Luna Studios Deutsch →

Luna Studios Research · The second metric

The Reading Index.

AI systems read the web — whether a website wants it or not. People increasingly ask their questions there and get summarized answers built from whatever is readable. The Reading Index measures what share of business websites take part at all — annually, with an unchanged method, as the counterpart to the Speaking Index.

Introduced August 2026 4 countries 659,095 reachable websites checked

Reading Index 2026 · all four countries

13.36

out of 100

Of 659,095 reachable business websites in Switzerland, Germany, Austria and the United States, 88,073 provide AI systems a readable fact base without locking them out — one in 7.

Definition

What the index measures.

The Reading Index of a market is the share of business websites, in percent, that provide AI systems a machine-readable fact base and do not block their access.

The count is strict: only a genuine, substantive fact file for AI systems in the website's root directory counts — error pages, redirects and empty files are weeded out individually. A website that provides such a file but locks AI systems out via its crawler rules does not count. The German name of the metric is Lese-Index.

The starting point

AI reads anyway. The question is what it finds.

That is good news, first of all. More and more buying decisions begin with a question to an AI system instead of a search: for the restaurant, the contractor, the practice, the provider. The systems answer with what they can read — whoever is readable can appear in those answers with their own prices, hours and services.

But the large majority of business websites do not appear in them at all: they provide nothing readable. To AI systems these companies either do not exist — or only second-hand, through directories, portals and outdated listings the company has no influence over. What cannot be read disappears from the answers, the way websites without a mobile version slowly disappeared from everyday life after 2010.

2026 values

The current values.

MarketReading Index 2026One readable website inFact file presentAI actively blockedBase
United States19.79520.1 %3.54 %275,439 reachable websites
Switzerland14.24714.4 %1.72 %37,862 reachable websites
Austria9.09119.2 %1.25 %44,151 reachable websites
Germany8.01128.5 %1.76 %301,643 reachable websites
All four countries13.36713.7 %2.46 %659,095 reachable websites

For comparison: the Speaking Index — can the website answer a visitor on its own? — sits between 0.05 and 0.12 in the same markets. Websites can thus be read by machines roughly two hundred times more often than they can answer people. The gap is widest in the US: the most readable market of the measurement (19.79) is also the one whose websites are least able to answer (Speaking Index 0.05).

The finding

And where reading happens, almost nobody controls it.

The most remarkable finding is not the level of the index but its origin. A deep sample of 1,587 readable websites shows: about 74 percent of the fact files were not created by the website owners but generated automatically — by website builders (44 %), content-management-system extensions (23 %) and shop systems (7 %). This automation summarizes whatever is on the website — nothing more, nothing less.

The web is becoming readable without its owners ever deciding it — the platforms are rolling it out. What the AI actually reads is in almost nobody's hands.

Combined with the Speaking Index measurements this paints a clear picture: most websites state no prices, no response promises, often not even unambiguous opening hours. An automation that summarizes such pages carries exactly these gaps into AI answers — and the systems fill them second-hand. Being readable is the first step. Controlling what is read is the real one.

This is precisely where the two metrics of the series differ: the Reading Index rises on its own, because platforms ship readability as a default. The Speaking Index only rises when a company decides to teach its website to answer. One number describes infrastructure — the other an attitude towards the visitor.

The other direction

Who locks the AI out?

2.46 %explicitly block AI crawlers (website crawler rules)
8.0 %mention AI crawlers in their crawler rules at all
92.0 %have visibly never engaged with the question

The vast majority of business websites has no position on the defining distribution question of the coming decade — may AI read and relay my content? Not out of rejection, but out of absence.

Perspective — clearly separated from the data

What this means for your business.

The order has reversed: first the AI reads — then it decides what to say about a company. Whoever does not maintain their own fact base leaves that answer to an automation, or to chance. Websites without structure, prices or current information get summarized anyway — just incompletely, outdatedly, or not at all.

Three things can be done right away: provide and maintain a fact base for AI systems instead of leaving it to an automation; consciously allow AI crawlers instead of ignoring the question; and structure the content so that prices, hours and services are unambiguously readable.

Where does your website stand?

Luna, the assistant on this page, looks at your website in real time when you tell her your domain. She is also the counter-model to uncontrolled readability: a Luna speaks exclusively from a curated knowledge base approved by the company — the same foundation that can be provided to AI systems deliberately. That way the company itself decides what is said about it: to visitors and to machines.

Measurement rules

How it is measured.

1Population: the same reachable business websites as the Speaking Index — openly licensed geodata (OpenStreetMap, ODbL), region by region
2Two requests per website: the fact file for AI systems in the root directory, and the website's crawler rules
3Strict authenticity check of the fact file: error pages, redirects and empty responses do not count; a deep sample verifies content and origin
4Only websites that are readable AND do not block AI crawlers enter the index — checked against 14 common AI crawler identifiers

Limits: the measurement checks the common standard location of the fact file; other forms of machine readability (structured data in the page source) are not covered here. The crawler evaluation follows the website's published rules — whether a provider honors them is a different question. The values are approximations with a documented method, not a full census of every form of readability.

Time series

2026 is the baseline.

The Reading Index is updated annually together with the Speaking Index. The interesting question of the coming years is the distance between the two: how quickly does the web become readable — and how long does it stay mute while it does? The 2027 edition is scheduled for summer 2027.

Whoever defines both metrics of a category and owns their time series gets cited for as long as the category exists.

Usage

How to cite the index.

All values may be used freely. Recommended citation:

"Reading Index 2026 (all four countries): 13.36 — Luna Studios, lunastudios.ai/en/research/reading-index"

Data and reports

The aggregated values are available as open data; charts and quote blocks live in the press area (German, with language-neutral data).