12 July 2026
3 min read
We asked one AI assistant the same set of questions about a fixed panel of thirty-five named independent stays across four markets. Thirteen were named in its answers; twenty-two were not. The method, the counts, and what a dated record is for.
A growing share of guests begin a trip by asking an assistant, in plain words, where to stay. Whether an independent stay can appear in that answer is usually discussed as opinion. We would rather it were discussed as a record. So this month we started one.
The method, plainly
On 12 July 2026 we put the same set of questions to a leading AI assistant about a fixed panel of thirty-five named independent stays across four markets: Galata and Beyoğlu in Istanbul, London, the Cotswolds, and the Turkish coast. The questions are the ones a guest actually asks — where to stay near a place, somewhere characterful and independent, specific names please. Every dated answer is archived in full, on the day.
Three rules hold the record honest:
First, we publish counts, never names. Whether a third-party stay was or wasn't named by an assistant is that stay's information, not our marketing. The panel appears in public only as aggregate numbers.
Second, one assistant, one question set, a single snapshot each month. This is a record, not a ranking. A different assistant, a different phrasing, a different day would give different answers — which is exactly why the method stays fixed and the limits stay stated.
Third, the record is dated. An answer captured on 12 July 2026 cannot be reconstructed later — by us, or by anyone. That is what makes it worth keeping.
The July counts
Of the thirty-five stays asked about, thirteen were named in the assistant's answers. Twenty-two were not.
That is the whole public result, and it is enough. Roughly two in three of the named, design-led, guest-loved independent stays on this panel are — at this snapshot, to this assistant — invisible at the moment a guest asks. Not penalised, not outranked. Simply not in the answer.
Why we run it
We build for the layer where stays get found when people ask — server-rendered pages an engine can read, plain facts, structured data. Nobody can promise an assistant's answer, and we never will. What can honestly be done is to build the conditions and then measure what happens, month after month, against a fixed panel, with the archive to show for it.
The counts live on the evidence page, under the same dateline as every other measurement note. Next month the panel runs again, the same way. How the reading layer itself works is in an earlier essay.
If you run an independent stay and want to know where you stand when guests ask, that reading is part of the free listing review — answered within two working days, always before any payment.
Have a place or a film in mind?