AI for publishers and literary agencies: rights, catalogs, royalty statements and manuscripts
Three Italian publishers in four already use AI tools, according to the Italian Publishers Association. The most useful application is also the easiest to defend: getting AI to answer from your own contracts, royalty statements and metadata, with the clause and the page, without handing anyone your books to train a model.
- 8,645 translation rights sold abroad in 2025, a record since 2001
- 75% of publishers already use AI tools (AIE survey, late 2025)
- 28% have been asked to license their content for a language model
Key takeaways
- In 2025 Italian publishers sold 8,645 translation rights abroad, a record since the Italian Publishers Association (AIE) began counting in 2001, and bought 9,470: more than 18,000 contracts a year, each with territories, formats and deadlines.
- Three publishers in four already use AI (AIE survey, late 2025), 92% of them in editorial work. 28% have been asked to license their content for a language model; of those, only 4% have signed.
- Letting AI read your own contracts and statements is not handing your books over for training: it consults the documents when you ask and cites the source. Whether your content may train a model is up to the contract with the provider.
- Three rules govern training: the text and data mining opt-out (EU Directive 2019/790, Art. 4), the transparency duties of general-purpose models (AI Act, Art. 53, since 2 August 2025) and Art. 70-septies of Italy’s copyright law, added by Law 132/2025.
- In Italy, whoever licenses an author’s rights must report to them at least every six months on who exploits the work, how and with what revenue (Art. 110-quater, Law 633/1941).
01 What Italian publishing is worth, and why rights are a memory job
According to the Report on the State of Publishing in Italy 2026 by the Italian Publishers Association (AIE), presented in Frankfurt on 7 October 2026, total book sales in Italy came to €3.198 billion in 2025, down 1.2%, counting trade, educational, professional and academic publishing, exports and sales to libraries. 2026 started better: in the first nine months, print trade books grew 3.2%.
The number that matters to anyone working on rights is another. In 2025 8,645 contracts were signed to sell translation rights to Italian books abroad, almost five times the 1,800 of 2001 and the best result since AIE began its surveys, and rights to 9,470 foreign titles were bought. Italian books went into 45 languages and 64 countries.
| Item | Value | Change |
|---|---|---|
| Total sales, 2025 | €3.198 billion | -1.2% |
| Print trade books in trade channels, 2025 | €1.484 billion | -2.1% |
| Translation rights sold abroad, 2025 | 8,645 | +1.9% |
| Translation rights bought, 2025 | 9,470 | -3.5% |
| Print titles published, 2025 | 84,737 | |
| Titles in print (the living catalog) | 1.53 million | |
| Print trade books, first nine months of 2026 | €1,026.5 million | +3.2% |
Sources: AIE, Report on the State of Publishing in Italy 2026 (summary) and releases of 30 January and 7 October 2026; market data by NielsenIQ BookData.
In practice that means more than 18,000 translation contracts a year, each with its own territories, languages, formats, deadlines and options. At this year’s Frankfurt Book Fair there were 134 Italian exhibitors, literary agents included. The question that matters most is also the simplest: are these rights still available?
02 Where the knowledge of a publisher or a literary agency lives
It doesn’t sit in one program. It sits in contracts from different years, statements in different formats, spreadsheets kept by one person and the memory of whoever has looked after an author for twenty years.
| Where it is | What it holds | The typical question |
|---|---|---|
| Publishing and agency contracts | Rights by territory and format, terms, options, advances | “Are the German rights still available?” |
| Foreign rights contracts | Translation licenses, expiry dates, royalties | “When does the Spanish license expire?” |
| Royalty statements | Copies sold, revenue, advance earned out | “How much does the author get this half year?” |
| Catalog metadata | ISBNs, formats, prices, title sheets | “Which backlist titles have an audiobook?” |
| Submissions and manuscripts | Synopses, reader reports | “Have we read this author before?” |
| Letters, emails, book fair notes | Promises, first refusals, editors’ wishes | “What did we promise this author in 2009?” |
| House style guides | Spellings, quotations, conventions | “How do we write dates in non-fiction?” |
Example questions written by us from the typical documents of a publisher and an agency.
Keeping this memory in order is also the law. Since 7 June 2022, Article 110-quater of Italy’s copyright law has required anyone who has licensed or acquired an author’s rights to inform the author at least every six months about who exploits the work, sub-licensees included, how, with what revenue and with what remuneration due. AGCOM, the communications authority, can fine up to 1% of turnover, and missing information makes the law presume the remuneration inadequate.
Article 110-quinquies adds a right to further remuneration when the agreed fee turns out disproportionately low; Article 110-septies lets the author terminate the contract or revoke exclusivity when the work is not exploited within the terms. Knowing whether a title might revert, or whether a statement is complete, means finding the clause in seconds.
