Skip to content

Research report ยท October 2026 edition

Real-World Data Collection Work

On 31 Oct 2026, 107 roles on 16 AI training platforms recruited people to collect real-world data for robots and physical AI, Claru the most (40), at a median advertised $10 an hour; 31.8% had to be done at a site.

Read the month's story: The AI Training Job Market in October 2026

Provisional figures. This month is complete, but late data can still arrive until 3 Nov 2026, when the edition is frozen for good.

Data window
1 Oct 2026 to 31 Oct 2026
31 days, UTC
Computed
1 Nov 2026, 01:16 UTC
Final once frozen
Source
The public job boards of the AI training platforms this site tracks, read every hour (each open role's title, location, workplace field, country list and advertised pay), and the public job boards of AI labs with a documented board API, read on the day the report is computed.
Method
Version 2. Definitions
CC BY 4.0, how to cite

Every figure on this page as CSV and JSON, sent to your work email.

real-world data collection roles open on 31 Oct 2026
107
median advertised rate
$10/hr
of them must be done at a site
32%

Key findings

  1. 1

    On 31 Oct 2026, 107 open roles on 16 AI training platforms recruited people to record real-world data or operate robots; Claru listed the most (40, 37.4%).

    Figure 1
  2. 2

    Read from their titles, the largest kinds are general video and photo capture (48 roles, 44.9%) and in-person speech recording (18, 16.8%); the 1 role that teleoperates or operates robots state too few rates for a median.

    Figure 2
  3. 3

    59 of the 107 roles state an hourly rate, with a median of $10 and typical pay of $7 to $30: 61.0% advertise under $15 an hour and 10.2% $30 or more.

    Figure 6
  4. 4

    31.8% of these roles must be done at a site (34, at TransPerfect DataForce, Welo Data, TELUS Digital AI and Mindrift, median $35 an hour), while 63.6% are remote, recorded where the contributor lives (median $10 an hour), and 4.7% do not say.

    Figure 3
  5. 5

    The roles are open in 70 named countries, most often the United States (38), India (23) and Brazil (17), and 8 are open worldwide without naming one.

    Figure 4
  6. 6

    On-site collection is called in a few cities, most often Pune (7), Los Angeles (6), the San Francisco Bay Area (4) and London (4); the 4 session calls that state a fee pay $55 to $170 for 2.5 hours ($22 to $68 an hour).

    Figure 5
  7. 7

    On the 18 platforms tracked since 1 Oct 2026, 7.0 new real-world data collection roles were listed a week on average in the week to 25 Oct 2026, against 5.0 a week in the week before (+40%).

    Figure 7
  8. 8

    A further 35 open roles in Robotics & Physical-AI Data label, review or edit footage (or record nothing) rather than collect it, at a median of $35 an hour against $10 for collection; they are not counted here.

    Figure 10

Figure 1

Who lists real-world data collection work

Which AI training platforms are recruiting people to record the physical world and operate robots?

PlatformOpen rolesShare of allMedian advertised (per hour)
Claru
40
37.4%$8
TransPerfect DataForce
26
24.3%$30
micro1
11
10.3%$20
Welo Data
7
6.5%n/a
Mindrift
6
5.6%n/a
TELUS Digital AI
3
2.8%n/a
DATAmundi
2
1.9%n/a
Lifted (Upwork)
2
1.9%n/a
OneForma
2
1.9%n/a
Perle Labs
2
1.9%n/a
Appen
1
0.9%n/a
Handshake AI
1
0.9%n/a
Innodata
1
0.9%n/a
Invisible
1
0.9%n/a
Sama
1
0.9%n/a
Terac
1
0.9%n/a

Sample. 107 real-world data collection roles open at the end of 31 Oct 2026 on the AI training platforms this site reads, by the platform that lists them; new listings over 1 Oct 2026 to 31 Oct 2026.

A median is the job board's: shown where the platform's roles state at least 3 distinct hourly rates; n/a otherwise.

Listings, not hires: a platform that posts one role per country or language counts each posting.

New listings: roles first seen in the window on a platform already tracked before that week (a platform's first week is its back catalogue).

