Microtasking (part 1): AI training and data tasks
Microtasks are small, well-defined online tasks that computers can't do reliably on their own: labelling images, rating AI answers, checking search results, recording speech. Companies building AI systems, search engines, maps and online shops pay people around the world to do them, usually through task platforms.
This lesson explains each task type in detail. Part 2 covers surveys, testing, research and local tasks.
Before you start: what all microtask work has in common
- Qualification tests are normal: you read guidelines and pass a test before getting paid work.
- Quality is checked constantly (hidden test questions, reviewers). Accuracy matters more than speed.
- Work availability goes up and down. Projects open and close; some weeks have no tasks.
- Country eligibility changes. Platforms open or close to Kenyan workers; Remotasks left Kenya in 2024. Never rely on one platform.
- Payment is usually by PayPal, Payoneer, bank transfer or the platform's own method, weekly or monthly.
- Legitimate platforms never charge you to join. Accounts for sale, "verified account rentals" and "training fees" are scams or rule-breaking.
Examples of platforms in this space include Appen (CrowdGen), TELUS Digital (AI community), Toloka, Clickworker, OneForma, DataAnnotation, Outlier and others; availability for Kenyan residents changes, so check each site's eligible countries yourself.
See also: AI training and data work overview.
1. Image tagging and classification
What it is: Looking at images and choosing the correct labels: "Does this photo contain a vehicle?", "Is this product image blurry?", "Which category does this item belong to?" How it works: You get batches of images with a list of labels and detailed rules. Each image takes seconds to a minute. Hidden "gold" questions check your accuracy.
| Skills | Attention to detail, following written guidelines, basic English |
| Cost & equipment | Laptop or smartphone (some platforms), stable internet |
| Where to find work | Data-labelling and crowd-work platforms |
| Payment | Per task or per hour; PayPal, Payoneer or bank |
| Realistic earnings | Very low to low; each task pays little |
| Time to first income | 1–4 weeks (sign-up, tests, payment cycle) |
How to start: Sign up on 2–3 platforms open to Kenya; complete your profile honestly; study guidelines slowly; pass the qualification; start with small batches and prioritise accuracy. Pros: Easy to learn, flexible hours, a good first step into AI data work. Cons & risks: Low pay, repetitive, irregular availability, accounts can be suspended for low accuracy. Scam warning: Nobody needs to "unlock" image tasks for a fee. Kenya notes: Kenya has a large data-annotation workforce; build accuracy records to move into better-paid projects (methods 2–4, 9).
2. Bounding box annotation
What it is: Drawing rectangles around objects in images (cars, pedestrians, products, road signs) to train computer-vision systems such as self-driving cars and shop-shelf scanners. How it works: In a web tool, you draw tight boxes around every object of the required types and choose labels. Rules cover overlapping objects, partly hidden items and minimum sizes.
| Skills | Precision with a mouse, patience, strict rule-following |
| Cost & equipment | Laptop with a mouse (a phone isn't practical), good screen, stable internet |
| Where to find work | Annotation platforms and managed annotation companies (some hire teams in Nairobi and other cities) |
| Payment | Per image/object or hourly |
| Realistic earnings | Low; can rise to low–medium on long-term managed projects |
| Time to first income | 2–6 weeks |
How to start: Practise with free annotation tools (e.g. open-source labelling software) to build speed and precision; apply to platforms and annotation firms; keep a record of projects and accuracy scores. Pros: Skills transfer to better annotation work; some companies offer contracts. Cons & risks: Eye strain, strict quality thresholds, project-based work. Scam warning: Fake "annotation companies" that charge for "training" or "laptops". Kenya notes: Some AI data companies operate offices in Kenya; check reviews and worker experiences before joining.