03 Which questions can AI answer at a publisher or an agency
An AI agent, a program that reads the documents you give it and answers citing its source, replaces no one. It removes the first step, the hunt for the right clause or statement:
- Foreign rights: what is available, where and in which format; what is optioned and until when.
- Accounts: what an author is owed, whether the advance has earned out, which line doesn’t match the contract.
- Editorial: house conventions for names, dates and quotations; which version of a text was approved.
- Sales and marketing: title sheets, formats and prices consistent with the catalog metadata.
- An agency at a book fair: the rights guide with the territories still free; what an editor has bought lately.
Two conditions decide whether people keep using it. Every answer must cite the document and the page, because no agent confirms rights to a foreign publisher on a sentence they cannot check. And when a document is missing, the AI must say so: rights confirmed as free that turn out to be optioned cost more than a slow answer.
04 What the rules on AI and copyright say
The debate is almost always about one thing: using works to train general-purpose models, the large models behind assistants such as ChatGPT, Gemini or Claude. Four rules matter today.
EU Directive 2019/790, Article 4. It allows text and data mining, the automated analysis of digital text and data to extract information, on works one has lawful access to, unless rightholders have reserved that use “in an appropriate manner”, for example by machine-readable means for content online: the opt-out. In Italy this is Article 70-quater of the copyright law.
The AI Act, Article 53. Since 2 August 2025, providers of general-purpose models must have a copyright policy, respect opt-outs and publish a “sufficiently detailed summary” of their training content on a Commission template, which asks among other things for the top 10% of internet domains by content scraped. Models placed on the market before then have until 2 August 2027; since 2 August 2026 the Commission can fine up to 3% of worldwide turnover or €15 million, whichever is higher. The July 2026 Digital Omnibus on AI did not change these duties.
The EU code of practice. Published on 10 July 2025, voluntary and signed by more than twenty providers, it commits them to respect robots.txt, the file a website uses to tell automated programs what they may read, and the other machine-readable opt-out protocols. The Commission consulted on those protocols until 23 January 2026.
Italy’s AI law. Law no. 132 of 23 September 2025, in force since 10 October 2025, protects works of “human” intellect, AI-assisted ones included when they result from the author’s intellectual work; allows text and data mining through AI models within the existing exceptions, opt-out included (new Article 70-septies); and punishes mining in breach of them (Article 171). What else the law changes for companies and professionals: Italy’s AI law explained.
| Rule | What it provides | Since |
|---|---|---|
| Directive (EU) 2019/790, Art. 4 | Text and data mining allowed unless rightholders opt out | In Italy, Legislative Decree 177/2021 |
| AI Act, Art. 53 | Copyright policy, opt-out respected, public training summary | 2 August 2025; fines from 2 August 2026 |
| EU code of practice | Voluntary commitment to robots.txt and opt-out protocols | 10 July 2025 |
| Law 132/2025, Arts. 25 and 26 | Human authorship; AI mining within the existing exceptions; penalties | 10 October 2025 |
| European Parliament resolution | Not binding: remuneration, transparency, presumption of use | 10 March 2026 |
Sources: EUR-Lex, Normattiva, European Commission, European Parliament Legislative Observatory. Texts in force on 10 October 2026.
The debate is still open. On 10 March 2026 the European Parliament adopted, by 460 votes to 71 with 88 abstentions, a non-binding resolution asking for fair remuneration, full transparency on sources and a presumption that works were used when providers fail their transparency duties. The directive provides for a Commission review no sooner than 7 June 2026.
Companies that use AI have different duties, starting with the staff training required by Article 4 of the AI Act: see You are already an AI deployer.
This article describes the rules and is not legal advice: for a contract or a specific case, talk to a copyright lawyer.
05 Using AI on your own documents is not handing your books over for training
All these rules concern training: a model learns from millions of texts, which shape how it works for anyone who uses it. Letting AI read your contracts, statements and metadata is something else.
It is called RAG, short for retrieval-augmented generation. RAG is the method by which an AI, before answering, searches the documents you gave it for the relevant passages and builds the answer on them, showing where it took them from. The model learns nothing: it reads when you ask, like a colleague with the archive open.
RAG does touch copyright when a service uses it on other people’s works, say books or articles summarized for its users, and the European Parliament names it next to training. A publisher querying its own contracts is the opposite case: the company’s documents, read for the company. That line belongs in writing. Before choosing a tool, ask:
- Is our content used to train models? It must be no, in writing, for the model providers behind the tool too.
- Where is it stored? Contracts hold authors’ personal data: an archive in Europe is safer.
- Who sees what? Agents their authors, accounts the statements: permissions must bind the AI too.
- Do answers cite the source? Otherwise nobody can check them before they reach a foreign publisher.
- What about manuscripts? The rights to texts sent to you for reading are not yours: decide how to treat them, and say so in your submission guidelines.
