How I assess the quality of Italian text: audit methodology

Methodology · Version 1.0 · Published October 6, 2026 · Updated October 6, 2026 · by Alessio Di Rubbo

A text can rank well, be cited by an AI assistant and still read like a translation. This page describes how I assess that layer: the linguistic quality of an Italian text, whoever wrote it.

What the audit checks and what it does not

The audit checks the quality of the Italian, from grammatical correctness to naturalness. It does not check factual accuracy and it does not replace a technical SEO or GEO audit. It complements both where they do not look: whether the text is written the way a native speaker would write it.

The six error categories

Agreement

Wrong gender or number agreement in articles, adjectives and participles.

Example: “le informazioni sono stati aggiornati” instead of “le informazioni sono state aggiornate”.

Register

Switching between tu, Lei and voi, or a tone that shifts within the same text. The error is the inconsistency, not the choice of register: a text can use tu or Lei, as long as it sticks to it.

Example: a text that opens with “Le consigliamo di verificare” and two lines later says “ti basta cliccare”.

Calques

Non-idiomatic constructions carried over from English.

Example: “questa soluzione fa senso per il tuo team”, from makes sense, instead of “ha senso”; “ho realizzato che il file era sbagliato” instead of “mi sono accorto”.

Syntax

Structures that are grammatically correct but sound artificial to native speakers, usually because of word order or heavy use of the passive and the gerund.

Example: “L’ordine verrà inviato da noi entro 24 ore” instead of “Spediremo l’ordine entro 24 ore”.

Terminology

Terms used inconsistently within the same text, false friends, technical terms left unadapted.

Example: “carrello” and “cestino” for the same element in two steps of the same flow; “eventualmente” used in the sense of English eventually.

Fluency

Correct but unnatural text that “sounds translated”. This is the most subjective category, so I always pair it with a suggested rewording.

Example: “Il tuo ordine è stato ricevuto con successo” instead of “Abbiamo ricevuto il tuo ordine”.

The three severity levels

Critical: affects the comprehension or the credibility of the text.
Medium: lowers the perceived quality without blocking comprehension.
Minor: a stylistic note that does not get in the way of communication.

How I report the findings

Each finding includes the passage, the category, the severity, an explanation and a proposed correction. The report sums up the figures by category and by severity, because a single score hides where the problem lies.

The score in the online demo is an estimate produced by a language model. It works as a quick indicator, but it is not a calibrated measurement. In a real audit, what counts is the list of findings, reviewed by me.

Recurring patterns in Italian text from AI models

These are the areas where I most often find problems in my evaluation work:

the generic masculine, even where Italian requires explicit agreement;
the subjunctive, systematically replaced by the indicative;
the gerund, used more often than in natural Italian text;
punctuation borrowed from English, such as a colon followed by a capital letter;
no distinction between the tone for a B2B setting and the tone for a B2C one.

Limits of the method

The audit is based on a sample of text, not on the full corpus of a website or a product. The syntax and fluency categories partly depend on the reviewer’s judgement. That is why the report always states the criterion applied, and where a choice is legitimate, such as a register or an approach to gender-inclusive language, I assess how consistently it is applied, not the choice itself.