What We Found Auditing 16 European Tour Operators | AI Visibility Study 2026
Why this page exists
Original research is the most citable asset a brand can publish and the hardest to fake. This page is not a testimonial wall. It is a dated, reproducible measurement of a named industry, published with its methodology, which is the form assistants cite and competitors cannot copy without building the measurement infrastructure first.
The commercial mechanism is the same one that made the outreach work: every operator in the table is a prospect who now has a public number next to their name.
What we found auditing sixteen European tour operators
Published 31 July 2026. Next run: monthly.
Method version 1.0. By Georgios P. Katsonis, CTO, SyeniteLabs.
We measured sixteen major outbound tour operators across four European markets for one thing: how well AI assistants can actually read them. One methodology, one day, so every measurement is comparable to every other.
We are not publishing the names or the individual scores. What follows is the pattern, which is the part that transfers to your business anyway.
The Headline
Nobody has claimed this category. Nothing in the field cleared 70 out of 100. The field averaged in the mid-forties. Across four countries, in an industry moving billions of euros a year, not one operator has built a position an AI assistant would treat as the authority on the category.
An operator moving now would not climb a ranking. They would set the reference point assistants learn the category from.
Most of the field has no llms.txt at all, and several of those that do cannot be reached by a crawler anyway, which makes the file decorative.
The largest source market in Europe was the least prepared of the four. Its strongest operator sat well behind the strongest in two smaller markets. Size of business predicted nothing about readability.
A quarter of the field scored zero in practice despite inviting crawlers in robots.txt, for two entirely different reasons. Both are invisible to anyone who reads the file instead of testing the fetch.
The two ways a good site scores zero
This is the finding that transfers to every industry, and it is the reason a robots.txt review is not an audit.
1. The crawler is refused at the edge
The robots.txt file welcomes GPTBot, ClaudeBot and PerplexityBot by name. Everything on the page is correct. Then the CDN or the WAF, sitting in front of the site and configured by a different team for a different purpose, answers those same crawlers with HTTP 403.
The marketing team reads the robots.txt and sees an open door. The crawler hits a wall it never reports. Search Console shows nothing, because Googlebot is usually allowed through while the AI crawlers are not.
We found this on a site whose technical content was among the strongest we measured. Best-prepared content in the study, zero effective visibility, and nobody inside the company knew.
2. The page arrives empty
The crawler is allowed. It gets HTTP 200 and the full payload, byte for byte what a browser receives. The payload is a JavaScript shell.
Three quarters of the weight is script. The HTML contains no title tag, no heading, and no body text. A human browser runs the JavaScript and sees a polished site. An AI crawler, most of which do not execute JavaScript, sees an empty document.
We tested this by asking an AI reader to summarise one such page. It returned the site’s analytics script, because that was the only text in the file.
Neither failure appears in any file-based audit or any SEO tool. Both need a real fetch, as each named agent, checking what actually comes back.
What separates the top of the field from the bottom
The useful question is what gives one operator the edge over another. Four things did, and none of them were content quality.
- Whether a crawler can complete the fetch. Binary, and it overrides everything else. A site with excellent structure that answers 403 is worth less than a mediocre site that answers 200. The gap between the best-prepared and the best-performing site in this field was entirely this.
- Whether the text exists without JavaScript. Server-rendered operators beat client-rendered ones by a wide margin regardless of how good the copy was, because the copy was not there when it counted.
- Whether the brand has a machine-readable identity. Two thirds of the field had no usable Organization schema, meaning no structured statement of what the company is, where it operates, or what it sells. Engines resolve brands, not pages. A brand a model cannot resolve is one it describes vaguely or confuses with a competitor.
- Whether the site says plainly, in text, what the business does. More sites than we expected never state it in a form a machine can extract. The information lives in imagery, in navigation labels, and in the assumed knowledge of a returning customer.
The operators at the top of this field were not the ones with the best marketing. They were the ones whose sites happened to be built in a way that survives being read by a machine. In most cases that looked like an accident rather than a decision, which is why the lead is available to anyone who makes it deliberate.
Methodology
Published so the study can be checked and repeated. This section is the citable artifact on this page.
What was measured. The public homepage of each operator, on 31 July 2026, between 12:21 and 12:47 CET. One run per site, no retries, no cherry-picking.
Technical readiness is a weighted score out of 100 across eight checks: robots.txt and AI crawler rules (18), llms.txt (18), Schema.org structured data (16), meta and Open Graph tags (14), content quality (12), freshness and language signals (6), AI discovery files (6), brand entity signals (10), less a penalty for negative signals. The breakdown sums to the headline number by design, so any score can be explained line by line.
