Case Study: An AI Agent Manages Haus Hana's Visibility
The most convincing tool is the one you use yourself. We manage a real business through the same connector we sell.
Sunday evening, a new Google review comes in for Haus Hana. Monday morning, the draft reply is ready. Approve, send, done. This is exactly the workflow we sell. So we run it on ourselves first.
Haus Hana is a small accommodation business. We manage its local visibility through an AI agent, connected via the Honeyfield Marketing MCP, our own connector. No test account, no demo dataset. A real profile, real guests, real reviews.
For an accommodation business, local search decides bookings. Anyone looking for a place to stay in the area sees the map with three listings first. Google describes the logic behind it officially:
"Local results are mainly based on relevance, distance, and popularity."
We can't change distance. Relevance and popularity we manage directly in the profile.
What Used to Fall Through the Cracks
The marketing work for a business like this is rarely big. It's small and constant. A review needs a reply. The holiday hours need tending. A new photo belongs in the profile.
None of these tasks has a deadline. That's exactly why they slip through. Before, it was manual. Someone logged in, clicked through, forgot half of it. How often it happened depended on whether anyone happened to think of it, and mostly no one did.
And guests pay close attention. 89 percent of consumers expect a business to respond to reviews, the good ones and the bad (BrightLocal, Local Consumer Review Survey 2025). A profile with old photos and silent reviews costs bookings before a guest ever gets in touch.
The Setup: Five Sources, One Agent
Five of Haus Hana's data sources feed the agent:
- Google Business Profile for reviews, categories, opening hours and photos
- Search Console for the search queries guests use to find the website
- GA4 for behavior on the website
- Microsoft Clarity for the recordings of clicks and scrolls
- WordPress for the site's content
On its own, each source says little. Together they form a picture: what guests search for, what they do on the site, where they drop off. How five sources turn into shared context is covered in a separate post.
More important than the access is the limit. Every write action is draft-first. The agent proposes a draft, we approve, and only then does anything go out. An approval gate sits in front of every change, an audit log records every action. Everything is hosted in the EU. Why this GDPR-compliant approach is mandatory rather than optional when it's someone else's business is obvious.
The Routine Today
The process is a weekly routine. Here's how it runs.
The agent checks new reviews and writes draft replies in the voice of the house. It only hits that voice because we gave it the context of the house once, cleanly. That's exactly what context engineering in online marketing means. Before, most went unanswered: of 29 reviews, only four carry a reply, the most recent of them from 2019 (as of July 2026). Now the agent proposes a draft for every new review, and approval takes minutes. It reconciles the opening hours with the upcoming holidays before they arrive. It checks the categories against what the business actually offers.
On top of that comes the local rank check. The agent queries where Haus Hana sits in the local pack. That means the search terms that bring bookings. For "pension Karlobag", the profile currently sits at position 3 (as of July 2026), right in the middle of the local pack. If the position slips, we see it early.
We wrote up the routine behind the profile separately: how to automate a Google Business Profile. At the end there's no dashboard anyone has to open. There's a short list of drafts. We cut two sentences, approve, done.
What the Agent Can't Do
Honest stays honest. The agent has clear limits.
It doesn't know the new holiday hours until we tell it. It doesn't decide the tone for an angry review. Choosing the primary category is strategic, not a routine click. And it publishes nothing without our approval.
That's not a flaw. The draft comes from the machine. The decision stays with us.
What We Learned
Two things stay with us.
First, the routine holds because it's delegated. As long as a human has to trigger every small task themselves, it gets left undone. Once the agent lays out the drafts, approval only takes minutes.
Second, we see the product differently now that we run it ourselves. Every limit we run into at Haus Hana flows back into development. A business we actually look after is the most honest test case we have. Plenty of things can be sold. Using it yourself is the better proof.
The Honeyfield Marketing MCP connects your AI agent to the Google Business Profile and ten more marketing sources. Draft-first, with EU hosting. Try it free at marketing-mcp.honeyfield.at.