■ CRITICAL RISK ■ Science & Research
Substantially, yes — the craft is being squeezed between DIY survey platforms that automate the mechanics and passive data that answers questions nobody had to ask. Human researchers persist where rigor is legally or scientifically non-negotiable.
“AI analyzes sentiment without needing to design a questionnaire.”
Our AI replacement risk score — how we score jobs
Survey researchers design questionnaires, draw samples, field studies, clean data, and translate responses into findings — for polling firms, market research agencies, government statistical programs, and academia. The craft is subtler than outsiders assume: question wording that doesn't lead, sampling frames that don't skew, weighting that repairs who actually answered versus who was supposed to. The industry's existential problem predates AI: response rates have collapsed for decades, making every survey costlier and shakier — the profession was already treating a chronic illness when automation arrived.
Automation now hits every stage at once. Platform tools generate questionnaires, program logic, field panels, and produce dashboards — work that once justified a research team now ships as a SaaS subscription any product manager can drive. Large language models draft and translate instruments, code open-ended responses in minutes (previously an army-of-interns task), and summarize findings passably. More fundamentally, passive and behavioral data substitutes for asking: why survey purchase intent when transaction streams, app analytics, and review sentiment show behavior directly? Text-analysis of social platforms delivers rough opinion reads without a single questionnaire, and clients increasingly accept rough over rigorous when rough is 90% cheaper.
The resistant territory is where being wrong has consequences and rigor is the product: official statistics, election polling (battered but irreplaceable), regulated pharmaceutical and public-health research, and litigation surveys where methodology gets cross-examined. Measuring attitudes and reasons — the why behind the behavioral data — still genuinely requires asking humans well, and sampling expertise matters more, not less, as response rates fall. But that's a specialist stratum atop a commodity market that no longer needs many researchers. Our risk score of 87 reflects a middle being hollowed out from both ends.
Automatability: our editorial assessment of current and near-term AI capability
The compression is underway: DIY platforms and LLM-assisted analysis have already commoditized routine studies, and passive-data substitution grows every year. Through the late 2020s, generalist survey roles at agencies keep contracting while demand concentrates in methodological specialists — sampling statisticians, questionnaire scientists, and researchers in regulated domains. The profession shrinks to its rigorous core rather than disappearing.
Shrinking and polarizing rather than dying. Routine market-research studies are being absorbed by DIY platforms, AI analysis, and passive behavioral data, hollowing out generalist roles. Meanwhile the hard core — sampling statistics, official surveys, regulated research, polling methodology — still needs experts, arguably more urgently as response rates keep falling. The safe ground is technical depth; the exposed ground is running routine studies.
It writes a plausible one instantly, which is exactly the trap. LLMs produce clean-sounding questions but don't reliably catch leading wording, order effects, double-barreled items, or whether the construct being measured is the one the client needs. Used as a drafting assistant under an expert's review, it's a genuine time-saver; used unsupervised, it mass-produces professional-looking bias.
Behavioral data, mostly. Transactions, app analytics, search trends, and review or social-media text answer many questions surveys used to ask — measuring what people do rather than what they claim. The catch: passive data shows behavior, not reasons, and its samples skew in ways nobody controls. Attitudes, motivations, and populations without digital exhaust still require asking — which is the surviving case for the craft.
Toward the statistics or toward the synthesis. Path one: become a sampling and measurement specialist — scarce, technical, and needed wherever rigor is mandatory. Path two: become the insights lead who combines survey, behavioral, and AI-analyzed text data into decisions, using the machines rather than racing them. The role to escape is the middle one: fielding routine studies a platform now does end-to-end.