17 guesthouses tested. The best one managed 61 out of 100.
Between 19 July and 20 August 2026 we ran a full AI visibility audit on 17 South African guesthouses and lodges. Every one was scored out of 100 and tested live against ChatGPT and Gemini.
Sixteen of the seventeen scored under 60. One cleared it, with 61. Not one reached the level where an AI assistant can confidently find a guesthouse, describe it correctly and recommend it to a traveller.
We score question coverage out of 10, meaning whether a site answers the plain things a guest asks before booking. Is there parking? Is breakfast included? What time is check-in? Are dogs allowed? Is there a family room?
Average question coverage across 17 guesthouses. Fourteen of them scored zero.
Wonderful reviews. Nothing to quote.
Reputation was by far their strongest category, averaging 7.5 out of 10 against 5.8 for the rest of the businesses we have tested. Ten of the seventeen scored 8 or better. These are places guests genuinely rate.
Eight of those well reviewed guesthouses scored zero on question coverage. So an assistant can tell a traveller that people liked staying there, and then go quiet the moment it is asked anything a person actually needs to know before booking.
Reviews are not the problem. Reviews tell an AI that you are liked. They do not tell it whether you have parking, whether the rate includes breakfast, or whether a family of four can stay.
The door is open and the shelves are bare
Almost none of these sites were blocking anything. Crawler access averaged 8.5 out of 10, the best score of any category in the group, and better than the wider study.
Thirteen of the seventeen had the door wide open to AI systems and no question and answer content behind it. Sixteen of the seventeen had no guide file of any kind telling an AI what the business is.
Why this costs you a booking
Travel is one of the first things people stopped researching in twelve browser tabs. They ask an assistant where to stay near a town, for a weekend, with a dog, and they get back one answer with two or three names in it.
To be one of those names, the assistant has to be able to say something specific and safe about you. A guesthouse it cannot describe accurately is a risky recommendation, and the safe thing to do with a risky recommendation is leave it out and name a competitor instead.
There is a second cost that is easy to miss. If your own site says nothing, the assistant answers from whatever else it can find, which usually means a booking platform listing. Your guest arrives through somebody else, and you pay commission on a booking that started as a question about you.
It is not the platform
Thirteen of the seventeen were built on WordPress, and most of them look good. We tested that question directly across 65 WordPress sites and found the platform makes no measurable difference, so this is not a software problem and it is not a criticism of whoever built the site.
A site built to make a room look inviting to a person is built differently to one built to be read and quoted by a machine. Both are needed now. Almost nobody has caught up, which is why a whole sector sits in the 40s.
The encouraging part
The two weakest areas here, a guide file and question coverage, are also the two cheapest to fix. Neither one requires redesigning your site or changing a single thing a guest sees.
You already know the answers. You repeat them in emails and on the phone every week. The work is writing them down somewhere a machine can read them.
This article draws on our wider study of 135 South African businesses tested for AI visibility. No business is named, and all figures are published as aggregates. See also: we tested 65 WordPress sites, the best scored 83 and the worst 19.