No row with any open roles is a small sample: the rows differ so much that each reads on its own from its first.

Source. AITraining.jobs Research, Real-World Data Collection Work, October 2026 edition.

Figure 2

What kind of real-world data collection

Are platforms paying people to teleoperate robots, film first-person video, record household tasks, capture motion or record speech in person?

Kind of collection, read from the titleOpen rolesShareMedian advertised (per hour)
All real-world data collection107
100.0%
$10
General video and photo capture48
44.9%
$10
Speech and voice recorded in person18
16.8%
n/a
Household and at-home tasks16
15.0%
$10
Egocentric (first-person) video4
3.7%
n/a
Motion capture and movement4
3.7%
n/a
Teleoperation and robot operationSmall sample1
0.9%
n/a
Other and unspecified collection16
15.0%
$18

Sample. 107 real-world data collection roles open at the end of 31 Oct 2026, each counted once.

Each role counts once, in the first kind whose words its title contains, in this order: teleoperation and robot operation (teleoperation, robot operators, pilots, trainers and wranglers, data-collection operators); egocentric (first-person) video (egocentric, first-person, POV, head-mounted, wearable and smart-glasses capture, embodied data); household and at-home tasks (household, at-home and in-home recording, chores and kitchens); motion capture and movement (motion capture, mocap, movement and gesture recording); speech and voice recorded in person (speech, voice and speaker sessions held at a site); general video and photo capture (360-degree, car-mounted, smartphone, exocentric and third-person video, photographs, faces); other and unspecified collection (on-site collection projects and moderators, hardware testing, robotics data roles that name no method).

A role counts as collection when its title operates a robot; records, collects or captures (in the robotics and physical-AI data kind of work, or naming a camera medium in any other kind); or holds speech or other collection sessions in person. Roles that label, review, transcribe or edit recorded data, and work on screens, files, documents or cards, are never counted. Speech recorded at home is speech data, not real-world collection.

A median is the job board's, from at least 3 distinct hourly rates; a kind with fewer than 3 open roles is marked thin.

Source. AITraining.jobs Research, Real-World Data Collection Work, October 2026 edition.

Figure 3

On-site or remote

Does real-world data collection have to be done at a site, or where the contributor lives?

Work arrangementOpen rolesShareMedian advertised (per hour)
On-site or hybrid (at a studio, lab or session site)34
31.8%
$35
Remote (recorded where the contributor lives)68
63.6%
$10
Not stated5
4.7%
n/a

Sample. 107 real-world data collection roles open at the end of 31 Oct 2026.

On-site or hybrid: the title or location names a site ('Onsite', 'In-person', 'Hybrid'), a job board with a workplace field (Lever, Ashby, Mercor's own board) marks the role not remote, or the title calls speakers to a place ('Speakers Needed in Pune'), as in the on-site report.

Remote collection is recorded wherever the contributor is (their home, street or car), not at a site the platform runs; it is not desk work.

Not stated: neither the title, the location nor a workplace field says.

Source. AITraining.jobs Research, Real-World Data Collection Work, October 2026 edition.

Figure 4

Where the collection work is, by country

In which countries are platforms recruiting people to collect real-world data?

CountryOpen rolesOn-site thereMedian advertised (per hour)
United States
38
15$15
India
23
10$5
Brazil
17
0$7
Colombia
16
0$7
Mexico
16
0$7
Argentina
15
0$7
Chile
15
0$7
Canada
14
0$10
Peru
14
0$7
Indonesia
12
0$5
Vietnam
11
0$5
Philippines
10
0$5
United Arab Emirates
5
0n/a
Australia
4
0n/a
Costa Rica
4
0$9
United Kingdom
4
4n/a
Kuwait
4
0n/a
Malaysia
4
0n/a
Oman
4
0n/a
Panama
4
0$9
Saudi Arabia
4
0n/a
Uruguay
4
0$9
Bolivia
3
0n/a
Dominican Republic
3
0n/a
Ecuador
3
0n/a
France
3
1n/a
Guatemala
3
0n/a
Honduras
3
0n/a
Jordan
3
0n/a
Nicaragua
3
0n/a
Open worldwide (no country named)
8
0$20

Sample. 107 real-world data collection roles open at the end of 31 Oct 2026; 70 countries named, the 30 named by the most roles shown.