Protecting your catalog from other people’s training is different. Online, the reservation must be machine-readable: besides robots.txt there are protocols such as TDMRep, from a W3C community group (not an official standard), which also works in EPUB and PDF metadata. Write the reservation into contracts and licensing terms as well. Licensing is the other path: according to AIE, 28% of publishers have received requests, but only 4% of those have signed.
This is how ekory works. Ekos, our AI agents, read contracts, options, statements and the catalog where they already are, in Google Drive, SharePoint, OneDrive or Notion, and answer with the clause and the date; each person sees only what is shared with them. Your content is never used to train AI models, as our contracts with the providers, including Google, rule out, and the archive is stored in Europe. Examples are on the pages for literary agencies and publishers; the details in security and data.
06 How to start: four steps for a publisher or an agency
- Pick one family of questions. Rights first: “are they available?”, “when does it expire?”. They are frequent, and mistakes there cost the most.
- Connect the folders the contracts are already in. No migration: the right tool reads files where they are, scanned PDFs included.
- Decide who sees what, by agent, territory or department, before opening the tool to everyone.
- Test it on real questions, including ones where it must say it doesn’t know, and count how many questions still reach your most experienced person.
You need contracts in order and one person who can tell a right answer from a wrong one. That person already exists: it’s the one everyone asks today, as in The bottleneck is your best person.
07 Methodological note
- Sources: AIE documents, legal texts on EUR-Lex and Normattiva, European Commission and European Parliament pages, read in the original on 10 October 2026.
- More than 18,000 contracts: our sum of rights sold (8,645) and bought (9,470) in 2025.
- Use of AI: AIE survey of its members, late 2025, reported in the 2026 Report; it covers members, not the whole industry. The 4% is on publishers that received requests.
- Typical questions: the table in section two is our illustration, not a survey.
We left out e-book and audiobook sales, because the January release and the Report summary give different figures, and the number of literary agencies in Italy, which no public survey measures.
Frequently asked questions
Can AI tell whether translation rights are still available?
Yes, if it reads the contracts and options for that title: it gives the status, the expiry date and the clause it comes from, so the agent can check before confirming. If a contract is missing from its documents, it must say so instead of guessing.
If we use AI on our catalog, do our books end up training AI models?
Not if the contract with the provider rules it out: an AI that reads documents when you ask learns nothing from them and doesn’t reuse them for other customers. Get that in writing, and for the texts of the works check what your author contracts say.
How can a publisher opt out of AI training?
By reserving the right “in an appropriate manner”, as the EU directive says: for content online, by machine-readable means such as robots.txt, which signatories of the EU code of practice commit to respect, or TDMRep, which also works in EPUB and PDF metadata. Write the reservation into contracts and licensing terms too.
Is a book written with the help of AI protected by copyright in Italy?
Since 10 October 2025, Italian law protects works of “human” intellect, even when created with AI tools, as long as they result from the author’s intellectual work (Article 25 of Law 132/2025). Protection follows the human contribution, not the tool.
What do companies that build AI models have to disclose?
Under Article 53 of the AI Act, since 2 August 2025 providers of general-purpose models must have a copyright policy, respect opt-outs and publish a summary of their training content on the Commission’s template. Models placed on the market before that date have until 2 August 2027; fines apply from 2 August 2026.
How often must a publisher report revenue to authors in Italy?
At least every six months, under Article 110-quater of the copyright law: who exploits the work, how and with what revenue, sub-licensees included. AGCOM supervises and can fine up to 1% of turnover.
Sources
- AIE, Report on the State of Publishing in Italy 2026, summary, October 2026.
- AIE, Frankfurt press release on the record sales of translation rights, 7 October 2026.
- AIE, release on the 2025 trade book market (in Italian), 30 January 2026.
- Regulation (EU) 2024/1689 on artificial intelligence (AI Act), Arts. 53, 101, 111 and 113, EUR-Lex.
- Regulation (EU) 2026/1744 (Digital Omnibus on AI), 8 July 2026, EUR-Lex.
- Directive (EU) 2019/790 on copyright in the Digital Single Market, Arts. 2, 3, 4 and 30, EUR-Lex.
- European Commission, template for the public summary of training content, 24 July 2025, and its questions and answers.
- European Commission, General-Purpose AI Code of Practice, 10 July 2025.
- European Commission, consultation on protocols for reserving rights from text and data mining, 1 December 2025.
- Law no. 132 of 23 September 2025 (in Italian), Arts. 25 and 26, Normattiva.
- Law no. 633 of 22 April 1941 on copyright (in Italian), Arts. 70-quater, 110-quater, 110-quinquies and 110-septies, Normattiva.
- European Parliament, Copyright and generative artificial intelligence, procedure 2025/2058(INI), resolution of 10 March 2026, Legislative Observatory.
- W3C, TDM Reservation Protocol (TDMRep), Final Community Group Report, 10 May 2024.