3% to 13% direct, in three months.
Maxine and I own MINT @Naschmarkt: 18 serviced apartments in Vienna, in five apartment types. In April the website could not take a single booking, and every guest arrived through a platform taking 15% and keeping the relationship. This is what changed between 1 May and 31 July 2026, read out of the systems that recorded it, and what it does not prove.
direct share of booking value, in three months
booked direct in the quarter
sessions a month from AI assistants
clicks a month from Google search
A good business with no way to be found on its own.
Almost every guest arrived through an online travel agent. The apartments were excellent. The way people found them belonged to somebody else.
commission on every platform stay, taken before I see a cent.
extra discount funded on platform loyalty rates, on top of that.
guest emails owned, so no second stay to earn.
So what I wanted was not more bookings. It was to change where the booking starts: own the question a traveller asks before they have picked a platform.
First a booking engine, not a brochure. Then the pages that bring people to it.
A website only breaks the dependency if it can take the money, issue the invoice and hold the calendar. We built all of that before writing a single page.
The booking engine
- The site: English and German, five apartment types, live availability. The booking step moved from four clicks to two.
- The calendar: the property system stays the single source of truth, so direct and platform bookings never collide. No double bookings at MINT since the sync went live in May.
- Payments: Stripe, cards and Apple Pay, with a stored card charged again later for the refundable rate.
- Invoicing: an Austrian compliant invoice issues itself on every booking, with the full trail on a cancellation.
- Two rate plans, priced honestly. Non-refundable charges in full at booking. Refundable takes half, then collects the balance five days before arrival.
- City tax named in the breakdown instead of sprung at checkout, and long-stay exemptions kept separate from the operator’s own discounts.
- A reason to book direct that survives EU law: length-of-stay discounts applied automatically, a welcome code in exchange for an email address. No fake timers, no hidden fees.
60 pages, 30 topics, two languages
- Not blog posts. Answers to what a traveller asks before they pick a city, written so a search engine and an assistant can both quote them.
- Hubs own the big question: where to stay in Vienna, day trips from Vienna, the Christmas markets.
- Spokes close the narrow one that follows: monthly rates for a furnished apartment, is Vienna expensive, Vienna to Bratislava.
- Ten of the thirty are a template set: one page per commercial term, each anchored to demand we could measure first.
- Every piece opens with the direct answer, carries a table a machine can lift, and marks its questions up so they can be read as data.
- We check claims against a source before publishing, because an assistant only keeps quoting a page it can trust.
- Germany is Vienna’s largest source market, so every piece shipped in English and German on the same day.
Part two is the half most operators skip. A booking engine with no demand of its own just sits there, and pages that land on a site which cannot take a card are a donation to the platform.
From a site nobody saw to 43,000 impressions a month.
Ninety days, no advertising, no link buying. The site started answering questions it had never answered before.
clicks a month from search, about 5.7 times May
impressions a month, about 17 times May
average position across the period, roughly, and steady
| Query | Impressions |
|---|---|
| vienna christmas market 2026 | 2,123 |
| weihnachtsmarkt wien 2026 | 545 |
| christkindlmarkt wien 2026 | 237 |
| möblierte apartments wien | 173 |
| apartment wien langzeitmiete | 85 |
| tagesausflüge von wien mit zug | 82 |
| when does vienna christmas market start 2026 | 51 |
| day trips from vienna by train | 25 |
Google Search Console, unsampled, for mintaustria.com.
Narrow questions, in two languages, and not one of them is the property’s own name. That is the whole trick: far more questions answered than before, not a ranking stunt.
74 sessions in July from people who were asking an assistant.
In May the site was invisible to answer engines. By July, one visitor in eighteen arrived from one. These are sessions with a referrer, from four different assistants, landing on specific pages. Not crawler hits.
| Assistant | Sessions |
|---|---|
| ChatGPT | 62 |
| Gemini | 8 |
| Copilot | 3 |
| Claude | 1 |
| All assistants | 74 |
GA4 sessions by referring source, 5.6% of all sessions in July. A second system wired independently, the pixel, counted 38 assistant-referred visitors in its first ten days, 7.7% of visitors over that window. The two were never tuned to agree.
| Landing page | Sessions |
|---|---|
| /en/blog/stay-vienna | 16 |
| /en/blog/day-trips-from-vienna | 10 |
| /en/guides/where-to-eat-in-vienna | 7 |
| /en (the homepage) | 5 |
| /en/blog/boutique-apartment-hotels-vienna-overview-2026 | 4 |
| /en/blog/complete-vienna-activities-visitor-guide-2026 | 4 |
| /en/blog/furnished-apartment-monthly-rates-vienna | 4 |
| /en/blog/is-vienna-expensive-visitor-budget-guide | 4 |
| /en/blog/vienna-3-day-itinerary | 3 |
GA4 landing pages, assistant referrers, 1 May to 30 July 2026.
