10 min read11 hours ago

–

Press enter or click to view image in full size

Screenshot of the AI Overview result in Google for my search “what’s on at acmi this weekend”

Background

Late last year, the website production function of acmi.net.au moved into the Digital Products & Interaction Design team, which I lead. With that shift, we’re now looking at a new website strategy. Previously, Digital handled UX and software development, but not production, this was in the Brand & Marketing team. The “org chart” version of this change is one website producer moving teams. The practical version is that we can now manage the website as a single digital product — aligning decisions, improving workflows, and ensuring it connects cleanly with our other systems and user‑centred strategies.

Context: AI and museum visibility

In September 2024, ACMI’s Head of Research, Dr Indigo Holcombe‑James, presented Evaluating the Multiplatform Museum in the Age of AI Automation at AMAGA in Ballarat. She challenged the room to consider how AI‑driven shifts will affect how museums evaluate their impact.

Then, in mid‑2025, Seb Chan (ACMI’s Director & CEO) shared a report from Trakkr.ai — a tool that tracks website visibility across major LLMs. Seb asked for a review of how AI is reshaping search. I set myself the task of reading what peers were publishing, looking at our Google Analytics and Trakkr.ai and pulled these insights into a report: where things stand, how AI might affect cultural organisations, and what ACMI should do to stay visible in LLM results.

Around the same time, Michael Weinberg published Are AI Bots Knocking Cultural Heritage Offline?, sparking discussions about cultural collections online being scraped for data by bots for training AI models and what to do about it. Since then, helpful reports from One Further, Ash Mann, Colleen Dilenschneider, and Lauren Pope have helped shape our thinking. (All articles linked at the end of this article).

Our approach

Our team has been watching the sector experiment with LLMs for a while now. At ACMI we have been experimenting with LLMs for collection access, exhibition interactives, copywriting and editing, writing code, visitor research and more. In an environment marked by uncertainty, declining trust and rising misinformation, we’ve been thinking carefully about how and where we show up for audiences — both the ones who already know us and those who are trying to find us for the first time.

In this article, I want to share what we’ve learned about how AI is changing how people find web content, what this means for cultural organisations, how it affects ACMI’s website and goals, and how we’ve chosen to respond. I’m focusing specifically on the core user journey of planning a visit — not engagement with online collections (that’s its own conversation).

The research in summary

How is AI changing how people find web content?

1. Users are increasingly using LLMs like ChatGPT and Perplexity to research what to do, read, play and watch.
2. Search habits are shifting, including the rise of AI‑powered discovery on social platforms like TikTok.
3. AI Overviews and email summaries mean users often get what they need without clicking through to the original webpage.
4. People are using ad blockers and privacy tools more heavily, reducing what organisations can track.

Changing search behaviours

Press enter or click to view image in full size

AI usage graph from Colleen Dilenschneider, Know Your Own Bone, 2025

Research by Colleen Dilenschneider How AI is Reshaping Visitor Engagement Strategies for Cultural Organizations* *shows emerging trends that are shaping information seeking behaviour across generations. Of particular interest for us in the graph above, shows that Gen Z is more likely to adopt AI search than Baby Boomers and for Gen Z traditional search is in decline.

One Further’s graph below on AI Overview coverage from The Impact of AI Overviews on the Cultural Sector shows that the summaries that LLM provides are becoming increasingly common in the cultural sector.

Press enter or click to view image in full size

How might this impact cultural organisations websites?

Here are some of the ways we are likely to see this play out.

1. **Less organic website traffic to cultural organisation’s websites **— LLM responses and AI overviews can often reduce the need to go to an organisation’s website. This change is likely to be slow. This graph below from Benedict Evans AI Eats the World talk that shows weekly users of ChatGPT from December 2022 to December 2025. Only 5% of users are paying and there’s no data on daily use which would show a more true behaviour change.

Press enter or click to view image in full size

Graph showing ChatGPT weekly active users Dec 2022 — Dec 2025

2. Possible impact on brand awareness and museum visitation if our organisations are not coming up in results responding to relevant prompts. There are tools available, like Trakkr.ai, to see how visible our website and brand are in LLMs so it’s worth plugging into one of them and downloading the reports.

3. Possible undermining of our broader objectives and missions. If people are relying on or going to other sources for information about your organisation they might not be getting what they need, it might not be correct and you have less of a chance to communicate to them directly when you need to.

4. Google Analytics is becoming even less reliant for evaluating website goals. Privacy regulations like** **GDPR are leading to lower data collection rates when users opt out of tracking. Browsers like Safari and Firefox are increasingly block third-party cookies by default making cross-site and long-term tracking more difficult. The widespread use of ad blockers can prevent tracking scripts from loading, resulting in incomplete data for affected users. This also impacts re-marketing chances.

Get Lucie Paterson’s stories in your inbox

Join Medium for free to get updates from this writer.

5. Decrease in email sign ups and ability to re-market. Following on from the above, less people coming to our websites means less chance to subscribe them to our databases for news and therefore less ability to re-market to them.

What have we observed at ACMI?