3. Polygon and semantic segmentation
What it is: Outlining the exact shape of objects point by point (polygons), or colouring every pixel of an image by category (road, sky, person, building). How it works: More detailed than bounding boxes; each image can take several minutes or longer. Used for medical images, satellite imagery, autonomous driving and agriculture.
| Skills | High precision, patience, good eyesight, domain knowledge for specialist projects (e.g. medical) |
| Cost & equipment | Laptop, mouse (or drawing tablet), large screen helps |
| Where to find work | Annotation platforms, specialist data companies |
| Payment | Per image or hourly, often higher than simple tagging |
| Realistic earnings | Low to low–medium |
| Time to first income | 2–6 weeks |
How to start: Learn the basics with free tools; start with bounding boxes, then apply for segmentation projects when available. Pros: Better pay per task than simple tagging; valued skill. Cons & risks: Slow, tiring, strict review. Scam warning: As above: no fees for training. Kenya notes: Agricultural and satellite-imagery projects sometimes value local knowledge (crops, land use).
4. Video annotation and object tracking
What it is: Labelling objects frame by frame across a video, tracking the same object over time, or marking events ("person enters", "car turns"). How it works: Tools interpolate between keyframes; you correct the boxes and labels as objects move.
| Skills | Precision, concentration over long sessions |
| Cost & equipment | Laptop, mouse, fast and stable internet (video files) |
| Where to find work | Annotation platforms and companies |
| Payment | Per video/minute or hourly |
| Realistic earnings | Low to low–medium |
| Time to first income | 2–6 weeks |
How to start: Build image annotation experience first; mention it when applying for video projects. Pros: Often longer projects. Cons & risks: Data-heavy (internet costs), tiring. Scam warning: Requests to pay for "high-speed annotation software". Kenya notes: Budget for data bundles or a reliable home internet connection.
5. Text classification and sentiment labelling
What it is: Reading short texts (reviews, tweets, customer messages) and labelling their topic, intent or sentiment (positive, negative, neutral), or flagging spam. How it works: You read each text and choose labels according to guidelines; some tasks ask for short explanations.
| Skills | Strong reading comprehension; for some projects, Kiswahili or other languages |
| Cost & equipment | Laptop or phone, internet |
| Where to find work | Crowd-work and AI data platforms |
| Payment | Per item or hourly |
| Realistic earnings | Very low to low |
| Time to first income | 1–4 weeks |
How to start: Apply to platforms; highlight languages you speak fluently (language projects can be less crowded). Pros: Simple, phone-friendly on some platforms. Cons & risks: Low pay; you may see offensive content. Scam warning: "Pay to get Kiswahili text tasks" offers are scams. Kenya notes: Kiswahili, Sheng and other African-language data is increasingly in demand for AI; mention your languages.
6. Named entity recognition (NER) and text annotation
What it is: Highlighting names of people, places, organisations, dates, products or medical terms in text, and marking relationships between them. How it works: You select words or phrases and assign tags; detailed rules handle tricky cases.
| Skills | Grammar knowledge, careful reading, domain knowledge for legal/medical projects |
| Cost & equipment | Laptop, internet |
| Where to find work | AI data platforms, annotation companies, language-data firms |
| Payment | Per document or hourly |
| Realistic earnings | Low; higher for specialist (legal, medical) projects |
| Time to first income | 2–6 weeks |
How to start: Learn basic linguistics terms (nouns, entities); pass qualification tests carefully. Pros: More interesting than image tasks for language lovers. Cons & risks: Demanding accuracy standards. Scam warning: As above. Kenya notes: Professionals (nurses, lawyers, accountants) can qualify for better-paid domain projects.
7. Search engine evaluation (search quality rater)
What it is: Judging whether search results and web pages are useful, trustworthy and relevant to a query, following long published guidelines, to help improve search engines. How it works: Usually part-time contract work (often 10–20 hours a week) through vendors; you pass a demanding exam on the guidelines, then rate tasks in a web tool.
| Skills | Excellent English, internet research skills, cultural knowledge of your market, patience with long guidelines |
| Cost & equipment | Laptop/desktop, reliable internet |
| Where to find work | Vendors that recruit raters by country (programmes and eligible countries change) |
| Payment | Hourly, monthly, by bank or other methods |
| Realistic earnings | Low–medium for part-time hours, when available in your country |
| Time to first income | 1–3 months (application, exam, onboarding) |
How to start: Watch for openings for Kenya or East Africa; study the publicly available search quality rater guidelines before the exam. Pros: More stable than pure microtasks when available. Cons & risks: Limited openings, demanding exams, contracts can end. Scam warning: Fake recruiters asking for exam "processing fees". Kenya notes: Local-market rater roles depend on demand for Kenyan search results; openings are occasional.