A role counts in every country it is open to: the country list its job board publishes, else the countries its location or title names (a site, 'Remote - Japan', '(US-based)'). A role open in several countries counts once in each, so the rows add up to more than the roles.

Open worldwide: the board says anyone may apply and the posting names no country.

A median is the job board's, shown where the country's roles state at least 3 distinct hourly rates.

Source. AITraining.jobs Research, Real-World Data Collection Work, October 2026 edition.

Figure 5

Cities where collection work is located

Which cities have real-world data collection roles, and what do they pay?

City or metro areaCountryOpen rolesMedian advertised (per hour)Session fee stated in the title
PuneIndia
7
n/a$55 for 2.5 hours ($22 an hour)
Los AngelesUnited States
7
n/an/a
San Francisco Bay AreaUnited States
5
n/an/a
LondonUnited Kingdom
4
n/a$170 for 2.5 hours ($68 an hour)
HyderabadIndia
3
n/a$55 for 2.5 hours ($22 an hour)
AmmanJordan
3
n/an/a
DubaiUnited Arab Emirates
3
n/an/a
BergenNorway
2
n/an/a
AtlantaUnited States
1
n/an/a
BarcelonaSpain
1
n/an/a
ParisFrance
1
n/a$160 for 2.5 hours ($64 an hour)
San DiegoUnited States
1
n/an/a
Seattle areaUnited States
1
n/an/a
SingaporeSingapore
1
n/an/a
Hong KongHong Kong
1
n/an/a
JobIndia
1
n/an/a
New York CityUnited States
1
n/an/a

Sample. 41 of the 107 real-world data collection roles open at the end of 31 Oct 2026 name a city in their location; the 25 cities named most.

Read from the location the posting publishes, with the on-site report's reader: cities of one metropolitan area are merged (Sunnyvale and San Francisco count as the San Francisco Bay Area). A role open at several sites counts once in each city.

A median is the job board's, shown where the city's roles state at least 3 distinct hourly rates. Some in-person session calls state a fee for a few hours in the title instead of an hourly rate; it is shown with its hourly equivalent (the fee over the hours stated) and is not in the medians.

A city named only in the title (some session calls leave the location empty) is not placed here; its country still counts in the country figure when the board lists it.

Source. AITraining.jobs Research, Real-World Data Collection Work, October 2026 edition.

Figure 6

What real-world data collection advertises per hour

How much do platforms advertise per hour to people who record data or operate robots?

Advertised hourly rateRolesShare of roles stating a rate
Under $1024
40.7%
$10 to $14.9912
20.3%
$15 to $19.9912
20.3%
$20 to $29.995
8.5%
$30 to $49.993
5.1%
$50 and more3
5.1%

Sample. 59 of the 107 real-world data collection roles open at the end of 31 Oct 2026 state an hourly rate (median $10, typical pay $7 to $30).

Advertised pay: the hourly figure a posting states, the middle of a stated range. A fee per session, per video or per project is not counted here; session fees stated in titles are shown in the cities figure.

Platforms often set a recording role's rate per country or region, so a band reflects where roles are open as well as the work.

Source. AITraining.jobs Research, Real-World Data Collection Work, October 2026 edition.

Figure 7

New real-world data collection listings, week by week

Are platforms listing more real-world data collection work, or less?

WeekNew listings (fixed panel)New listings (all tracked platforms)
1 to 4 OctPartial week
4
4
5 to 11 Oct
2
2
12 to 18 Oct
5
5
19 to 25 Oct
7
7
26 to 31 OctPartial week
13
14

Sample. Real-world data collection roles first listed in each ISO week (Monday to Sunday, UTC), 1 Oct 2026 to 31 Oct 2026. The fixed panel is the 18 platforms tracked since 1 Oct 2026.

