First Order

The question every diner asks but no guide answers

London has no shortage of restaurant guides. Michelin, Harden's, TimeOut, The Infatuation, Hot Dinners. They all answer the same question: where should I go? What none of them answer is the question you actually have once you have booked: what should I order?

The best answer to that question comes from the restaurant itself. Not a food critic who visited once. Not a diner who had the wrong thing. The person who answers the phone, who knows what gets sent back and what gets photographed, who has heard the chef's pride and felt the kitchen's rhythm. That person knows the one dish.

First Order sends an AI voice agent to make that call. Alice calls the restaurant, introduces herself, asks the question, listens to the answer, and surfaces it on the site. Simple. Human. Genuinely useful.

AI as restaurant researcher

The voice agent, Alice, calls restaurants during the gap between lunch and dinner service when someone is usually near the phone. She introduces herself honestly ("I'm putting together a free food guide for London"), asks the one-dish question, and wraps the call in under 60 seconds if the answer comes quickly.

The key prompt insight, borrowed from the Guinndex model, is restraint: one question, one dish. No follow-ups that pressure the caller. No elaboration that loses them. The discipline of a single question is what gets real answers from busy restaurant staff.

Each confirmed recommendation becomes a card on the site: the dish as the hero, the restaurant and neighbourhood, a direct quote from the call transcript. A "First Order" tag for signature dishes, a "Sleeper Pick" tag for things that surprised us. The transcript excerpt lets you hear the exchange.

Voice AI as data collection infrastructure

Independent venues answer. Chains do not. The big-name restaurants with central reservations and IVR phone trees are effectively unreachable for this kind of call. Turkish grills in Dalston, neighbourhood bistros in Hackney, family-run cafes in Walthamstow: they pick up, they answer, and they tell you exactly what to order. The most useful food data is in the independents.

Scale changes everything. Nine calls returns nothing useful. Ninety-eight calls returns real conversations, real recommendations, real data. Voice AI as a research method only works at scale. The infrastructure cost is low enough that scale is achievable.

The quotes are the product. Wait times are structure; quotes are soul. "The adana. Definitely the adana." "The cornbread – it's like a signature starter." "The breaded butter pudding." These are what people share. The data is useful; the voice is what makes it memorable.

3pm is the sweet spot. Between lunch and dinner service, the phone gets answered by someone who has a moment to spare. Timing a calling campaign around restaurant service patterns dramatically improves connection rates.