SurveyMonkey analyzed over 25,000 surveys, each with more than 100 respondents. The finding that should worry anyone who sends forms: surveys with 10 open-ended questions have a 78% completion rate. Surveys with one open-ended question hit 88%. Each open-ended field you add actively repels the people you need answers from.
The average open-ended response is 5 words long and takes 15 seconds to write. “Fine.” “Nothing to change.” “Looks good.” These aren't answers. They're the respondent signaling they want to close the tab.
The same question · two cognitive modes
The form industry has spent two decades solving this problem from the wrong end. Typeform made forms beautiful. Google Forms made them free. SurveyMonkey built analytics dashboards. None of it changed what the respondent types into the box.
That's because the problem isn't the form. It's the cognitive task the form demands.
The recall problem
In 1885, Hermann Ebbinghaus published research showing that humans forget roughly 66% of new information within 24 hours. His forgetting curve has been replicated consistently for over a century, most recently in a 2015 study published in PLOS ONE. The shape of the curve hasn't changed: memory degrades fast, and what remains is lossy.
Every open-ended form field is a recall task. “What went well this sprint?” You're asking someone to reconstruct two weeks of work from memory, usually at the end of a long day. The tickets they closed, the PRs they reviewed, the Slack thread where the deploy failed — all of it sits in systems they could search. But the form gives them a blank textarea and asks them to synthesize on demand.
Compare that to recognition. Roger Shepard's 1967 study showed that recognition accuracy for previously seen items is 98%. Lionel Standing pushed it further in 1973 — participants who viewed 10,000 photographs for 5 seconds each still recognized 83% of them later. Recognition is easy. Recall is hard. The gap between them is enormous.
Recall · hard
Review · easy
This is why pre-filled answers change everything. You aren't asking the respondent to recall anymore. You're asking them to recognize and verify. Review a draft, fix what's wrong, submit. The cognitive load drops from “reconstruct your week” to “read two paragraphs.”
Satisficing: the real reason your data is bad
In 1991, Jon Krosnick at Stanford published a theory that explains why survey data degrades even when people do respond. He called it satisficing — a portmanteau of “satisfy” and “suffice.”
Krosnick's model describes a spectrum. At one end, respondents optimize: they carefully retrieve information, form a judgment, and map it to an answer. At the other end, they strongly satisfice: they skip retrieval entirely and pick the first plausible option. In the middle is weak satisficing, where the respondent goes through the motions but settles for “good enough.”
Krosnick's satisficing spectrum · Stanford, 1991
Up to 22% of survey responses are straight-lined (Survey Research Methods, 2020). Fatigue increases “don't know” responses by 18%.
Satisficing isn't laziness. It's a rational response to cognitive load. Tourangeau, Rips, and Rasinski laid out the full model in The Psychology of Survey Response (2000): every survey question requires four cognitive steps — comprehend the question, retrieve information, form a judgment, map it to a response. When any step is effortful, people cut corners.
A 2020 study in Survey Research Methods measured the damage: up to 22% of survey responses are straight-lined, where the respondent selects the same option for every row. Removing those responses shifted regression coefficients by 13-39%. The data wasn't just noisy — it was directionally misleading.
Fatigue compounds the problem. Move the same questions from the beginning to the end of a survey and neutral/“don't know” responses rise 18%. The survey didn't change. The respondent's willingness to think did.
The form industry optimized the wrong side
Every major form tool has invested in the creator experience. Drag-and-drop builders. Conditional logic. Branded themes. Integration marketplaces. These are features for the person sending the form.
The person filling the form still gets a blank textarea and a deadline.
This asymmetry explains why completion rates haven't improved despite twenty years of product innovation. The average form abandonment rate is 67%. People who start a form abandon it after an average of 1 minute and 43 seconds. These numbers haven't meaningfully moved.
The constraint was never the form's design. It was the respondent's cognitive budget. You can make a textarea more beautiful, but you can't make recall less tiring.
What changes when AI does the recall
When a respondent hands a form to their AI assistant — Claude, ChatGPT, a custom agent — the AI can synthesize context the respondent already has: tickets, docs, messages, code. It drafts answers grounded in what actually happened, with dates and specifics. “Shipped the auth migration Tuesday, zero rollbacks. Reused 4 components from the new design library.”
The respondent reads the draft. Tweaks one line. Submits. Ten seconds.
The improvement isn't speed alone — it's that the cognitive task changed. Recall (generate from memory) became review (verify and edit). Shepard's 98% recognition accuracy applies here: people are exceptionally good at spotting what's wrong in a draft they didn't write, as long as it's about their own work.
The respondent still owns the answer. They read every word. They correct inaccuracies. But they don't stare at a blank textarea trying to reconstruct their week.
That's what Pluck does. One URL — respondents can fill it in a browser or hand it to their AI. The form has a native JSON interface, so agents don't need a browser extension or screen scraper. They read a structured spec, draft answers from context, and submit. The respondent reviews and approves.
The result: responses with dates, names, and specifics instead of “fine.” Not because people tried harder — because the form stopped asking them to do something humans are bad at.