Why We Built AIscore — and What It Changed About How We Work

1. The Question That Changed Everything

About eighteen months ago, a client — a four-star hotel in the south of France — asked us something we had never been asked before: "Does ChatGPT recommend us when someone searches for hotels in our area?"

We did not know. We ran some manual tests on ChatGPT and Perplexity. The hotel appeared occasionally, inconsistently, with inaccurate details — the phone number was wrong, the check-in time was not mentioned, one response described a restaurant the hotel no longer operated. The hotel's Google ranking was strong. Its Booking.com profile was well-maintained. But in AI-generated responses, it was essentially a ghost.

We looked for a tool to measure this properly. Something that could audit a hotel website against the signals that AI systems use to evaluate a property — structured data quality, bot access permissions, content legibility, metadata completeness. Nothing credible existed. There were generic SEO tools, there were anecdotal blog posts, there was a lot of speculation. There was no instrument.

So we decided to build one.

2. The Industry's Blind Spot

The more we looked at the problem, the larger it became. Hotel digital marketing has sophisticated measurement frameworks for almost everything: Google rankings tracked daily, OTA conversion rates monitored weekly, social media reach and engagement reported monthly. Revenue managers live inside dashboards. The data culture in hotel distribution is, in many ways, exemplary.

And yet: AI search, which is growing at roughly 27% per year and already accounts for a meaningful share of how younger travellers begin their accommodation research, has no equivalent metric. It is not tracked. It is not reported. In most cases, it is not even discussed.

This is not negligence. It is a measurement gap. When something is not measured, it does not get managed. When it does not get managed, it does not improve. The absence of a score means the absence of a baseline, which means the absence of accountability, which means the absence of progress. The hotels scoring D and F on AI visibility are not doing so because they have decided AI does not matter. They are doing so because they have never had a reason to check.

This is the industry blind spot. It is large, it is systematic, and it is entirely addressable — but only once you can see it.

3. What We Wanted to Measure

We did not want to build another single-metric tool. The existing conversation about AI search optimisation tends to focus on one or two signals — typically robots.txt permissions or schema presence — in isolation. That is not how AI systems actually evaluate a website.

What we wanted was a composite score: a single number, like a credit rating or a PageRank, that synthesised the full picture. Something a hotelier could look at and immediately understand where they stood, without needing to know what JSON-LD is or why hreflang matters.

The methodology we developed analyses 91 distinct signals across five categories: bot access permissions, structured data quality, content structure, technical metadata, and AI-specific signals like llms.txt presence. These are combined into a single 0-100 score with letter grading.

The most important architectural decision was the gate multiplier. Bot access — whether the major AI crawlers are permitted to index the site at all — does not just add to the score; it scales the entire score. A site that blocks three of the four principal bots will never score above a certain ceiling, regardless of how good everything else is. Because a blocked bot will not see your schema, your headings, your metadata, your llms.txt — it will not see any of it. The gate comes first.

This model means the scoring is not just a checklist. It reflects the causal structure of how AI systems actually interact with websites.

4. What 9,500 Scans Taught Us

We expected the results to be mixed. We did not expect them to be this bad.

An average score of 38 out of 100. 65% of hotels at D or F. 18% blocking all four major AI bots entirely — including some well-known, well-resourced properties that would be horrified to know this. Less than one in ten hotels with an llms.txt file. Nearly a quarter of hotel homepages with no H1 heading.

The scale of the problem was sobering. But the nature of the problem was actually encouraging. These are not hard problems. A robots.txt file with an unnecessary block rule can be fixed in five minutes. An incomplete Hotel schema can be completed in an afternoon. An llms.txt file can be drafted and deployed in a day. The technical barrier to improvement is low. Almost everything in the bottom quintile of AIscore results could move to the top half with a focused week of work.

What is missing is not capability. It is information. The hotels at the bottom of the distribution do not know they are there. They have no score, no baseline, no benchmark, no dashboard that tells them "your AI visibility is poor and here is exactly why." That information gap is what we are trying to close.

5. What We Want Next

The goal is not to sell audits. We do sell audits — and consulting, and implementation services — but the existence of a paid service layer is not why we made the tool free and public.

We made it free because the problem is an industry problem, not a competitive advantage problem. A hotel scoring F on AI visibility is not losing market share to us; it is losing to its competitors who happen to have better schema, or who accidentally have open robots.txt rules, or who worked with an agency that knew about llms.txt. The information asymmetry hurts the whole market.

What we want is for AI visibility to become a standard metric in hotel digital performance — as routine as Google rank, as expected as a Booking.com score. We want it in monthly performance reviews, in agency briefs, in RFPs. We want it to be something every hotelier knows, the way every hotelier knows their occupancy rate.

We are not there yet. 91% of hotel websites still do not have an llms.txt file. The average score is still 38. But the trajectory is clear, and the tools to change it exist. The only thing left is to use them.