Cookiy ResearchAI Meal PlanningApril 2026

Dinner is decided fast. Deciding takes forever.

Five Tier-1 city professionals run through a daily 10-to-15-minute scroll before eating, even when they already know what they want.

5 of 5 had last dinner plan disrupted
FIG. 01 · HERO IMAGE · WEEKDAY-DINNER-AI-TIER1
5 of 5had plan disrupted last week
10–15 mintypical decision scroll
2–3max alternatives wanted
2–3 weekstrial window for trust
Executive summary — three sentences
Weekday dinner is a decision problem, not a cooking problem: every participant described 10–15 minutes of scrolling, switching apps, and second-guessing before the first click. Last-minute disruptions — late Zooms, client asks, a sick child, a busy kitchen — hit five of five participants in their most recent workweek, and each retreats to the same single trusted meal (congee, noodle soup, dumplings, a chicken bowl) to stop the spiral. An AI weekly plan earns trust only if it hands over two or three adaptive alternatives on the fly, stays out of health data and real-time tracking, and can be paused with one tap.
5 of 5plans disrupted last week
4 of 5name scrolling as biggest drain
5 of 5have one trusted fallback meal
5 of 5prefer subscription over pay-per-use

Four archetypes

● The Scroller 2 of 5 "I probably considered like four or five options seriously — two usual spots closed, a couple others had super long delivery times or bad ratings."
— P01 · Beijing knowledge worker
■ The Optimizer 1 of 5 "I'd tolerate a miss here and there, maybe one or two meals in a week not fitting perfectly. But if issues happen often or cause waste, I'd lose confidence quickly."
— P02 · Shanghai consultant
▲ The Parent-Pivoter 1 of 5 "The hardest part was juggling cooking a second safe dish while answering work messages. It felt like I was pulled in two directions."
— P03 · Guangzhou working parent
◆ The Friction-Avoider 1 of 5 "I like clear choices, so I don't waste time deciding. Either cook a simple dish, eat semi-prepared food, or order delivery. Then pick the fastest or easiest."
— P05 · Shenzhen apartment-sharer
01

The decision is the drain, not the dinner.

Four of five participants named the choice step — scrolling apps, filtering, second-guessing ratings — as the longest and most draining part of the evening. Typical estimates cluster at 10–15 minutes; cooking or waiting for delivery felt shorter than getting to the first click.

Counter: 1 of 5 (P03, the parent) said deciding was quick; her time drain was executing the second backup meal while answering work messages.

Implication. Design for the 'first click in under 60 seconds' moment. Surface a two-or-three-item shortlist pinned to the current hour and energy level; do not present a catalogue.

"Definitely the scrolling and second-guessing part — going back and forth between apps or friends' suggestions, checking ratings to avoid a bad meal. It feels like wasted energy, especially when I'm already drained."— P01 · Beijing knowledge worker
02

Every plan breaks; fallback is the plan.

All five participants described their most recent workday plan being overridden — a late Zoom (P01), a client ask at 6:45pm (P02), a daycare health flag (P03), a review call (P04), a crowded kitchen (P05). None executed the original plan.

Counter: 2 of 5 (P02, P05) had explicitly pre-staged a named fallback (grilled-chicken-breast kit; trusted noodle shop) and reached it in under two steps; the other 3 improvised under fatigue.

Implication. Build the product around 'my plan just broke — what now?' Pre-compute 2–3 substitutes per slot that preserve the user's stated constraint (light; under 30 min; under 60 kuai) and deliver them the moment a disruption signal arrives.

"If I had a plan that accounted for possible delays — like some quick, healthy semi-prepared meals at the office or nearby — or maybe an app showing real-time open hours nearby, that would have helped."— P02 · Shanghai consultant
03

Everyone has one safe meal they fall back to.

All five participants named a single trusted meal they order or cook when too drained to decide: congee (P01), grilled chicken with pre-cut veg (P02), frozen dumplings (P03, P04), a specific neighbourhood noodle shop (P05).

Counter: 1 of 5 (P02, the optimizer) keeps two defaults rather than one, and rotates them; the other 4 collapse to a single option under fatigue.

Implication. In onboarding, ask for the one-meal default. Treat it as a reserved slot in every generated plan, and use it as the automatic 'skip tonight' substitute.

"Congee is my quick default — easy on the stomach, filling enough, and usually available pretty late. It's like a safe anchor when I have zero energy or patience to hunt for anything else."— P01 · Beijing knowledge worker
04

Trust is gated on control, not on accuracy.

Five of five drew the same privacy line: preferences, rough schedule, basic location are acceptable; health data, real-time location, calendar access, and payment history are disqualifying. Every participant asked for toggles and explicit override buttons before flexibility of the recommendations themselves.

Counter: 2 of 5 (P02, P04) said they would tolerate minor recommendation misses for several weeks if control was clearly theirs; the other 3 said one privacy surprise would end the trial immediately.

Implication. Ship granular data toggles on day one. Default every sensitive permission to off. Show, in plain language, what is used to produce each suggestion; make it one tap to revoke.

"I'd be okay sharing basic taste preferences, general timing, and maybe rough budget — I see those as helpers. But accessing my detailed location history, full payment records, or health data feels like way too much."— P01 · Beijing knowledge worker
05

Paid pricing is set by minutes saved, not features.

All five participants preferred a monthly subscription over pay-per-use, but anchored price to time reclaimed from the scroll. One participant named a specific band — 15 to 30 RMB per month — and the others described the same shape in words.

Counter: 2 of 5 (P02, P04) said they would only pay after two or three weeks of clear benefit; a paid offer before that would push them to quit.

Implication. Set a modest monthly anchor (sub-30 RMB) and delay the paywall until after the habit window. Show the 'minutes-of-scroll avoided' metric in every billing touchpoint.

"I'd say about 15 to 30 yuan per month feels reasonable, given that I'd be saving time and mental energy. If it's priced much higher, I'd expect really strong, proven benefits."— P04 · Shenzhen engineer
"Finally got congee and a small side around 10pm. Felt relief but honestly sort of resentful that even dinner felt like a negotiation." — P01 · Beijing knowledge worker

Dinner is never the problem. The scroll always is.

Five of five participants already know what they want; the product job is to stop making them prove it again every night.

Powered by Cookiy AI

The consumers in this report. You could talk to them.

Cookiy AI runs AI-moderated consumer interviews at scale — in hours, not weeks.

Learn more →