The fix, shown · agent-ready reservations
People are starting to hand this exact sentence to an assistant instead of typing it into Google. Below is that request, played out twice: once as a booking, once as a private-event inquiry, using the data Meteora's site would expose once it's agent-ready. Press play.
✓ Assistant → guest
Booked. Meteora, Melrose Ave, 1-star Michelin, live-fire. Friday, 7:30pm, table for two. Six-course tasting, $165pp, one seat flagged vegetarian for the choice-of-protein course. Total with the optional service charge: ≈$375. Confirmation and the $25/guest 72-hour cancellation policy are in your email.
Every fact above (price, format, the vegetarian substitution, the cancellation window) has to live somewhere machine-readable for this to work unattended. Today, none of it lives on meteora.la.
This isn't a future feature. Google and OpenAI already ship agentic restaurant booking: the only question is whether Meteora is the one that gets recommended, or the one the assistant can't quite read.
Where Meteora stands today
On April 10, 2026, Google took AI Mode's agentic restaurant booking worldwide, dropping the subscription requirement it had in the US since August 2025, powered by partner platforms that include OpenTable. Separately, OpenTable is OpenAI's first restaurant-tech partner for booking through ChatGPT's Operator. Meteora is on OpenTable, so a version of "already listed" is probably true here too, the same way Wally's was already inside Shopify's agent rollout without doing anything.
But that bookability lives entirely inside OpenTable's data, not Meteora's own. The lever restaurants actually control (the one that decides whether Gemini, ChatGPT, and Perplexity describe you accurately and recommend you with confidence instead of a thin, half-right scrape) is Restaurant / Menu / FAQPage schema markup (the invisible labels search engines and AIs read) on meteora.la itself. That's the gap finding #3 in the teardown points at, and it's the one thing here Meteora fully controls.
Listed ≠ ready: what wins the pick
Restaurant + Menu + FAQPage schema.org markup, in plain text on meteora.la: the 4-course/6-course format, both prices, and the full-party-participation rule. Right now that's invisible to anything that isn't a human clicking through OpenTable.
An agent won't confidently book a shared-menu tasting format for a mixed-diet party unless it can verify a substitution exists. State the policy once, in text, and it stops being a phone call for every party of two.
The teardown's finding #1: right now an assistant researching a Meteora buyout can't safely fill out that form on a guest's behalf: it doesn't know which of three restaurants it's actually inquiring about. A Meteora-scoped intake fixes the highest-margin path first.
This session hit OpenTable's own bot-wall outright. Exposing your own availability signal, even a simple "book on OpenTable, or call/email us" agent tool, means one vendor's front door isn't the only way in.
The two doors: the finding #1 fix, mocked up in your palette
✓ Assistant → planner
Sent. Meteora private-events inquiry submitted: 20 guests, September, full buyout, routed directly to Meteora's own events coordinator, not a shared multi-restaurant queue. Expect a reply within one business day.
Same scenario as finding #1 in the teardown: the only thing that changed is that the form now knows which restaurant it is.
One more piece
The demo above only works if the underlying facts exist somewhere an agent can fetch them. Here's the starter kit: llms.txt (the file AI assistants read first), an agent card and MCP tool schemas (the files that let an AI assistant use you correctly), and a concierge persona, scoped correctly to Meteora alone.
Open the Meteora Agent Kit → ← Back to the teardown