Nobody asked an assistant about MINT. They asked where to stay in Vienna, what to do on a day trip, and whether Vienna is expensive. The assistant answered with MINT’s pages, and the traveller landed on a site that could take the booking. The homepage came fifth. And the visits turn into people we can write back to: since the pixel went live on 21 July, 51 people have left an email address on the site, and four of them arrived from an assistant first, three from ChatGPT and one from Claude. The pixel ties each enquiry to that first touch and follows the contact through to the booking.
Copilot names the sources it used, and Bing reports those citations where Google Search Console never will. The line sits at zero until late June, ramps from 28 June, and peaks near 180 citations a day in mid July: roughly 1,100 across ninety days, with three pages cited on an average day. The ramp starts in the same fortnight as the German rollout.
Bing Webmaster Tools, captured 21 July 2026. Bing states this is a sample of overall activity, so we report it as directional, not as a precise count.
In July, one euro in every eight came through a channel that did not exist in April.
April is a true zero. The single booking attempt that month was cancelled, and the first reservation through the site landed on 8 May.
Website bookings nearly doubled, €5,597 to €10,616, and the direct share went from 2.8% to 12.7%. The euros and the share say the same thing.
| Bookings made | Direct | Direct share |
|---|---|---|
| May 2026 | €4,436 | 2.8% |
| June 2026 | €5,597 | 4.1% |
| July 2026 | €10,616 | 12.7% |
| Three months | €20,649 | 5.5% |
Read live from the property system. Reservations it marks BOOKED, at full stay value including tax, dated by when the booking was recorded.
commission not paid across the quarter, at my own 15% rate. That is margin, not revenue, so it lands whole.
average nightly rate on a direct booking, against €222 through the platforms, and that is after the direct discounts on offer.
stays booked through the site in three months, every one free of commission and with the guest’s details attached.
email subscribers captured by the welcome offer since 23 June. An audience the business can reach again without paying for the introduction.
The part that compounds is the relationship. On a platform booking I get a masked email address and no permission to write again. Direct, I get the guest, the preference and the second stay.
The weak parts of these numbers, before anyone asks.
Naming them is the only thing that makes the rest of it worth reading. Every line below was in the report before a prospect ever saw it.
This was never about leaving the platforms
The goal was never to leave them, it was to stop being unable to sell without them. On lead time the channels are level, 98 days direct against 97 through a platform, so this is not a different kind of guest.
The Copilot count is a sample
Bing says so itself in the tool that reports it. We quote the figure because the surface is invisible everywhere else, and we treat it as directional rather than as a count.
Three months is three months
Points per quarter, not overnight. The European serviced-apartment average sits near 45% direct, counting walk-in and corporate, so there is more runway ahead than ground covered. Anyone promising the whole distance in a quarter is selling something else.
We will read your numbers the same way: what moved, what did not, and what nobody can prove yet. Book the call and that read is yours whether or not you hire us.
Where each number comes from.
Every figure on this page traces to a named system I can open myself.
Bookings 1 May to 31 July 2026. Google figures run 2 May to 30 July, the last fully settled day.
Read live from the property system itself, not from a report. Reservations it marks BOOKED, priced at the order total, dated by when the booking was recorded, at full stay value including tax, across the whole portfolio.
Cancelled and blocked leads, and 8 direct bookings taken by phone or email, which the website did not produce.
Google Search Console, property sc-domain:mintaustria.com, unsampled.
GA4 sessions by referring source, corroborated by a second pixel wired independently of it.
Bing Webmaster Tools, captured 21 July 2026. Bing states the figure is a sample.
The operator’s own configured rate of 15%, not estimated and not scraped.
MINT @Naschmarkt, Vienna. mintaustria.com. Case study period 1 May to 31 July 2026, published 31 July 2026. The same method runs in any city. Yours starts on the call.
Twenty minutes, and we’ll tell you what’s leaking.
The audit is free, and the read is yours whether or not you hire us.
Or count the commission first, on the calculator