Crawler access was tested by requesting each homepage as GPTBot, ClaudeBot and PerplexityBot and recording the HTTP status and payload. A real fetch, not an inference from robots.txt, which is why the zero-scoring failures surfaced here and would not surface in a file-based audit.
Effective score is readiness weighted by crawler accessibility. A site no crawler can read cannot realise its readiness, so the effective score falls to zero.
What this study does not measure. Live citations. These are technical readability figures, not counts of how often a brand appears in ChatGPT answers today. Measured citation data needs a per-brand prompt set and belongs to a full engagement rather than a sector scan.
Limitations. Homepage only. Single run. One date. Scores from different dates are not comparable.
Reproducibility. The scoring is deterministic. Same site, same day, same result. The full method is published at /methodology/ with a version number, so anyone can rebuild it and check our arithmetic against their own site.
Why we publish the findings and not the companies
Four things worth taking away, whatever industry you are in.
The bar is on the floor and it will not stay there. An average of 44 out of 100 across sixteen serious companies means the work required to lead is smaller today than it will ever be again. Whoever fixes their llms.txt, their structured data and their crawler access first is not competing for a marginal ranking. They are becoming the source the assistants learn the category from.
Being technically good is not enough on its own. The single best-prepared site in this study scores zero in practice. Two of the sixteen are invisible to AI crawlers while their robots.txt says otherwise, and neither company knows.
First position compounds. Assistants reinforce what they already cite. Every month a competitor holds the answer in your category, the cost of taking it back rises.
Nobody in this field is measuring any of it. We found four broken sites in an afternoon. None of those four had found the problem themselves, which tells you how much attention this is getting inside even large operators today.
What this means for your own site
Four things worth taking away, whatever industry you are in.
The bar is on the floor and it will not stay there. A field average in the mid-forties across sixteen serious companies means the work required to lead is smaller today than it will ever be again. Whoever fixes their crawler access, their structured data and their machine-readable identity first becomes the source assistants learn the category from.
Being technically good is not enough on its own. The best-prepared site in this study was worth nothing in practice. Its robots.txt said otherwise, and nobody inside the company had checked.
First position compounds. Assistants reinforce what they already cite. Every month a competitor holds the answer in your category, the cost of taking it back rises.
Nobody in this field is measuring any of it. We found four broken sites in an afternoon. None of those companies had found the problem themselves, which tells you how much attention this gets inside even large, well-resourced businesses.
What we would fix first, in order
Applies to most of the field, and to most sites outside it.
- Test crawler access with a real fetch, as each named agent. A quarter of this field had an access or rendering fault and none could see it from their robots.txt.
- Render key content server-side. A page that needs JavaScript to show text is a page assistants read as blank.
- Publish an llms.txt. Most of this field had none. It was the single largest scoring gap.
- Add Organization schema with a stable identity. Two thirds scored zero or close to it, meaning no machine-readable statement of what the brand is.
- State plainly, in text, what the company does and where it operates. Say it in a sentence a machine can lift.
Frequently Asked Questions
There is no established industry benchmark yet, which is part of why we measure. In this field nothing cleared 70 out of 100 and the average sat in the mid-forties. Treat the seventies as the current practical ceiling in European travel rather than as a good score in absolute terms.
We are not publishing that, and we would not publish yours either. The study covers sixteen major outbound operators across four European markets, selected by size and market presence. The findings above are the whole of what we release.
Yes, and the first measurement is free. You get the same technical audit the sixteen received, in full, and it goes to nobody else. No call is required to get it. If it shows nothing worth fixing, we will tell you that and leave you alone.
Technical readiness is what a site would be worth if engines could read it. Effective score is what it is worth given whether they can. When the two differ, something between the content and the crawler is failing, and that gap is usually the cheapest thing on the list to fix.
Monthly, against a fixed method, so movement is provable rather than anecdotal. Source overlap in AI answers between consecutive days runs 34% to 42%, which is why a single snapshot proves nothing.
Cite this study
Cite this study
Katsonis, G. P. / SyeniteLabs (2026). AI Visibility in European Outbound Travel: a sixteen-operator technical audit. Method version 1.0. Published 31 July 2026.
https://thysanos.com/research/european-tour-operators-2026/APA: Katsonis, G. P. (2026). AI Visibility in European Outbound Travel: a sixteen-operator technical audit (Method v1.0).
https://thysanos.com/research/european-tour-operators-2026/Plain text: In a 2026 SyeniteLabs audit of sixteen major European tour operators, no operator scored above 70 out of 100 for AI readability, the field averaged in the mid-forties, and a quarter of the sites were unreadable to AI crawlers despite permitting them in robots.txt.
Method published under CC BY 4.0. Per-company data is not released.