There’s a new ‘share of’ in town. ‘Share of model’. A new marketing metric for the GenAI age. More in the article below. Includes a bit of history on the ‘share ofs’ it follows on from - share of market, share of voice, ESOV and share of search. Jellyfish The Brandtech Group Marketing Week https://lnkd.in/edzVVd_q
This is a great idea, but I'm wary of producing a number from LLMs. Test prompt: List all the brands in the car industry. You'll get a list. Ask AI to rewrite. You will get a slightly different list. Which one should be in the number?
Great bit of thinking - I believe the (slow) speed LLMs are built/refreshed means that such a way of measurement would lead to out of date findings? Unless we see a future of OPEN AI doing weekly updates. Kind of my issue with using AI for research, hard to understand how it got to that outcome/figure and you've no idea of the time periods. But cheap and fast so has it's uses!
Self-confessed Luddite who is far from expert in anything AI but this one will take some convincing - For starters, how will ‘share of model’ metrics differentiate the ‘organic’ references to brand from those generated by the many professionals using LLMs to support their work…? And how will the likes of Search and Web Traffic rankings skew what LLMs pull and feature? Appreciate early days but this has thrown up so many questions in my little mind…
thanks for writing and posting! I though 'share of model' (also) meant share of the attribution model for publishers, i.e. to what extent does the attribution model recognizes your contribution Gabriel Mohanna
Really love the idea, Tom. As it stands, there's no clear, reliable way to actually measure what's being proposed with the tools at hand. But I look forward to someone taking up the torch to make this a practical reality!
I have nothing to say at this stage other than “this is excellent” 👏🏻
This really shows how little marketers understand LLMs, starting from the most obvious fact which is that most brand mentions happen in spaces like Social Media, YouTube, Media outlets, etc., all of which are currently NOT part of a model’s pretraining data (And they also happen in regular conversations which are ofc not part of that data either). If you don’t understand the basics forget about the rest.
Hi Tom. This bit is wrong: The key insight in all this work was that brands investing more in media than their market share implied were much more likely to grow their market share over the longer term.
Tom Roach Fascinating. If I'm reading this correctly, it seems like this could be a solution for measuring the effectiveness of marcom activities across owned, earned, paid and shared. And also build in KPIs around brand trust. Am I getting that right? The trouble with ESOV is it leaves out owned, earned and organic social content. And the trouble with measuring share of search/mental availability is, it leaves out if people know about the brand for the right reasons (because they like it and want to buy it). This new model seems like it has the potential of being a magic bullet that PR and owned content folks have been desperately waiting for.
Head of B2B Marketing at Vega IT
3wTom, just recently a B2B inbound lead told us that we’ve been recommended to them by the ChatGPT 🤖 That’s when I realized that, sooner or later, everyone’s going to opt in for a “Model Optimisation Strategy”, in order to increase what you call a brand’s “Share of Model”. But this would have a different use than the Share of Search, which is used as a proxy metric for assesing the existing mental availability of a brand. It seams that the “Share of Model” would indicate how likely it is for an audience to be exposed to the brand while using an LLM. So, it seams to be similar to SEO ranking — in a sense that it enables discoverability. But the “Share of Model” would at the same time count as earned media, which we can’t say about a brand’s own SEO ranking. And, better than just being discovered, “Share of Model” would also serve as a source of social proof — and even as an effective recommendation engine (which we experienced firsthand at Vega IT). I don’t know if these☝️random thoughts make sense, but I’d love to hear your opinion 🤔