Huge huge thanks and kudos to AI legal legend, Dr. Barry Scanell for not only highlighting the Getty Images vs (Stability AI) Stable Diffusion legal case in the UK, but also for providing a significant level of valuable insight here, and even more with his colleague, Leo Moor, here.
In brief: Getty Images suggested in court that new AI-generated content, specifically the models that power image generators like Stability Diffusion, break copyright rules when trained upon, and asked to reproduce, copyrighted materials.
The key issue is that when asked to create content that may be copyrighted (produce an image that looks like [person, place, thing] that the model knows what criteria it needs to consider, and what calculations it needs to perform, to carry out that request. That knowledge is the copyright material.
The result, yesterday: Justice Joanna Smith ruled that as the models themselves do not store or reproduce training data, the output they generate merely reflects patterns they have learned during training, not the data itself.
Theres a few really important bits that stand out and have significant consequences:
Teaching from copyright material is not illegal: Model weights (learned behavior) are the “distilled result of learning patterns”, not actual storage of the original photographs or text. They do not store the content itself.
This creates the important distinction between:
Data Ingestion/Training: Copying can occur when data is initially materialized and stored to be used for training.
The Trained Model: The resulting AI model is not itself a copy of the training data.
Couple of other things caught my eye this week, and both relate to the ongoing discussion around AI Search.
Someone I love to read, for both the cheekiness (and brilliance) of his takes is Parry Headrick of Crackle PR who last week launched an AI-Engine Optimized News Release template. At a time when nobody truly knows the answers to this area of search, it’s a bold and admirable move from Parry to put his neck out there and see what the reaction is to the template. “Great so far” is the answer!
As I remarked to Parry, it reminded me very much of the importance of the Social Media Press Release template way back when brands’ social media newsrooms were a novel and emerging platform for owned news. It looks to be a really good template and I’m glad that someone has dared to put something out there rather than pontificate about thin air.
That said, when I look at the SEO industry’s approach to the same challenge, I see a more technical, quantified, and measurable approach. Both sides seem to have a similar grasp about what factors matter, and where they should be on a page—but at this stage, I’d still prefer to put my AI-optimized content into an SEO page template built using an innate understanding of search intent, and which can give me demonstrable results.
Related to this, James Crawford of Agency One released a fascinating and extensive report called “Prove me wrong” looking at the state of “GEO” and PR. Same thing as AIO, different acronym.
Like anyone exploring this space, kudos to James for pulling such a comprehensive review together.
One of the most interesting ideas was “Share of Model” (ie. the % of times our brand is mentioned out of all prompt results or prompts that relate to our industry/product etc.). It offers a commonly-understood approach to deciding whether our brand is being seen more than others. This, plus any associated traffic from the GenAI tool seems a very reasonable way of determining PR impact.
However, as James rightly points out, the fact that each AI tool has different approaches to context interpretation and prompts can vary significantly (yet return the same or similar result), it’s nearly impossible to know what prompt (or tool) we should optimize for - and that’s after we’ve decided what it is we’re trying to promote or achieve.
Herein lies the source of the SEO vs AEO/GEO debate. Where in the funnel should comms look to operate?
SEOs have long operated with intentionality about funnel behaviors, the result of years of click and keyword optimization. Gen AI tools are no different. Asking AI any sort of “what” or “which” question will result in a zero-click answer which is as close as you’ll get to the kind of low-funnel, high purchase intent behavior of a normal search engine (that would have been the last click before site visit on Google search).
Comms (and the platforms/media environment we call home) has always played best in the upper ends of the funnel - the areas that raise awareness and consciousness, the triggers of the “what or which” request. Sew the seed, spread the word, pique the interest, generate the recall. I just don’t see how that’s different now than two or three years ago. Sure, SEO can do (and does) that, too, but it feels like using a surgeon’s scalpel to carve a brisket - way more precision and accuracy than you need.
I just don’t see Comms being skilled or experienced enough to ever be able to compete for optimized last/zero click behavior better than an entire SEO industry. I do, however, believe that the work Parry and James have begun is helping us (well, me at least) clarify what ideas we can drop or where we can play effectively. Sometimes knowing something is not for you is as important as knowing what is.
Mirroring SEOs structured approach for upper funnel intent and content can only be a good thing. For example, if we think of a product launch, have we really stopped and asked what we are trying to achieve with the announcement? Raise awareness? Inform? Explain? Inviting trial? Sales? Each of these outcomes has an extensive number of short and long-tail search terms associated with them that we rarely, if ever, contemplate using - but if we’re thinking about AI Optimization, we need to include the moment we begin thinking about completing Parry’s template or a brief.
Which is where this brilliant tool called Listening Mind from Seyong Park plays such a crucial role in empowering us with user intent insights that can be applied directly to content creation. Rand Fishkin has a brief demo here. SEO tools that empower comms people. Win-win.
But as a genius friend of mine said:
Write for users
Optimize for Search Engines
Structure for AI
Any work that can improve the way we think about adapting to these new tools and modernize our industry have to be thoroughly digested and applauded. Until we try, we won’t know.
No news this week because, well, there’s already plenty above!
Please, please, please, do consider sharing this piece. It’s a labor of love, but is so helpful to me and many other communications professionals looking for work.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.