Written for Whom? The Readability Crisis in National AI Strategies

A national AI strategy is, in theory, a public document. It announces a government's intentions to citizens, industry, and civil society. But the language in which these documents are written tells a different story — one that raises uncomfortable questions about who AI governance is actually for.

We computed Flesch-Kincaid Grade Level scores for 72 strategies in the AI Folio corpus where full text extraction was possible. The mean grade level is 19.2. To put that in context: Grade 8 corresponds to an average newspaper; Grade 12 to a secondary school leaver; Grade 16 to an undergraduate degree. Grade 19.2 is postdoctoral territory.

The distribution is almost entirely above Grade 12. Seventy-one of the 72 scoreable strategies are classified as "Technical" — the highest readability band in our scheme. Only one strategy scores in the "Specialist" band. Not a single national AI strategy in the corpus achieves "Accessible" or "General" readability.

The most readable strategy in the corpus is Sweden's AI Strategy at Grade 14.5 — still "Specialist" level, still requiring university reading ability. The least readable strategies include those from Portugal, Pakistan, Peru, the United Kingdom, and Scotland, all scoring at Grade 20.

Two caveats are important. First, 32 of the 104 strategies are written in languages other than English, and our analysis runs on the original text. Translation artefacts and cross-linguistic variation in readability metrics mean these scores should be interpreted carefully for non-English documents. Second, Flesch-Kincaid was designed for English prose and systematically produces higher scores for technical and legal language regardless of true comprehension difficulty.

These caveats do not, however, explain away the core finding. Even accounting for methodological limitations, a mean grade level of 19.2 across documents that are nominally addressed to the public represents a significant democratic accountability gap. AI governance decisions — about surveillance, automated public services, labour market disruption — affect everyone. The documents that announce how governments will manage these decisions are readable by almost no one outside the specialist community.

The democratic implication is uncomfortable: most national AI strategies are not written for the publics they claim to serve. Whether this matters depends on whether you believe AI governance is a technical or a political question. The evidence of this corpus suggests most governments have already answered that question.

Chart: Distribution histogram — number of strategies at each Flesch-Kincaid grade level band, with reference lines for newspaper (8), secondary school (12), undergraduate (16), postgraduate (19.2 mean).

Data: AI Folio Corpus Metrics, Flesch-Kincaid analysis of 72 national AI strategy documents with extractable full text.