AI Product Psychology

Engineering metrics pass. Why do users still call it slow?

Earlier chapters taught you to cut latency, squeeze cost, and drive hallucination rates down. This chapter swaps the stopwatch: the one in the user's head. It ticks differently from the server's. Ratings, renewals, and uninstalls all follow it. Lesson one runs an experiment, then unpacks this watch's timing quirks one by one.

Experiment 1 · Same 5 seconds, four ways to live

The four panels below share one request: same question, same 5 seconds to a full answer. The only difference: what you show the user during those 5 seconds. Hit the button to start all four at once, and notice how each feels.

Four options, side by side Start the race
When they finish, tap the panel that felt worst.

Option A · Blank freeze

Give nothing
Tap the button below to start
0.0s

Option B · Spinner wait

Shows it's alive
Tap the button below to start
0.0s

Option C · Streaming

Give part first
Tap the button below to start
0.0s

Option D · Visible steps

Show what it's doing
Tap the button below to start
0.0s
Wall-clock time identical: all 5 seconds
Experiment 2 · Calibrate your body stopwatch

Most people found the blank panel worst. Blank isn't just unpleasant—it has a hidden cost: it inflates your estimate of how long you waited. Perceived-time research calls timing while you attend to time itself prospective timing: the more attention you park on time, the more subjective duration balloons, and anxiety stretches it further. Zakay and Block's attentional-gate model is the most-cited account. Here's the conclusion up front—measuring it yourself is more convincing.

Feel out 8 seconds Not started
(Ready? Tap Start on the right)

After you tap Start, the left panel enters a wait with no progress, no spinner, no copy. Stop when it feels like 8 seconds. Try not to count beats in your head—that's cheating.

You thought8.0s
Actually elapsed0.0s
Error0%

That's the bill blank freeze runs that never shows in the backend: the server logs 5 seconds, the user logs six and a half. Overestimation on no-feedback waits—20–30% is common in experiments. Your product's “felt latency” is worse than the P99 on the monitoring panel, and the gap is decided by the UI.

Experiment 3 · How long a lever is time to first token

If the stopwatch runs in the user's head, spend the engineering budget on the interval they time. Below, this chatbot's total duration is locked at 5 seconds; you move one knob only: TTFT—time to first token (from send to seeing the first character). Drag the slider, hit Replay; the right side replays the typing stream at that TTFT, and the feel score moves with it.

TTFT slider · total duration fixed Drag, then replay
2.4s
Total fixed at 5.0s—only first token moves
Feel score
64
Summarize this meeting note into three to-dos
Tap Replay to feel this TTFT tier
Why engineering pushes TTFT first: Cut first token from 5s to 0.5s and you save zero seconds of total time—yet the feel score jumps 2–3×. Leave TTFT alone and cut total from 5s to 4s, and this stopwatch barely moves. Streaming, outline-first, chunked returns all beat buying bigger machines on cost-effectiveness.
Experiment 4 · Three timing quirks of the user's stopwatch

Package the feel above into three reusable rules. Each tier has a mini demo—switch through all three.

Three timing quirks 1 / 3 tiers
User's stopwatch
0.6s · first character seen, stopwatch stops
Server's stopwatch
5.0s · full answer lands

The user's clock starts on Send and stops at the first character. After that, attention goes to reading—tokens arriving while they read barely enter the waiting ledger. Engineering watches total-duration P99; users only remember first token. When the two watches disagree, bad reviews slip through the crack. How long the TTFT lever is in Experiment 3 comes from this quirk.

Bare
With updates
Same 6 seconds—which felt worse?

Same wait: answer lands at 5 seconds. Flip the two endings and watch remembered duration change.

“Came back in a moment, answer was on point—this tool works.”
Remembered duration
~4 sec

Physically it's the same 5 seconds. Retrospective time is rebuilt after the fact—the worse the mood, the longer the reconstructed wait. So latency and hallucination often share one bad-review sentence: slow and dumb. Answer quality gets blamed onto speed; conversely, a good answer can cover for some of the slowness. Lesson 4 on the peak-end rule puts this into ending design.

Thesis · Users score you on the perceived stopwatch

AI products have three native flaws: slow (inference takes time), wrong (hallucination never fully dies), opaque (users can't see what it's doing). Engineering can only ease these in the short run, but ratings come from perception—and perception can be designed. Streaming, model routing, semantic cache, Agent progress reports from earlier chapters each also have a psychology-switch identity. This chapter teaches when to flip them for psychological effect.

Perceived performance ≠ actual performance. The server's 5 seconds is a physical fact; the user's 5 seconds is a psychological event. Ratings, renewals, uninstalls all happen in that second 5 seconds.
Exit quiz · budget for one cut only Pick one
Users call your AI assistant slow. Measured total is 5s, first token 4.2s. Sprint capacity for one change—which cut first?
AUpgrade the inference cluster—cut total from 5s to 4s
BAdd streaming—first token to 0.5s, total unchanged
CShip a fancier loading animation—spinner upgrades to particles
DPop a tip explaining the model is thinking deeply
Chapter map · how the sixteen lessons continue

Slow, wrong, opaque—each flaw has lessons behind it; later come relationships and money. Tap a card to flip and see which user complaint each lesson targets.

Sixteen lessons, one map 0 / 16 flipped
Sources and further reading: The attentional-gate model of perceived time is in Zakay and Block's series (from the 1990s); overestimation under no feedback and high arousal keeps showing up in prospective-timing experiments. The classic survey of waiting experience is David Maister, The Psychology of Waiting Lines (1985)—“occupied time feels shorter; unexplained waits feel longer” comes from there. The 0.1 / 1 / 10 second tiers for UI response are in Nielsen, Usability Engineering (1993). Experiments showing progress indicators raise preference and patience: Myers (1985). The mechanism by which emotion rewrites time memory shares a root with Lesson 4's peak-end rule (Kahneman et al.).

✅ What this lesson wants to share