Survey programming — also called survey scripting — is the work of building skip logic, quotas, rotations, and piping into a questionnaire: traditionally the slowest, most specialized part of getting a study into field. Here's what it actually involves, what it costs today, and how an AI agent now does it in minutes instead of days.
By the Wavefield Research team · Published Jul 28, 2026 · Updated Aug 9, 2026
A questionnaire document — the version a client signs off on — is just text. Scripting is the step that turns it into a survey that actually runs: every branch, every counter, every rotation has to be programmed by hand or generated correctly. That typically covers:
Scripting assumes the questionnaire itself is already sound — picking between multiple choice, Likert scales, rankings, and open-ends is its own craft, covered in survey question types, with examples.
Swipe the table sideways to see all columns →
| A programmer scripts it | Wavefield's agent | |
|---|---|---|
| Turnaround | Days to weeks, depending on programmer or agency availability and queue | Minutes — upload a brief, the agent programs it in the same session |
| Cost | Billed separately — commonly a few hundred to several thousand dollars depending on complexity | Included in the $99–$299 project price, no separate scripting fee |
| Skill required | A dedicated survey programmer fluent in the platform's scripting language | None — write a plain-English brief or upload an existing questionnaire document |
| Making changes | Change requests go back through the programmer or agency's queue | Chat with the agent or edit directly in the visual editor — same document, no re-briefing |
| What's supported | Depends on the platform and the programmer's skill — logic, quotas, and piping vary by shop | Skip logic, quotas, randomization, piping, and translations generated directly from the brief |
Generation runs in passes rather than one giant request: first an outline of blocks and question counts, then the real questions and logic generated block by block, then quotas and embedded data as a final pass. Every output is validated against the same schema the rest of the platform enforces — if something doesn't parse, the agent retries with the validation error until it does.
Translating a scripted survey is where budgets quietly double. The naive approach — copy the survey per language and translate each copy — forks the logic: fix a skip pattern in the English version and the Spanish copy silently keeps the bug, and at analysis time the same question carries different ids per language, so merging the data becomes its own project.
The approach that survives contact with reality is an overlay: one master survey holds all structure — questions, logic, quotas, ids — and each language is a set of text substitutions applied at render time. Ids never change, so logic, piping, and analysis are language-free, and a questionnaire fix lands in every language at once. That's how Wavefield implements translations, and it's worth demanding from any platform: ask specifically whether fixing logic after translation requires touching every language version.
A scripted survey is software, and it ships with bugs unless tested like software. The classic escapes: a skip pattern that strands respondents on an unreachable question, a screener that terminates the wrong arm, quotas counting screen-outs, and a rotation that never shows one choice. Every one of these has ruined real fieldwork — after the sample was paid for.
A note on the traditional tooling: dedicated scripting platforms — Forsta (Confirmit and Decipher), Voxco, and the scripting layers inside enterprise suites — remain the standard at large fieldwork agencies, operated by specialist programmers. They're genuinely powerful; the trade is that every change routes through a specialist's queue. The AI-agent model doesn't replace those shops at the high end — it removes the queue for the studies that never needed one.
Scripting (also called questionnaire programming) is the technical work of turning a written questionnaire into a functioning survey — skip and display logic, quotas, randomization and rotation, piping answers between questions, and translations. It's traditionally done by a dedicated survey programmer using a platform's own scripting language or tools like Confirmit, Decipher, or Voxco.
It varies by complexity and who does it — an in-house research-ops hire, a freelance programmer, or a full-service agency's scripting team. Quotes commonly range from a few hundred dollars for a short, simple questionnaire to several thousand for a long one with heavy branching, quotas, and multiple languages, often billed hourly or as part of a larger project fee.
Yes, within the same guardrails a human programmer would use. Wavefield's agent reads your brief or an existing Word-doc questionnaire and builds it in passes — first an outline, then the actual questions and logic block by block, then quotas and embedded data — validating every step against the same schema the platform enforces everywhere else, so it can't produce logic the platform itself couldn't run.
You fix it the same way you'd fix anything else in the platform — by chatting with the agent ("add a quota on region") or editing directly in the visual editor. Both write to the same underlying survey document, so nothing gets out of sync, and every change is versioned.
See the numbers a study needs with the sample size calculator, or how respondents actually get selected in quota sampling vs. random sampling. The programmed questionnaire still has to reach people — survey data collection methods, compared covers the channels.
Upload your brief or an existing questionnaire — the agent programs it the same session, no separate scripting fee.
Start a project