OpenAI shows how ChatGPT builds a catering pitch for a Toronto restaurant

OpenAI has put a short video on its YouTube channel showing how ChatGPT helps a restaurant find catering clients. The hero is Richard, owner of The Dumpling Prince in Toronto. Host Nick Baumann visits him and takes on two specific tasks: build a catering presentation and set up automated outreach. The video description gives no revenue, no order count and no time saved. It offers only the scenario and a joke about ChatGPT never being issued an apron.

What the video actually shows

The starting problem is stated plainly: a restaurant demands attention around the clock, and hunting for catering clients takes as much time as the kitchen does. Clients for this kind of work (offices, event organizers, private events) don't walk in on their own. Someone has to find them, write to them and send a clear commercial offer.

OpenAI builds its demonstration on that. The first part is the presentation: a document that explains what the restaurant cooks for groups, on what terms, and why a client should pick it. The second part is outbound communication: ChatGPT writes and sends messages to potential clients. The description does not say how the automation is built, which tools it uses or how much a human is involved. You would have to watch the video itself to find out.

What the announcement tells us:

• Client: The Dumpling Prince, Toronto, owner Richard
• Tool: ChatGPT
• Task 1: a catering presentation
• Task 2: automated outreach
• Format: a video on the OpenAI channel, host Nick Baumann
• Results in numbers: not disclosed

Why catering suits an LLM

The task plays to the strengths of language models. It involves a lot of text, templated documents that still need tailoring, and repetitive emails where only the name, event type and group size change. A small business owner usually doesn't hire a copywriter and a sales manager. ChatGPT fills both roles at a "good enough" level, as long as someone checks the output.

The context of the release matters too. OpenAI made the video itself, so it is marketing material, not independent research. The description invites viewers to submit an idea or send a direct message so the team can "come and build this together." It looks like a regular series in which the company shows practical scenarios at real small businesses. There is no data yet to judge the real payoff of this format.

The weak points of such setups are well known. The model can get a price or a dish's ingredients wrong if it wasn't given the current menu. Automated outreach without a vetted recipient list turns into spam. A restaurant in Canada also faces a legal side: the local anti-spam law (CASL) requires recipients' consent for commercial electronic messages, with certain exceptions.

How to repeat the scenario at your own place

Strip away the promotional framing and the approach carries over to any small food business. The workflow looks like this:

• Prepare the source data yourself: the current menu with prices, minimum order, delivery radius, photos and reviews. The model will rely on what you give it.
• Ask ChatGPT to draft a presentation structure for a specific segment, such as office lunches or corporate events, and make a separate version for each.
• Build the recipient list manually or from sources where you have grounds to contact people, and only then generate the emails.
• Keep a human check before sending: prices, dates, the recipient company's name.
• Track the outcome: how many emails went out, how many replies came back and how many orders were closed. These are exactly the numbers missing from OpenAI's announcement.

Judge the idea by these metrics, not by a slick video. If a hundred emails bring a few replies and one order from a large group, the payback can already be noticeable. If not, the problem most likely lies in the contact list or the offer, not in the wording of the email.