This report summarises the data collected in the “Writers and Publishers AI Survey”. Throughout the report, all percentages are rounded to whole numbers. Key terms are in bold and defined in the footnotes. The survey was structured into three parts.
Data collection
Responses were collected via a weblink distributed through electronic direct mail and social media (Instagram, X, Facebook, and LinkedIn). Distribution targeted writers and publishers across the combined audiences of Copyright Licensing New Zealand (CLNZ), New Zealand Society of Authors (PEN NZ Inc) Te Puni Kaituhi o Aotearoa (NZSA), and Publishers Association of New Zealand Te Rau o Tākupu (PANZ). A total of 120 complete responses were gathered between Thursday, 28th November 2024 and Tuesday, 14th January 2025.
SUMMARY OF FINDINGS
The first part of the survey asked participants to identify as a writer or publisher. Of the 120 complete responses, most participants identify as writers (81%) and this group is predominantly composed of authors, academics, poets and scriptwriters.1 Publishers (19%) were largely composed of independent publishers, academic publishers and publishing houses, with a small set of data from self-publishers and audiobook production.2
The second part of the survey requested information about participant’s attitudes toward, and experience with, generative AI systems.3 Most participants are concerned about their works possibly being used to train AI (71%), however more than half are unsure if their work has already been used in this way. Of those that know that their work has been used to train AI models, most did not consent to this use (73%).
Many participants have not used AI in their work as a writer or publisher (70%). Those that have used AI in their work as a writer or publisher have, at some point, used AI systems to assist their output (69%), used AI systems in “another way” (46%), or input their own work (36%). We did not ask about the frequency of this use of AI systems, nor did we offer an alt field to specify how AI was used in alternative ways. Should CLNZ want to understand how writers and publishers are using AI systems, further research would be required.
The final part of the survey sought responses regarding compensation and possible forms of AI licensing such as a prompt licence and a training licence or levy on output. Attitudes toward compensation are nuanced. Most participants would like to receive compensation for historic use of their work to train AI models (77%).
While many participants are in favour of receiving compensation should CLNZ be able to secure this in the future (55%), a significant portion (30%) were either unsure whether they would want to receive compensation in the future (18%) or felt that their response was dependent on contextual factors (12%). These factors include the purpose of use, extent of use, ongoing ability to give or withhold permission, potential loss of income, and the broader ethics of AI usage. Some participants specified that while they theoretically support compensation in this context, they feel it would be difficult to prove and manage.
Participants are unsure about the possibility of a prompt licence (36%). Remaining responses are slightly in favour of this option (yes: 33%; no: 31%). Participants are also unsure about the possibility of training licence or levy on output (36%). Remaining responses were slightly against this option (no: 34%; yes: 30%). This data is largely inconclusive.
Finally, the survey asked for more information regarding concerns about writing being copied and broader views on AI. This section was open field and garnered a variety of responses which have been summarised below.
- Some had strong feelings against AI and its use due to what is perceived to be an affront to intelligence and the human soul. In contrast, others strongly advocate for creators to embrace and use AI.
- Many comments indicated a limited understanding of AI, and the risks associated with it. These comments are characterised by a healthy suspicion toward, and lack of transparency from, Big These comments also indicate a lack of public forum on AI, its use, and risks.
- Many shared concerns over the possibility of their work being copied, with or without compensation, and had particular attention to less tangible aspects of written work such as themes, characters, and voice/style. Also concerned about the meanings of their works changing as a result of works being removed from context.
- Most advocated for greater guardrails against the misuse of AI and pointed to limited existing legal frameworks. A few participants suggests that copyright law already provides the tools to object to misuse or cite their BY-NC-ND Creative Commons licence as a protection. Some expressed concerns that publishers would profit from
- to summarise the content, so that they did not have to read it all. A prompt licence would not allow the inputted work to be used to train the system that is summarising the work, it would only allow the work to be used to generate an answer.
Footnotes:
1 Other groups include journalists, author/illustrators, copywriters, bloggers, game writers, teachers, and retired writers.
2 Options were provided for participants to identify as an editor, translator, news/media agency, and a manager/representative of an estate (either as a writer or publisher); however, no responses were received from these groups. Therefore, they are not represented in this data.
3 Generative AI systems, such as ChatGPT or Gemini, produce responses to questions and instructions (prompts) provided by users. These systems can do this because they have been trained to recognise works and patterns of words. The training process involves the ingestion of millions of documents into the system.
4 A prompt licence would permit content from written works to be used to prompt AI systems. For example, people working for licensed organisations could input an extract (like a chapter from a book) of a rightsholder’s work into an AI system, and ask it to summarise the content, so that they did not have to read it all. A prompt licence would not allow the inputted work to be used to train the system that is summarising the work, it would only allow the work to be used to generate an answer.
5. A training licence or levy on output would permit written works to be used to train AI systems. This means that a rightsholder’s work would be ingested in order to train the AI system itself. The work would be used, alongside thousands of others, to train the system to recognise patterns of words, so that it can then output fully formed sentences in response to user prompts. AI systems are only as good as the information that is used to train them, so organisations are looking to train their AI systems on the best source material that they can.
