AI tools like ChatGPT and Claude can generate a survey first draft in seconds when given a clear research objective. Nielsen Norman Group's finding is direct: generative AI handles foundational survey writing tasks reasonably well, but it introduces subtle design flaws that degrade data quality if no experienced human reviews the output.
The critical detail is in the gap between 'solid first draft' and 'research-ready instrument.' NNG identifies specific areas where AI underperforms on survey design, and those weaknesses are not obvious to researchers who lack survey methodology training. Speed without that judgment does not save time, it moves bad data faster.
The full article maps exactly where AI succeeds and where it fails across survey design dimensions. That breakdown is the reason to read it, not just the conclusion that human review is still required.
[READ ORIGINAL →]