You will classify 225 short social-media posts in six manageable sessions. Every selection is timestamped and saved continuously in this browser.
Allow approximately 2.5–3.5 hours for careful work. Your four-hour coding window begins only after you pass the required practice. Each coding session is expected to take about 20–30 minutes.
We recommend a three-minute break between sessions. Before starting, keep some water nearby and use a laptop or desktop. Keep this browser and device for the entire study.
0 / 225posts completed
225posts remaining
0 / 6sessions completed
4:00:00submission time remaining
You cannot continue
You did not reach the practice requirement after two chances per question. Please follow the return/screen-out instruction shown here.
Step 1 — Read the guide, then practise
Practice contains seven fictional posts closely matched to the real task, with two questions per post. Each question has two chances. After a first incorrect choice, you receive a hint and try that question once more. After a second incorrect choice, the correct answer is explained. You must answer at least 11 of the 14 decisions correctly (78.6%). The Coding guide button stays available at the top of every practice and coding page.
Step 2 — Six coding sessions
Complete sessions in order. Each contains 37–38 real posts and one simple visual attention check. Take the recommended three-minute break after each session.
All sessions complete
Your complete CSV should download automatically. Email it to before the four-hour deadline.
Find the CSV in your Downloads folder. Use Download CSV again if needed.
Open your usual email website (such as Gmail or Outlook) or email app and start a new message to .
Attach the downloaded CSV and send it. Check that the message appears in your Sent folder before finishing the study.
If an email app stalls, try sending from your email website instead. The optional shortcut below opens a draft but cannot attach or send the CSV. Do not rename or edit the file.
Code what the post says or clearly implies—not your opinion. Do not infer an outcome from a hashtag, product category, or outside knowledge, and do not decide whether the claim is true.
For the first question, focus on the person or group speaking in the post (for example, “I,” “we,” or “our”). Treat that speaker as the human side even when they are a worker, artist, marketer, student, citizen, or organization. A headline, advertisement, feature list, or report about somebody else is normally neutral for the human question unless the speaker clearly describes or adopts the experience as their own.
Answer the human and AI questions separately. A post can describe a human experience without a clear AI outcome, or a clear AI outcome without a human experience.
Choose one answer for each question. If several signals appear, choose the outcome that is most explicit and central in the post.
You can reopen this guide from the top-right button at any time during practice and all six coding sessions.
1. What happens to the person or group speaking in the post?
Choose this explanation
Meaning
AI helps the speaker do something they already do
AI amplifies an existing task or ability: the speaker does it faster, more easily, better, or at greater scale. Example: using AI to write code faster or reach more clients.
AI gives the speaker a new ability, perspective, or possibility
The speaker gains a genuinely new capacity, perspective, learning opportunity, creative path, or possibility through AI.
AI reduces the speaker’s control, privacy, access, or freedom to act
The speaker’s autonomy or agency is limited through dependence, surveillance, lock-in, refusal, privacy risk, manipulation, or reduced control. Annoyance or guilt alone is not sustained damage to well-being, but it can be evidence of reduced freedom when the post describes pressure or manipulation.
AI diminishes the speaker’s skills, livelihood, creativity, thinking, identity, or well-being
The post reports an actual deterioration in an existing human capability or condition, such as lost work, weakened skills, poorer thinking, damaged identity, or sustained harm to well-being.
No clear AI-related change for the speaker
No enabling or constraining experience of the speaker is stated or clearly implied. Headlines, event announcements, product features, AI-art credits, and reports about other people can be neutral here.
2. How is the AI system, company, or ecosystem represented?
Choose this explanation
Meaning
AI works reliably or produces a useful result
The system performs reliably, produces a useful completed output, or otherwise serves the human–AI interaction well.
AI or its ecosystem gains data, learning, users, profit, adoption, or reach
The model, company, operator, or ecosystem benefits or grows—for example through added training data, learning, more users, subscriptions, profit, adoption, or market reach.
AI refuses, blocks, is limited by guardrails, or withholds a capability
The system or its operator is represented as censoring, refusing, gatekeeping, limiting access, imposing guardrails, or withholding a requested action or capability. This answer describes the AI-side action or limit; the resulting loss of human control can separately receive the human-control answer above.
AI fails, gives poor output, deteriorates, or is represented as damaged or exploited
The system hallucinates, is unreliable, degrades, produces poor or harmful output, or is represented as damaged or exploited as a disposable tool. Human harm alone does not establish this answer unless the post also represents AI output or performance as defective or harmful.
No clear role or outcome for AI or its ecosystem
AI may be mentioned, questioned, advertised, credited, or discussed, but no clear successful function, ecosystem growth, withheld capability, failure, or damaging outcome is represented.
Fast decision rules
Existing task improved → AI helps the speaker do something they already do. Genuinely new capacity or path → AI gives the speaker a new ability, perspective, or possibility.
Less autonomy, privacy, access, or choice → reduced human control. Actual deterioration in skill, livelihood, creativity, thinking, identity, or sustained well-being → diminished human capability or condition. When both appear, use the outcome that is most explicit and central.
A product or feature list alone normally gives no clear human experience. If the speaker explicitly says AI makes their existing work faster, easier, or better, choose the existing-task answer even when the post is promotional. A concrete successful function or output can still show that AI is represented as working.
“Smarter,” “better,” “faster,” or a useful completed output concerns system performance. Use the growth choice only when data, learning, profit, users, adoption, or reach is explicit and central.
“AI was tested” with no result, an event announcement, an unrelated #AI tag, or a general question about AI usually gives no clear system outcome.
AI restriction versus AI failure: choose restriction when the system or operator refuses, blocks, gatekeeps, applies guardrails, or withholds capability. Choose failure when the system attempts the task but is inaccurate, unreliable, degraded, or produces poor or harmful output.
For AI-generated content, “I created this with [AI tool]” can support an existing human activity and a successful AI output. An explicitly new creative path can support a new human possibility. A bare credit, repost, or hashtag alone may remain neutral.
Do not ask what AI wants, feels, or deserves. Judge only the action or outcome that the post attributes to the system, operator, or ecosystem.