8. Ads quality and relevance rating
What it is: Rating whether online adverts match search queries or user interests, whether landing pages are trustworthy, and whether ads break policies. How it works: Similar to search evaluation: guidelines, exam, then rating tasks.
| Skills | Good English, judgement, understanding of online shopping and advertising |
| Cost & equipment | Laptop, internet |
| Where to find work | Rating vendors and crowd-work platforms |
| Payment | Hourly or per task |
| Realistic earnings | Low to low–medium when available |
| Time to first income | 1–2 months |
How to start: Apply when projects open; study advertising basics (ads and analytics). Pros: Builds marketing understanding. Cons & risks: Project availability varies. Scam warning: As above. Kenya notes: Knowledge of Kenyan brands and consumer habits helps for local-market projects.
9. AI chatbot response rating (human feedback / RLHF)
What it is: Comparing and rating answers from AI chatbots for accuracy, helpfulness, safety and style, often choosing the better of two responses and explaining why. This feedback trains AI models. How it works: You read a prompt and AI responses, check facts, rate on several criteria and write short justifications.
| Skills | Excellent English writing, fact-checking, critical thinking; specialist knowledge (maths, science, law, medicine, coding) pays more |
| Cost & equipment | Laptop, internet |
| Where to find work | AI training platforms that hire "AI trainers" or "raters" |
| Payment | Hourly or per task, often weekly via PayPal or other methods |
| Realistic earnings | Low to medium; higher for experts and strong writers when projects are available |
| Time to first income | 1–4 weeks after passing assessments |
How to start: Prepare for writing and reasoning assessments; read the guidelines carefully; be honest about your expertise areas. Pros: Among the better-paid microtask types; interesting work. Cons & risks: Very strict quality checks; projects end suddenly; assessments are tough. Scam warning: Telegram/WhatsApp "AI training jobs" requiring deposits are scams. Kenya notes: University graduates and professionals are well placed; a strong command of written English is essential.
10. AI prompt and response writing
What it is: Writing prompts (questions, instructions, scenarios) and ideal responses that teach AI how to answer well, or writing creative and challenging tasks to test AI models. How it works: Projects specify topics, styles and difficulty; reviewers check originality and quality.
| Skills | Excellent writing, creativity, subject knowledge, following style guides |
| Cost & equipment | Laptop, internet |
| Where to find work | AI training platforms |
| Payment | Hourly or per task |
| Realistic earnings | Low to medium, depending on skill and project |
| Time to first income | 1–4 weeks |
How to start: Build writing samples (writing work); learn good prompting (prompting). Pros: Rewards writing talent. Cons & risks: Using AI to generate your submissions usually breaks rules and leads to bans. Scam warning: As above. Kenya notes: Kenyan English writers are competitive; careful, original work matters.
11. AI coding evaluation and code writing
What it is: Evaluating AI-generated code, writing correct solutions and explanations, and finding bugs, in languages like Python, JavaScript, SQL and others. How it works: Tasks include rating code correctness and style, writing test cases, and solving programming problems.
| Skills | Solid programming in at least one language, debugging, clear explanations |
| Cost & equipment | Laptop, internet |
| Where to find work | AI training platforms recruiting coders |
| Payment | Hourly, often higher than general tasks |
| Realistic earnings | Low–medium to medium for skilled developers when projects are available |
| Time to first income | 1–4 weeks after coding assessments |
How to start: Strengthen fundamentals (Python, JavaScript, algorithms); pass coding tests. Pros: Good use of coding skills while building a career. Cons & risks: Demanding tests; work availability varies. Scam warning: As above. Kenya notes: A good option for computer science students and junior developers.
