Research agent · Restaurant & Hospitality
Instagram & TikTok content research for restaurants
The Research agent reads your own performance, other restaurants and bars nearby and the questions diners keep sending — together — and turns them into a short, reasoned recommendation for what to make next.
The problem
Most restaurants post from gut feel, then check the numbers later.
Native analytics show views and follows one post at a time. They don't tell you that tonight's special, from the person who made it keeps outperforming everything else, or that diners asked about it three different ways in your DMs this week.
Putting that together takes an hour you don't have — which is why most restaurants skip it and fall back to posting whatever is easiest. The problem isn't effort; it's posting during service, when nobody has a free hand.
Tuesday's special, from the person who made it
We only make this one when the fish is right, not on a schedule. Tonight it's right. Table for two, or ask for it to go.
Signal · A tracked competitor switched to kitchen-first reels last week.
How it works
What the Research agent does, step by step.
Restaurants on Instagram and TikTok. Every step ends in a draft, not an action.
- 01
Connect read-only
Link Instagram and TikTok. The agent reads posts, results and comments — it cannot post or reply at this stage.
- 02
Read three signals together
Your last 90 days of results, other restaurants and bars nearby, and the recurring questions in your comments and DMs.
- 03
Explain the pattern
Every recommendation says what it saw and why it matters — e.g. that one dish from prep to pass is the format your diners save.
- 04
Hand it to the script agent
Accept a recommendation and it becomes a brief. Reject it and the agent notes why for next week.
What you get
Drafts you can use, with the reasons attached.
Part of the Research & growth side of Oryxia, reading from the same brand memory as the other six agents.
- A written weekly read-out: what moved, what didn't, and the one pattern worth acting on
- Recommended next posts for restaurants, each with the evidence attached
- Format and topic rankings against your own usual performance, not an industry average
- A running list of what diners are asking, grouped by theme
It recommends. You decide what gets made.
Research never creates or schedules anything on its own. A recommendation only becomes a brief when you accept it.
- 1Accept, edit or reject each recommendation
- 2Rejected ideas are remembered, so they don't come back next week
- 3Nothing reaches the calendar without your click
A good fit if
- You post for restaurants but rarely have time to work out why something worked
- You want reasons, not dashboards
- You're happy to spend a short weekly review deciding what to make
Not a fit if
- You want a report for a client deck or an agency PDF — Research is built to drive your next post, not to decorate a meeting.
- You want a delivery-platform or reservations integration — Oryxia drafts content and replies; your booking system stays yours.
FAQ
Fair questions.
01What data does the Research agent actually read?
Your own Instagram and TikTok posts and their results, the public posts of other restaurants and bars nearby you choose to track, and the themes in your comments and DMs. It doesn't buy third-party audience data.
02Is it useful if my account is small?
Yes — it compares each post against your own usual performance, so a small account still has a baseline. With very little history, it leans more on competitor and audience-question signals, and says so.
03Will it understand how a restaurant like mine talks to diners?
It starts from a 30-minute brand interview and your own past posts, aiming for warm, specific about the food and never over-polished. Your do-not-say list — for restaurants, usually allergen answers the kitchen hasn't confirmed, and menu or price details that have changed — is enforced on every draft, and you approve each one.
Related agents
See what your agents would draft first.
The free Social Audit maps your niche, competitors and hook gaps, and shows the agents working on a demo workspace.