/* global window */
// Content object for the article renderer (blog/blog-article.jsx).
// Produced by the Writer→Editor→Humanize→Proofreader chain; draft in blog/pipeline/drafts/where-do-i-click.md
window.__ARTICLE__ = {
slug: "where-do-i-click",
laneLabel: "FIELD NOTES",
kicker: "ONBOARDING",
readMins: 4,
dateLabel: "Apr 2026",
title: "The “where do I click” moment is more diagnostic than NPS",
deck: "The hesitation before a user finds the next step tells you more about churn than any satisfaction score. A short field note on behavioral signal over attitudinal scoring.",
tags: ["onboarding", "ux", "field-notes"],
toc: [
{ id: "erases", num: "01 · THE GAP", title: "What the survey erases" },
{ id: "latency", num: "02 · THE SIGNAL", title: "What latency tells you" },
{ id: "instrument", num: "03 · IN PRACTICE", title: "How to instrument it" },
],
body: [
{ t: "p", html: `There's a moment in almost every onboarding session that a survey will never catch. The screen loads. The participant goes quiet. Their cursor drifts, hovers over one thing, pulls back, drifts again. Two seconds pass. Four. Then, half to themselves: "So where do I... click?"` },
{ t: "p", html: `That pause is the most honest signal in the whole study. It's also exactly the thing a Net Promoter Score is built to miss.` },
{ t: "h2", id: "erases", num: "01 · THE GAP", text: "What the survey records, and what it erases" },
{ t: "p", html: `Ask that same person an hour later how the product felt and you might get a 7 out of 10. Ask "how likely are you to recommend this," the NPS prompt, and you might get a passive 8. Nothing in those numbers contains the four seconds of hesitation. The attitudinal score smooths it flat. The struggle gets recoded as a mild opinion, the mild opinion goes into a dashboard, and the dashboard says things are fine.` },
{ t: "p", html: `This isn't a knock on people being dishonest. It's how recalled attitude works. Nielsen Norman Group puts the gap plainly: they've "seen many users struggling to complete a task, yet rating a website as highly as someone who had no difficulty whatsoever" (Nielsen Norman Group, 2024). The satisfaction number and the behavior come apart, and when they do, the number is the one that lies.` },
{ t: "figure",
fig: { key: "transcript", props: { time: "00:14", speaker: "P-07 · ONBOARDING", children: `"It looks good, I like it... so where do I, um — where do I click to actually start one?"` } },
ref: "FIG 01",
caption: "The hesitation a satisfaction score never records. Four seconds of dead air, then the question itself." },
{ t: "pullquote", text: "A 7-out-of-10 with a four-second hover hides a worse story than a 5-out-of-10 that flowed." },
{ t: "p", html: `The same group ran the numbers on how tightly satisfaction tracks performance and found a correlation of only r = .53, moderate and real but far from the lock-step relationship most teams assume. About 30% of the time, satisfaction and actual task performance point in opposite directions: people perform badly and rate the design well, or sail through it and rate it poorly (Nielsen Norman Group, 2017). Three times in ten, the attitudinal score is actively pointing you the wrong way.` },
{ t: "p", html: `NPS in particular was never built for this job. It's a relationship-level, recalled-intent question. As NN/G notes, it "tells you how the experience is perceived but not why," and "it makes less sense to utilize NPS to assess users' satisfaction with granular details of UI design, such as a website's checkout process" (Nielsen Norman Group, 2024). The "where do I click" moment is granular. It lives at the level of one button on one screen. NPS can't see down there.` },
{ t: "h2", id: "latency", num: "02 · THE SIGNAL", text: "What the latency is actually telling you" },
{ t: "p", html: `Time-on-task is one of the oldest behavioral measures in usability work, and the reason it endures is simple: when someone takes much longer than expected on a step, that delay is usually the interface failing, not the person thinking deeply (Nielsen Norman Group, 2024). The gap between intent and action, where the participant clearly wants to proceed but can't find how, is a clean read on friction. No recall, no interpretation, no politeness filter. Just the cursor not knowing where to go.` },
{ t: "p", html: `Across our recent product studies, the pattern holds and sharpens. When decision-makers describe what they actually want from a research tool, the loudest theme isn't price and it isn't features. It's the demand to try before they trust: a risk-free proof-of-concept before they'll integrate anything new, the single strongest barrier we saw, present in roughly two-thirds of conversations. "Initially, we would like a demonstration to verify and see what type of data is collected and how it's collected," one participant said. That instinct, give me the thing in my hands before I commit, is the adult version of the same hesitation. People don't trust what they can't first try, and they don't proceed through what they can't first locate.` },
{ t: "p", html: `The studies also keep surfacing a quieter gap that mirrors the NPS problem exactly. In one client-experience study, every respondent rated their satisfaction high, tightly clustered at the good end of the scale. Read only that number and you'd change nothing. Read the transcripts and the dominant themes were all unmet needs: more proactivity, more anticipation, more of the next step surfaced before they had to ask for it. Satisfied today, quietly straining against the experience. The score said stay the course. The words said the opposite.` },
{ t: "pullquote", text: "Satisfaction tells you whether they'll say something nice. Latency tells you whether they'll come back." },
{ t: "h2", id: "instrument", num: "03 · IN PRACTICE", text: "How to instrument the moment" },
{ t: "p", html: `You don't need eye-tracking rigs or a lab. You need to stop throwing away the four seconds.` },
{ t: "p", html: `First, watch for the latency directly. In a moderated or AI-moderated session, the dead air between a screen loading and the first deliberate action is data. Mark it. A hover that pulls back is a behavioral signal; the spoken "where do I..." is a transcript moment; the same stall recurring across six participants on the same screen is a segment pattern. Those are three of the five evidence layers we anchor every claim to, and a survey gives you none of them.` },
{ t: "p", html: `Second, pair every attitudinal score with its behavioral shadow. NN/G's own guidance is to supplement satisfaction scores with task success rates and task times (Nielsen Norman Group, 2024). When a participant rates the flow an 8 but stalled for six seconds at the second step, you don't average those into "mostly fine." You keep both, and you trust the slower one.` },
{ t: "p", html: `Third, treat the hesitation as the headline, not the footnote. The instinct in most readouts is to lead with the score because it's a tidy number, then bury the struggle in an appendix clip nobody opens. Invert it. Lead with the clip of the cursor circling, the moment at 00:14 where the participant asks "where do I click," and let the satisfaction number play its real, smaller role: useful context, not the verdict.` },
{ t: "p", html: `A satisfaction score tells you how a person decided to feel about an experience after the fact, filtered through whatever they think you want to hear. The latency before they find the next step is the experience itself, caught in the act. One is a review. The other is the footage. When they disagree, and roughly a third of the time they will, play the footage.` },
{ t: "references", items: [
{ n: 1, html: `Nielsen Norman Group (2024). "Net Promoter Score: What a Customer-Relations Metric Can Tell You About Your User Experience." nngroup.com` },
{ n: 2, html: `Nielsen Norman Group (2017). "User Satisfaction vs. Performance Metrics." nngroup.com` },
] },
],
related: [
{ href: "/blog/five-flavors-friction.html", title: "Five flavors of UX friction agents catch before users do", meta: "6 min · Index" },
{ href: "/blog/what-counts-evidence.html", title: "What counts as evidence? A framework for AI research outputs", meta: "8 min · Index" },
],
};