12. Speech data recording (voice samples)
What it is: Recording yourself reading sentences, answering questions or having conversations, to train speech recognition and voice assistants in different languages and accents. How it works: An app or website shows prompts; you record in a quiet place; recordings are checked for quality.
| Skills | Clear speech, fluency in the required language or accent |
| Cost & equipment | Smartphone or laptop with a decent microphone, quiet room |
| Where to find work | Speech-data projects on crowd platforms and language-data companies |
| Payment | Per recording, per hour of audio, or per project |
| Realistic earnings | Very low to low; one-off projects |
| Time to first income | 1–4 weeks |
How to start: Sign up for language-data platforms; list all your languages (including Kiswahili, Kikuyu, Dholuo, Luhya, Kalenjin and others). Pros: Phone-friendly; values local languages. Cons & risks: Sporadic projects; read consent terms on how your voice will be used. Scam warning: Requests for your ID and money to "register for voice projects". Kenya notes: African-language speech data is in growing demand.
13. Short audio transcription and speech annotation
What it is: Typing short audio clips (seconds to a minute), correcting machine transcripts, or marking speakers, noise and emotions in audio. How it works: Listen, type exactly according to style rules, submit; accuracy is reviewed.
| Skills | Fast, accurate typing; good listening; grammar |
| Cost & equipment | Laptop, headphones, internet |
| Where to find work | Transcription and speech-data platforms |
| Payment | Per clip or per audio minute |
| Realistic earnings | Very low to low; better on full transcription work |
| Time to first income | 1–4 weeks |
How to start: Practise typing (typing skills); see transcription and audio work. Pros: Builds into full transcription and captioning work. Cons & risks: Hard audio, low per-clip pay. Scam warning: "Transcription jobs" requiring payment for software or tests. Kenya notes: Local-language transcription (Kiswahili and others) can be less crowded.
14. Translation and language evaluation microtasks
What it is: Translating short sentences, checking machine translations, rating translation quality, or adapting text for local audiences (localisation). How it works: Short tasks in a translation tool, with style guides and quality checks.
| Skills | Fluency in two or more languages, good writing in the target language |
| Cost & equipment | Laptop, internet |
| Where to find work | Language-data platforms, localisation companies |
| Payment | Per word, per task or hourly |
| Realistic earnings | Low; higher for professional translation |
| Time to first income | 2–6 weeks |
How to start: Take language tests on platforms; build towards professional translation work. Pros: Rewards language skills. Cons & risks: Machine translation reduces some demand. Scam warning: Fake translation agencies asking for "registration". Kenya notes: English–Kiswahili pairs, and French or Arabic, are useful.
15. Content moderation
What it is: Reviewing user posts, images, videos or comments and removing content that breaks platform rules (violence, hate, scams, adult content). How it works: Usually full-time or part-time roles through outsourcing companies, or smaller moderation tasks on crowd platforms.
| Skills | Judgement, knowledge of policies, emotional resilience, language skills |
| Cost & equipment | Often office-based; remote roles need a laptop and internet |
| Where to find work | Business process outsourcing (BPO) companies, crowd platforms, community managers for brands |
| Payment | Salary for employed roles; per task on platforms |
| Realistic earnings | Low to medium (employed roles offer steadier pay) |
| Time to first income | Weeks (hiring process) |
How to start: Apply to outsourcing firms and moderation projects; ask about wellbeing support before accepting. Pros: Employment opportunities exist. Cons & risks: Exposure to disturbing content can harm mental health; choose employers with proper support and breaks. Scam warning: Fake job offers requiring "medical" or "processing" fees. Kenya notes: Nairobi has hosted moderation operations; research employer conditions and worker experiences.