Like for like: platforms first tracked later are counted only in the all-platforms columns, so their arrival never reads as more collection work. Each platform's first week (its back catalogue) never counts as new listings.

The same title rules as the rest of the report, applied to every role listed in the span.

Source. AITraining.jobs Research, Real-World Data Collection Work, October 2026 edition.

Figure 8

AI labs hiring robot data collectors directly

Which robotics and AI labs recruit their own robot pilots, operators and data creators instead of going through a vendor?

No data in this window.

Sample. The public job boards of 0 AI labs read on 1 Nov 2026; labs with at least one robot data-collection role open.

A role counts when the AI labs hiring report classifies it as robot data collection: its title names a data-collection operator, teleoperation, a robot pilot or data creators. Robot operators whose title does not say they collect data are not counted, so this is a floor.

Stated pay is the range the posting publishes, as an hourly rate or a yearly salary; 'Not stated' where none is published.

Source. AITraining.jobs Research, Real-World Data Collection Work, October 2026 edition.

Figure 9

Every robot data-collection role open on the labs' own boards

Which roles exactly, where, and at what stated pay?

No data in this window.

Sample. The 0 robot data-collection roles open on 1 Nov 2026, each linked to its posting on the lab's own board.

The posted date is the board's first-published date, or its last update where that is all the board gives.

Source. AITraining.jobs Research, Real-World Data Collection Work, October 2026 edition.

Figure 10

What is counted, against the kinds of work

How does this count relate to the robotics and physical-AI data kind of work, and to other kinds?

RolesCounted hereOpen rolesMedian advertised (per hour)
Robotics & Physical-AI Data: recording, collecting or operatingYes
65
$10
Robotics & Physical-AI Data: labelling, editing or other work that records nothingNo
35
$35
Data Annotation & Labeling: recording or collecting in the physical worldYes
13
$30
Multilingual & Localization: recording or collecting in the physical worldYes
13
n/a
Transcription, Audio & Speech: recording or collecting in the physical worldYes
9
n/a
Research Studies & Surveys: recording or collecting in the physical worldYes
4
n/a
Math & STEM: recording or collecting in the physical worldYes
2
n/a
AI Evaluation & Rating (RLHF): recording or collecting in the physical worldYes
1
n/a
All real-world data collection (this report)Yes
107
$10

Sample. Roles open at the end of 31 Oct 2026.

The robotics and physical-AI data kind of work also holds roles that label, review or edit recorded robot, driving and egocentric footage, and expert roles whose title records nothing; they are physical-AI data work but not collection, so they are left out and shown here for comparison.

Recording roles in other kinds of work (a video collector listed under data labelling, an in-person speech session under audio) are counted by their titles.

Source. AITraining.jobs Research, Real-World Data Collection Work, October 2026 edition.

For labs, data vendors and researchers

Buy the data behind this report

The figures above are a summary. Behind them are records we collect ourselves, every day:

  • 11,482 roles tracked, each with its history

    Every role on 62 job boards, read since 13 Jul 2026 (every hour since 18 Oct 2026): its pay and place, and the first and last day it was open.

  • 58,632 apply clicks in 28 days

    Which roles people actually apply to, by field, pay and kind of work. It happens on this site, so nobody else can count it.

  • 39 lab and vendor hiring boards

    The job boards of 25 AI labs and 14 human-data vendors, read every day: who is hiring for human data, and who is hiring to buy it.

What you can buy

  • The records

    The rows behind any figure, with their history, not only the totals.

  • A regular feed

    The same figures every week or month, as files.

  • A custom cut or benchmark

    Your countries, languages or fields, or your own platform against the market.

Building AI and need the experts behind this data? (you contract them yourself) or (we deliver the expert work).

Investing in AI-data companies? Hiring radar is these records role by role, every day, as files.

Method and definitions

The public job boards of the AI training platforms this site tracks, read every hour (each open role's title, location, workplace field, country list and advertised pay), and the public job boards of AI labs with a documented board API, read on the day the report is computed. Every figure is computed by one database query over the window shown, the same query for the live report and for every edition. Each figure states its sample beneath it.

