I met Zhou in a restaurant owners' WeChat group. He posted: "Has anyone used ChatGPT to reply to negative reviews? How's the effect?" A bunch of people mocked him — "Are you running a restaurant or a tech company?" I didn't say anything in the group, but I DM'd him.
We talked for over an hour, and I found this guy fascinating. From Chengdu, owns 3 fast-food spots, 20 employees, can barely use Excel — but runs ChatGPT better than a lot of tech people.
He said something I still remember: "I can't code, but I know how to give orders. AI is just your worker — you tell it what you want."
Who Is Zhou
Zhou is 43, been in F&B for 15 years. Started as a line cook in a hole-in-the-wall, saved up, opened his first place, slowly expanded to 3 locations. Sells Sichuan-style fast food — rice bowls, noodles, snacks. Average ticket around 20 RMB. Combined monthly revenue across all 3 spots: about 400K RMB.
His biggest headache isn't bad business — business is fine. It's those "trivial but time-consuming" tasks:
- Every seasonal menu change means hiring a designer for menu layout and food photography — expensive and slow
- Delivery platform negative reviews — not replying hurts your rating, replying makes you furious and wastes time
- Scheduling 20 employees across 3 locations — every scheduling session eats a full day
Not huge issues individually, but they come up multiple times a month. Zhou said he used to spend at least 2 days a month just dealing with this stuff.
Thing 1: AI for New Menu Design
Zhou used to outsource menu design — 2,000 to 3,000 RMB per set, including food photos and layout, with two weeks of back-and-forth revisions. Four menu changes a year meant spending over 10,000 RMB just on this.
His new process has two steps:
Step one: use Midjourney to generate food imagery. Zhou says it's simple — he describes the dish to Midjourney, like "a bowl of Sichuan twice-cooked pork over rice, top-down shot, warm tones, distinct rice grains, glistening pork and garlic sprouts on top, white porcelain bowl." He picks a satisfactory image, then has his intern take real photos of the actual dish for comparison, ultimately using the real photos.
"Why not just use the AI image? Because when customers come in and the menu photo looks nothing like the real dish, they complain. AI images are for reference — composition, lighting. The final menu uses our own photos. But with AI as reference, the intern's photography leveled up instantly."
Step two: use Canva's AI features for layout. Canva has ready-made menu templates. Zhou picks one, fills in dish names and prices, and AI auto-adjusts the layout and color scheme. He finished in 30 minutes what used to take a designer two weeks.
The entire process — from kickoff to final menu — takes 2 days. Cost: one month of Midjourney subscription plus Canva's free tier.
Thing 2: AI for Negative Review Replies
This is what Zhou is most proud of.
Zhou used to reply to negative delivery reviews himself. As a hot-tempered Chengdu native, he'd often end up arguing with customers in the replies. Result: the negative review stayed, and the platform docked his score for "hostile merchant attitude."
Now he lets ChatGPT handle it. His prompt is simple:
"I'm a fast-food restaurant owner. I received a negative review: '{review content}'. Write a reply that: 1) is sincere but not groveling; 2) admits fault if it's ours, diplomatically explains if it's not; 3) under 100 words; 4) makes the customer feel 'we take you seriously.'"
ChatGPT's replies are consistently tactful — apologetic without being sycophantic, explanatory without deflecting, with a small incentive hook to bring the customer back. Zhou says since switching to AI replies, his store rating went from 4.2 to 4.6, and he no longer wastes time fuming.
One detail is particularly funny. A review once said "too little meat," and ChatGPT's reply included: "We'll reflect on our portion sizes, and we'd love to welcome you back to dine in — we'll add extra portions for you on the spot." Zhou slapped his thigh: "I'd never think of that myself, but the customer definitely feels good reading it."
Zhou now routes all 3 locations' negative reviews through ChatGPT. Each review, from reading to replying, takes 5 minutes. He used to spend 3-4 hours a week on reviews; now it's under 30 minutes a week total.
Thing 3: AI for Staff Scheduling
20 employees, 3 locations, day and night shifts, plus leave requests, shift swaps, and holidays — scheduling looks simple but can drive you insane. Zhou used to spend a full day every month on scheduling, and it still had errors. Employee complaints about unfair schedules were routine.
He handed this to ChatGPT too. His method: type out the 3 locations' staff lists, positions, available time slots, and leave requests in plain language, then ask ChatGPT to generate a schedule.
The prompt goes something like:
"Help me schedule one month. 3 locations, each needs: 2 people for day shift (9:00-17:00), 2 for night shift (17:00-23:00). Staff list and availability: {list}. Rules: everyone gets at least 1 day off per week, no more than 3 consecutive night shifts, don't schedule anyone on leave. Output as a table."
ChatGPT outputs a Markdown table. Zhou copies it to Excel, makes minor adjustments, and it's ready. From input to finished schedule: 20 minutes.
Of course, things went wrong sometimes. Once ChatGPT scheduled an employee who was on leave. Zhou didn't check carefully, sent it out, and one location was short-staffed that day. After that, he added a second check: "Verify this schedule — is anyone scheduled during their leave? Is anyone working more than 6 consecutive days?"
With this double-check, scheduling errors dropped to nearly zero.
The Ledger: Costs vs. Returns
Zhou broke down the numbers for me. Over 3 months:
- Midjourney subscription: ~200 RMB/month, 600 RMB for 3 months
- ChatGPT Plus: ~150 RMB/month, 450 RMB for 3 months
- Canva: free tier is enough, 0 RMB
- Total tool cost: ~1,050 RMB for 3 months
What he saved:
- Menu design outsourcing: previously ~12,000 RMB/year for 4 changes, now DIY — saved ~3,000 RMB in 3 months
- Review reply time: previously 3-4 hours/week at ~60 RMB/hour — saved ~2,880 RMB in 3 months
- Scheduling time: previously 1 day/month — saved 3 days in 3 months, worth ~1,800 RMB in daily output
- Better ratings from proper review replies → platform traffic boost → ~200 extra orders/month — ~12,000 RMB extra revenue in 3 months
Total benefit: ~20,000 RMB in 3 months, on 1,050 RMB investment. Zhou's words: "Better ROI than running a restaurant."
Zhou's 3 Honest Truths
Finally, I asked Zhou what advice he'd share with other restaurant owners. He thought for a while and said three things:
Truth 1: "Don't be afraid of AI — it's more useful than the intern you hired." Zhou says he was scared at first, thinking AI was high-tech that a cook couldn't understand. Then he realized: ChatGPT is just something that works when you talk to it. Like training a new hire, except this one's online 24/7, costs no salary, and never throws a tantrum.
Truth 2: "Always review what AI produces before sending it out." Whether it's review replies or schedules, Zhou always reads through before hitting send. AI occasionally makes stupid mistakes, but if you eyeball it, you're fine. "It's a good helper, but not your boss. You make the final call."
Truth 3: "Don't wait — just start using it." Zhou says many restaurant owners he knows are aware of AI but keep saying "let me wait and see." His attitude: tool subscriptions cost a few hundred RMB a month. The cost of trying is basically zero. What are you waiting for? "Just start using it. You'll figure out where it helps as you go."
After talking with Zhou, I had one deep realization: the biggest barrier to AI adoption isn't technology — it's mindset. Zhou can't code, can't use advanced Excel features, but he has something many bosses don't: willingness to try.
A 43-year-old restaurant owner who can barely pronounce the word "prompt" is saving more time and money with AI than many tech-industry people who scroll AI news every day.
This is what AI adoption actually looks like — not in a lab, but in a fast-food joint in Chengdu.