Case Study: An AI Agent Manages the Google Profile of Gasthaus Zum Wilden Mann

A traditional inn in the heart of Salzburg, more than 1,500 Google reviews from all over the world. Since July 2026, an AI agent has answered every new review, and we approve every reply.

Gasthaus Zum Wilden Mann has been around since 1864. It sits on the Getreidegasse, a few steps from Mozart's birthplace. In the dining room they serve Wiener Schnitzel, Zwiebelrostbraten and crispy roast pork, and in summer guests sit in the garden. Many guests travel from far away, and many of them leave a Google review afterwards.

We manage the inn's Google Business Profile. For that we use an AI agent that is connected to the profile through the Honeyfield Marketing MCP. All numbers in this article come straight from the profile, queried on 28 October 2026.

The starting point: many reviews, many languages

The profile has 1,545 reviews with an average of 4.2 stars. New ones keep coming in, on some days several. Guests write in German, English, Italian, Czech, Japanese or Chinese.

Answering every review takes time. In a foreign language it takes even more. In the daily life of a restaurant it slips, because the kitchen and the guests come first. Yet every prospect reads the replies before booking a table.

The current state, every morning

We look after several businesses with a Google profile. A routine in the cloud reads the profiles of all of them in the morning: new reviews, open replies, changes to the profile. That way we know every day where there's something to do, without logging into each profile separately. Zum Wilden Mann is one of these businesses.

Answering reviews in the guest's language

For every new review, the agent writes a draft reply. It answers in the language the guest wrote in. We read the draft, change a sentence if needed and approve it. Only then does the agent publish the reply.

Since July 2026, the agent has published 264 replies this way. The change log shows replies in German, English, French, Italian, Spanish, Swedish, Czech, Croatian, Japanese and Chinese. Among the last 250 reviews, none is without a reply. Of the last 20 replies, 17 went live within three days of the review.

A good reply picks up what the guest wrote. One guest, for example, was looking for a dish the inn no longer serves, and wrote (in German):

"… then the landlady told us they unfortunately no longer served it."

The guest then praised the venison goulash and the filled "Wilder Mann" schnitzel. The reply picks up both:

"What a shame the Beuscherl didn't work out, all the more we're glad you enjoyed the venison goulash with red cabbage and napkin dumplings and our 'Wilder Mann' schnitzel so much."

That way the reply doesn't read like a template, it reads like a conversation. The tutorial Respond to Google reviews with your agent shows how this works step by step.

Taking criticism seriously

Not every review is good. One guest gave three stars and wrote (in German):

"It took forever for us to get our food; the schnitzel was cold by then …"

The reply apologizes specifically and names both points:

"We're sorry you had to wait so long for your food and that the schnitzel was no longer warm when it reached you; that's not our standard."

Most replies address the guest informally ("du"). For criticism, the agent uses the formal "Sie". With critical reviews we read the draft especially carefully. Whatever the reply promises has to actually happen at the inn.

Keeping the profile up to date

Reviews are the most visible part. The agent looks after the rest of the profile by the same rule: it proposes, we approve, then it makes the change.

• Opening hours. On 30 August the agent entered the regular hours: Wednesday to Monday from 11:30 to 22:00, closed on Tuesdays. On 12 October it added a closing day as special hours.
• Description. On 14 September the agent rewrote the description. It now names the location, the history, the food, the opening hours and the ways to book.
• Posts. A summer post went live in August, and an autumn post in October with a photo of the dining room and a button to book a table.

A local SEO audit checks the profile for categories, description, opening hours, attributes, photos, posts and reviews. On 28 October the profile scored 125 out of 125 points.

How guests find the inn

The agent also checks which search terms people use when they see the profile. The result for the last three months is clear: most people don't search for the name. They search for "restaurants", "restaurant in der nähe" (restaurant near me), "biergarten", "kaiserschmarrn" or "schnitzel". "Restaurants" alone brought 44,456 impressions. Searches for the inn's name are at around 440.

For an inn in a tourist city, that means the profile is the first encounter. Someone searching for "restaurants" in the old town decides based on stars, replies, photos and opening hours. Those are exactly the parts the agent looks after.

What the numbers show

In the 90 days from 30 July to 27 October 2026, the profile had 81,613 impressions. They led to 2,053 website clicks, 1,846 direction requests, 794 menu clicks and 207 calls.

Compared with the 90 days before, there were fewer impressions but more actions: 58 % more calls, 36 % more website clicks and 36 % more direction requests. How much of that is down to the agent, we can't cleanly separate. The agent was already running in July, so in both periods. And Salzburg has a strong season that distorts any comparison.

What the agent can't do

• It doesn't decide alone. Every reply, every change and every post only goes out after our approval.
• It doesn't replace the team on site. A good reply shows that someone is listening. What happens in the restaurant is up to the kitchen and the service team.
• It doesn't know special hours on its own. The inn tells it about closing days and holidays. It only sees existing special hours as a count, not with dates.
• It doesn't change the name. A name change triggers a review by Google. That stays the business's decision.

What we learned

First: the routine holds because the agent takes the first step. Answering a review isn't a big task. It only slips when someone has to remember it first. When the draft is ready, approving it takes minutes.

Second: multiple languages are no longer a special case. A guest from Japan gets a reply in Japanese, a guest from the Czech Republic one in Czech. The inn doesn't need a translator for that.

Third: the most important searches are generic searches. Whoever looks after the profile wins on "restaurants" and "schnitzel", not only with guests who already know the name.

The article Automate your Google Business Profile covers how an agent looks after the whole profile.

The Honeyfield Marketing MCP connects your AI agent to Google Business Profile and 13 more marketing sources. The agent proposes every change, you approve. Hosted in the EU. Try it for free at marketing-mcp.honeyfield.at.