Real-world data collection role
An open role whose title records, collects or operates in the physical world: it teleoperates or operates a robot; it records, collects or captures data and is listed in the robotics and physical-AI data kind of work, or names a camera medium (video, photos, faces, 360-degree, egocentric, motion) in any other kind; or it holds speech or other collection sessions in person. A title that labels, reviews, evaluates, transcribes or edits recorded data, or works on screens, software, files, documents or cards, never counts. Speech recorded at home is speech data and is not counted.
Kind of collection
Read from the title, the first match in this order: teleoperation and robot operation; egocentric (first-person) video; household and at-home tasks; motion capture and movement; speech and voice recorded in person; general video and photo capture (360-degree, car-mounted, smartphone, exocentric and third-person video, photographs, faces); other and unspecified collection.
On-site or hybrid
The title or location names a site ('Onsite', 'In-person', 'Hybrid'), a job board that publishes a workplace field (Lever, Ashby, Mercor's board) marks the role not remote, or the title calls speakers to a place ('Speakers Needed in Pune'). Remote: the title, location or board says remote; for collection work this usually means recorded in the contributor's own home, street or car.
Where a role is open
The country list its job board publishes, else the countries its location or title names; a role open in several countries counts in each. Cities come from the location text, with the cities of one metropolitan area merged.
Advertised pay
The hourly figure a posting states (the middle of a stated range). A median is published where at least 3 roles state a rate (the one minimum of every median).
Session fee
A fee for one in-person session stated in the title ('$170 USD for 2,5 hours'), shown with its hourly equivalent (the fee over the hours). Fees are not mixed into the hourly medians.
Fixed panel
The platforms already tracked when the 12-week span opened (or by the end of its first week): new listings are counted on them so a platform added later never reads as growth. A platform's first week is its back catalogue and is never counted as new listings.
Lab robot data-collection role
A role on an AI lab's own job board that the AI labs hiring report classifies as robot data collection: its title names a data-collection operator, teleoperation, a robot pilot or data creators. Each is linked to its posting.
Open role
A role listed by a given day and still in its platform's feed that day.

What this data cannot tell you

  • Roles are read from their titles: a collection role whose title names no capture, robot or session is missed, and a title can describe the work loosely. Every rule is published and tested on real titles.
  • Platforms that recruit collection contributors without a public job board, and AI labs and robotics companies without a documented board API, are not counted.
  • The labs' figure counts only titles that say they collect data; some labs' generic robot-operator roles collect data by teleoperation according to their descriptions but are not counted, so it is a floor.
  • Listings are not hires: a platform posting one role per country or language counts each posting, and a role open in many countries counts in each country.
  • Pay is what postings advertise, not what contributors earn; in-person session calls often state a fee rather than an hourly rate, and a posting can state a figure above $100 an hour that reads like a one-off fee (kept as published; medians are not moved by it).
  • The labs' boards are read on the day the report is computed, so a monthly edition shows the labs' roles open when it was frozen, not over the month.

How the job feeds are read and roles are classified: site methodology. How fresh, complete and consistent the data is, checked every day: data quality and coverage.

Editions

The live report covers the last 28 complete days and changes every day. Each complete month is also published as an edition, frozen two days after the month ends, so a figure you cite from an edition never moves.

  • Live dataLast 28 complete days, recomputed daily
  • October 2026 edition (this page)Freezes on first reading

Cite and reuse

Free to quote, chart and republish under CC BY 4.0 with a link to this page. This edition is frozen, so its figures will read the same later.

Citation

AITraining.jobs (2026). Real-World Data Collection Work, October 2026 edition. https://aitraining.jobs/research/real-world-data-collection-work/2026-10

BibTeX

@misc{aitrainingjobs_real_world_data_collection_work_2026_10,
  author = {{AITraining.jobs}},
  title = {Real-World Data Collection Work, October 2026 edition},
  year = {2026},
  howpublished = {\url{https://aitraining.jobs/research/real-world-data-collection-work/2026-10}},
}
Data: every figure as CSV (one line per value) and JSON, sent to your work email.Pay and open-roles data

More research

All 81 research reports