How are arts and cultural organisations actually using AI?
Declaration: AI wasn’t used in the writing or editing of this article. For better or worse, a messy human brain made it.
ulysse.gks & FARI / https://betterimagesofai.org / https://creativecommons.org/licenses/by/4.0/
Over the past few years, I’ve worked with just over 90 clients in the arts and culture sector around AI, and it’s given me an interesting window into how individuals and teams are actually beginning to experiment with AI and what some common use-cases already are. I thought it could be helpful to condense some of those findings here.
These findings (all since 2023) emerge from a few sources. These are informal conversations with practitioners, group discussions during workshops, internal SLT and board discussions, internal staff surveys provided to me, long-form projects I’ve supported like The Audience Agency’s Let’s Get Real AI (see report), and from 398 pre-session questionnaire responses I’ve received from my own surveying efforts to the sector (I usually make a survey available before I run an internal programme at an organisation). I’ve also taken into account how these findings are shifting over time, and what those shifts tend to centre around.
It’s important to note that none of these findings are presented as best-practice suggestions or advice.
I’m presenting use cases these only as what I’m seeing actually take place with the arts and culture sector, rather than suggestions of positive or negative applications. Strictly speaking, I take no stance on that myself, and my work encourages each individual and each organisation to research, discuss, test and develop the work processes (or critical objections) which make most sense for their values, understanding of bias, and their perception of quality.
It’s also worth acknowledging that there is often a sense of stigma around using AI within a person’s role, particularly within arts and culture environments. I would suspect this leads to significant underreporting around actual use, although I’d imagine the actual categories of use may remain consistent.
Quick note: I sometimes hear the assumption that AI use is broadly coordinated/adopted by particular teams and wholesale used by members of that team (e.g., a whole team agrees to begin using a particular AI tool for some of their transcription efforts and this is built into the ‘official’ workflows for the team). Worth stating that I don’t generally see this with my clients (with the exception of experiments around accessibility, such as subtitling). When I do hear about genuine AI use or see it reported, it almost always comes from individuals who have made a decision to integrate it into their personal working process (either with or without official approval or guidance to do so). That said, Copilot tends to be the dominant suite of tools which are made available to staff, in the event that a subscription service is being paid for.
Important mention - Dyslexia and neurodiversity support - When I encounter individuals who most passionately state that AI has offered them genuine benefit within their professional roles, I’ve found that this often relates to how it interacts with aspects of self-reported neurodiversity. This often takes the shape of how they stay organised, schedule meetings, adjust their writing within emails, interpret ambiguous requests or support their own personal admin (filling in forms, in particular). I’ve included it as a special mention because this category isn’t about a specific tool or process, but a general demographic of AI users who seem disproportionately (when compared with other members of their team) engaged with AI tools to support their role.
#1: Copywriting - Copywriting was one of the earliest known professional applications for LLMs and it remains, by my interpretation, overwhelmingly the most common, and that’s reflected within the culture sector. I don’t believe this is often being used to create wholesale blog posts, brochure copy or articles (although exceptions abound), but it seems more common for internal emails, reducing word-counts for existing human-authored copy, and perhaps some forms of marketing ideation (the most common case being email subject lines for marketing emails).
#2 Research - This can take the shape of using an LLM a little like a search engine, to learn about and become familiar with a particular topic. It can also take the form of using Deep Research or some agentic tools to pull together diverse sources and produce a report or piece of analysis. In some instances I’ve heard of marketing staff researching particular events this way, especially if the speaker or performer might be considered controversial to some audiences.
#3: Summarising documents and reports - This seems to be common across the board, and often specifically flagged as a ‘green’ or acceptable use case for AI among many organisations. A common use for this seems to be summarising meetings and pulling out tasks which have been identified. For artistic practitioners, I’ve had many discussions on this process being used to better understand funding applications or calls for submissions, to help better identify the brief and identify whether they might fit the essential criteria.
#4: Audience segmentation - For marketing roles, a common use here is to do something like upload the PDF of the brochure for an upcoming season, and work with the model to identify links between shows or events relevant to a particular audience (to develop buyer crossover, personalised emails and ‘you might also like’). Some teams appear to be going further here, depending on the audience data they have available and what they’ve been told they can use (e.g. using Claude to create data dashboards to reveal unexpected booking trends or pathways and then use that information to reinforce their paid activity).
#5 Understanding how to begin a task - This could fall under ‘learning’ in general, but could also be considered brainstorming. The task is described to the LLM and a structured way of approaching it is suggested, giving some reassurance to the user that they are approaching it in a useful way. I’ve had conversations around this specifically in relation to stakeholder event planning, for instance.
#6 Subtitling, translations and alt-text generation - These seem to be increasingly common, particularly for subtitling videos and general efforts towards using AI for accessibility. From clients I’ve worked with, alt-text generation from LLMs is sometimes considered a helpful part of the process and sometimes considered too poor to deliver consistent results, so it seems to depend on the individual or team.
#7 Image cropping and editing - This has only really emerged this year as a consistent use case I’m hearing about. This seems to involve using LLMs like Claude or ChatGPT to create multiple resizes of a particular image for campaign assets, although I haven’t met anyone who has openly discussed creating new marketing assets directly with image models (although I’ve seen this in the wild so it does seem to be happening occasionally).
#8 Campaign/activity scheduling - Particularly relevant for marketing and comms teams, it seems that using LLMs alongside a calendar of planned marketing activity is becoming common practice, particularly when considering how to combine activities or schedule new opportunities into the marketing mix. This might also be useful for tracking paid marketing efforts, when they go live, what assets are needed at a particular time, etc.
Once again, I’m not looking to endorse or criticise any of these uses, but they do appear to be the most common among the widest sample of people within arts and culture organisations. Perhaps more interesting, however, are the more niche use-cases, although these are much more difficult to generalise as they’re usually most supportive of one person’s particular brain and particular style of working.