16. Data entry and e-commerce catalogue tasks
What it is: Entering or cleaning product information (names, prices, sizes, categories), matching identical products across shops, and categorising listings. How it works: Spreadsheet or web-tool tasks with specific formatting rules.
| Skills | Accurate typing, Excel/Sheets basics, attention to detail |
| Cost & equipment | Laptop, internet |
| Where to find work | Crowd platforms, freelance marketplaces, local businesses (shops moving online) |
| Payment | Per task, per hour or per project |
| Realistic earnings | Very low to low; better as part of virtual assistant or e-commerce support work |
| Time to first income | 1–4 weeks |
How to start: Learn Excel basics (Excel); offer catalogue setup to local online shops. Pros: Simple to start; local demand exists. Cons & risks: Very competitive and low-paid online. Scam warning: "Data entry jobs" that require buying a starter kit or software are a classic scam. Kenya notes: Kenyan businesses joining Jumia, Kilimall or building online shops often need catalogue help.
17. Receipt, document and OCR correction tasks
What it is: Reading scanned receipts, invoices or forms and typing or correcting the extracted data (store name, date, amounts) where software made mistakes. How it works: Side-by-side image and form; you fix errors per rules.
| Skills | Accuracy with numbers and text |
| Cost & equipment | Laptop or phone, internet |
| Where to find work | Crowd-work platforms, document-processing companies |
| Payment | Per document |
| Realistic earnings | Very low |
| Time to first income | 1–4 weeks |
How to start: Look for document-transcription projects on crowd platforms. Pros: Easy to learn. Cons & risks: Very low pay; handle personal data responsibly. Scam warning: As above. Kenya notes: Treat any personal data you see as confidential under the Data Protection Act.
18. Map and location data validation
What it is: Checking and correcting map information: business names, addresses, opening hours, road features, or verifying locations from street imagery. How it works: Desk tasks comparing sources, or field tasks visiting places (see part 2, method 28).
| Skills | Local knowledge, research skills, map reading |
| Cost & equipment | Laptop or smartphone, internet |
| Where to find work | Mapping and data companies' crowd projects |
| Payment | Per task or hourly |
| Realistic earnings | Very low to low |
| Time to first income | 2–6 weeks |
How to start: Learn digital mapping basics; volunteer mapping (e.g. OpenStreetMap community projects) builds skills and portfolio, though it's unpaid. Pros: Local knowledge is valuable. Cons & risks: Limited paid projects. Scam warning: As above. Kenya notes: Knowledge of Kenyan towns, estates and landmarks helps local mapping projects.
19. Web research and lead-list microtasks
What it is: Finding specific information online (company contacts, product details, event lists) and entering it into a spreadsheet to defined rules. How it works: Clients or platforms provide a list or criteria; you research and compile data.
| Skills | Search skills, Excel/Sheets, judging source reliability |
| Cost & equipment | Laptop, internet |
| Where to find work | Freelance marketplaces, crowd platforms, small businesses |
| Payment | Per record or per project |
| Realistic earnings | Very low to low; better as a virtual assistant service |
| Time to first income | 1–4 weeks |
How to start: Practise advanced Google search; build a sample research sheet; offer on freelance platforms (virtual assistance). Pros: Leads into virtual assistant and research careers. Cons & risks: Competitive; respect data-protection rules when collecting personal information. Scam warning: Clients asking you to collect data unlawfully (e.g. scraping personal phone numbers for spam). Kenya notes: Kenyan businesses need supplier and customer research; follow the Data Protection Act.
Summary
- AI and data microtasks include image, video and text labelling, search and ads rating, AI response rating and writing, coding evaluation, speech and transcription, translation, moderation, data entry and research.
- Earnings are mostly very low to low, rising with accuracy, specialist skills and better projects.
- Availability changes; join several legitimate platforms, never pay to join, and protect your accounts.
Check yourself
Drawing rectangles around objects in images is called what? (two words)
Show answer
bounding box
Rating which of two AI answers is better trains AI using human ___ (one word)
Show answer
feedback
Which equipment is essential for bounding-box work: a phone or a laptop with a mouse?
Show answer
laptop
Hidden test questions used to check accuracy are often called what? (one word)
Show answer
gold