participant consent process for each collection site.
7. Prolific
Prolific connects researchers with verified participants who are paid to take part in studies. Its customers range from universities to AI developers, and for AI teams it offers human feedback from representative populations for preference tuning, safety evaluations and benchmarks. Studies can be launched quickly, and participants sign up through the same platform.
Who it suits: university researchers, behavioral scientists and AI teams that need human judgments from a defined population. What to check: the fees on top of participant pay, and whether hard-to-reach professional audiences are available in the numbers you need.
8. Respondent
Respondent recruits participants for interviews, usability tests and surveys, and says it has more than 4.3 million verified participants. Its strength is professionals, verified through layered human checks to reduce fraud, with scheduling and an AI interview moderator built in. It has also added managed services for AI training and evaluation using its panel.
Who it suits: product and market research teams that need to talk to specific job titles, such as developers, finance leaders or clinicians. What to check: the incentive you will need to offer for senior professionals, which can exceed the platform fee.
9. Cint
Cint runs a programmatic research marketplace that connects survey buyers with panels around the world. It says more than 200 million surveys are completed through it each year, from over 800 integrated supply partners across 130 countries. It also offers campaign measurement and AI-moderated interviews. Researchers usually bring their own survey tool and use Cint for respondents.
Who it suits: research agencies and insight teams running multi-country studies. What to check: quality controls on each supply source, since sample quality can vary between panels and countries.
10. SurveyMonkey Audience
SurveyMonkey Audience lets you buy survey responses from inside SurveyMonkey. It says its panel reaches more than 335 million people in over 130 countries, with more than 200 targeting options and pricing that starts at one US dollar per response. Census balancing and screening questions help match the sample to your target group.
Who it suits: marketing and product teams that need a quick read from consumers. What to check: the price per response for narrow audiences, which rises quickly as targeting tightens.
11. Bright Data
Bright Data collects public web data rather than responses from people. It sells scraper APIs for more than 800 websites, pre-collected datasets from over 900 domains, real-time data feeds and a managed data acquisition service for enterprises, plus video and audio data for AI training. Its proxy network is the infrastructure underneath.
Who it suits: data teams tracking prices, listings, reviews or public company information at scale. What to check: legal review of each source’s terms and any personal data involved before you collect, because web data carries different risks from consented survey data.
Which type of data collection service do you need?
Start from the data, not the vendor. If you need speech, images or video to train a model, you need a managed AI data service such as Zilo AI, Appen, LXT, Defined.ai, Shaip or TELUS Digital. These companies recruit contributors, record or capture the data, and usually annotate it too. Price is quoted per project and depends on languages, demographics, environments and the level of review.
If you need opinions or behaviour from people, you need a participant platform or sample provider. Prolific and Respondent recruit individuals for studies and interviews, while Cint and SurveyMonkey Audience supply survey responses at volume. Here the main risk is fraud and inattentive respondents, so attention checks and identity verification matter more than headline reach.
If the data already exists on public websites, you need a web data provider such as Bright Data. The cost is mostly technical and legal rather than human.
What changed in this edition
We removed Amazon Mechanical Turk, which has closed. We also removed Typeform, KoboToolbox and Qualtrics. They are useful survey and form software, but you run the collection yourself, so they are tools rather than data collection services. Older market-size figures have been removed rather than repeated without a current source.
How to choose among data collection companies
- Speech data for Indian and other multilingual markets, with annotation included: Zilo AI.
- Speech and language data in hundreds of locales: LXT, Appen or Defined.ai.
- Medical or other regulated data: Shaip.
- Driving and robotics data captured in the field: TELUS Digital.
- Human feedback for AI evaluation: Prolific or Respondent.
- Professionals for interviews: Respondent.
- Survey sample across many countries: Cint, or SurveyMonkey Audience for quick consumer polls.
- Public web data: Bright Data.
Whichever you pick, ask for a small paid pilot, and specify consent wording, metadata fields, file formats and acceptance criteria before collection starts. Changing them halfway through usually means re-recording or re-labeling data you have already paid for. To talk through a speech or annotation project with Zilo AI, contact our team.
Frequently asked questions
What are data collection services?
They are companies that gather data for you, either by recruiting people to record, photograph or answer questions, or by collecting existing data from sources such as the web. Many AI data collection companies also annotate what they collect.
What replaced Amazon Mechanical Turk?
Mechanical Turk closed on 30 September 2026. For research participants and human feedback, Prolific and Respondent are common alternatives. For AI training data, managed services such as Zilo AI, LXT and Appen handle both the workforce and quality control.
How much do data collection services cost?
Survey panels usually charge per completed response, and SurveyMonkey Audience starts at one US dollar. Participant platforms add a fee on top of what you pay each person. Managed AI data collection is quoted per project, because cost depends on languages, demographics, recording conditions and review.
What is the difference between data collection and data annotation?
Collection creates or gathers the raw data, such as recordings, images or survey answers. Annotation adds the labels a model learns from, such as transcripts, bounding boxes or sentiment tags. AI projects usually need both.
Is web scraping a legal way to collect data?
It can be, but it depends on what is collected, from which sites, under which terms, and whether personal data is involved. Get legal advice for your jurisdiction before collecting at scale.