1. Google Analytics reports that 0.002% of 2025 website sessions to acmi.net.au came from LLMs with ChatGPT accounting for 81% of that. (Web traffic is not expected when users ask a LLM for information about ‘what is on at ACMI’ because the LLM will give them an answer. It’s only if they want to know more that they might click on the source link and if that’s acmi.net.au then we would see the traffic.)
2. We saw an increase in LLM referrals in September 2025 (and ongoing) with the opening of our major playable exhibition, Game Worlds. This coincides with an increase in web traffic and a large campaign being out in the world so is not LLM specific.
3. We were already planning to do work on our SEO and this is iterative as opportunities for enhancements come up as part of website production. From our research (linked above and below in References we know this will assist greatly in LLM visibility.
4. Trakkr.ai reports have given us insight into specific areas to focus on to increase our visibility in LLMs particularly around questions we want to appear in like ‘what museums should I visit when in Melbourne?’— with the most room for improvement being on answer-friendly formatting and structured data and semantics. When ACMI’s visibility improves in LLMs, it won’t necessarily result in more web traffic, it will just result in better answers for people asking those questions in LLMs.
5. We use Google Analytics for website reporting and have for many years identified that we need to not rely on these figures alone. We must account for a % of users that have ad blockers, don’t accept cookies and use browsers that block these tools. And this % is increasing. We have invested in using Askable for user testing to have another avenue for insights to improve the website UX.

Press enter or click to view image in full size

Graph showing traffic to acmi.net.au from LLMs in 2025

Press enter or click to view image in full size

List of LLM referrers to acmi.net.au in 2025

What’s our plan to respond to this?

There are two key principles that guide website decision making:

1. We respond to improve the experience: as a visitor-centered museum, we are committed to responding in a way that improves the user experience. It would be irresponsible for us not to respond to this shift in how audiences are finding our content. We are a public institution and access is a cornerstone of our mission. However, we will not compromise the UX for the machines! In doing this work, we do not need potential visitors to come to our website, but we do need them to get accurate information about what we do, have on and how to get here.
2. We need to remain sustainable: optimising the website for ticket sales and other revenue generation is essential for ACMI’s long-term health.

Goal

We need to optimise our online presence for AI consumption, not just human readers, but machines too

You need to write for humans and engineer for machines — making your claims explicit, structuring relationships, and publishing in ways that feed both symbolic systems (search engines) and learned models (LLMs). The brands that thrive won’t be those who only create great content; they’ll be the ones who encode their meaning in ways machines can trust, reuse, and amplify.

Jono Alderson, URL shaped web

Our progress

This work is ongoing. This is what we are currently up to.

1. Introduced an accordion module for FAQs into the CMS and published FAQs for collections and exhibition pages. The accordion is getting used in all sorts of ways so we need to ensure the mark up is correct for FAQs so the machines read it correctly and give it the right credit.
2. Focus on Jafar Panahi event page optimised for LLM and evaluated. Given the small audience for this event and the fact that most of film ticket sales come from our EDMs this was hard to measure success. It was most useful to use as an example for discussions across teams and understanding what compromises there might be leading with machine optimisation.
3. Added press quotes to Game Worlds exhibition event page to improve the credibility.
4. We have discussed updating the editorial guidelines for machines and AI as well as humans.
5. We have created a ‘project’ in our enterprise CoPilot account to experiment with AI copy editing event page copy.
6. We are reviewing SEO iteratively with each project or page build implementing improvements.

Transactional content is holding up well. Informational content, the kind that answers questions and provides educational value, is under pressure from AI Overviews. That doesn’t mean abandoning informational content. It means being more strategic: focusing on depth, originality and expertise that AI can’t easily replicate, and ensuring your technical SEO foundations are solid enough to compete for the visibility that remains.

Beth Downs, SEO visibility 2025–2026: how 100 UK arts organisations are performing

What’s next?

• We have talked about investing in deeper, longer pieces of writing again (post website strategy) and how we might put access to this content behind email address/phone number related to Lens engagement or membership. This conversation was inspired by an article where 404 Media did just this. The approach for us would be two fold, it would demonstrate to users that there is a cost and value to producing and hosting this content and it’s a blocker for LLMs harvesting this richer content without our permission. In extension to this we are considering what might we learn/borrow from alternatives like the fediverse and Indiweb.
• Specific research into voice assistants and AI agents and their possible impact for the website and opportunity for users. We are already gearing up for agentic shopping for the online shop. This would mean users on LLMs could ask for “gifts for under $50 for my nephew who like videogames and arthouse movies but don’t like books”.
• Getting a better grip on what evaluation of this might look like. The small amount of traffic from LLMs and so many external influences makes this challenging. We certainly have not had any negative feedback from users or internal stakeholders. So far we know improvements in visibility in Trakkr.ai without compromising the user experiment would be a good start.

Conclusion

At this stage, we’re treating this shift as an opportunity to keep learning in public. AI‑driven search is still evolving, and cultural organisations are learning alongside their audiences. Our approach at ACMI is to stay open, experiment thoughtfully, and share what we discover along the way. We may not get every step right, but by testing, questioning and collaborating across the sector, we can help shape a web ecosystem where culture remains visible, trustworthy and easy to find — whether the visitor is a person or a machine.

References

*Thank you to *

, Vera Samardzic and ** for reviewing the post. To ** for prompting that we do this research and for Jo Greggains who championed this work during her time at ACMI. I used CoPilot for copy editing and title suggestions. The copy editing was too reductive so I trusted myself and my colleagues instead!