LLMs, media plans and creators: are brands walking into a trap?

Justin Lebbon talks to Shane Skillen, CEO of Hotspex, an AI specialist working with Fortune 100 brands, about what AI is really changing in branding, media and the creator economy — and what it isn't. LLMs as the new "zero moment of truth": answer engines scan everything known about a brand to make personalised recommendations. Hotspex research found 71% of consumers overrode an LLM's recommendation when they had a strong emotional connection to a brand — weaker brands are left "at the mercy" of the model. The platform flywheel: Skillen argues Google and Meta benefit from weak brands, citing Meta's move to pull product catalogs onto-platform to capture sales and boost ROAS, and Meta's AI assistant Muse as a shopping gateway. AI bias in media planning: 63% of media buyers now use AI to plan media (Futuri study). Unprompted, a test plan pushed ~57% of a $10m QSR budget into Google, Meta and Amazon; adding IPA econometric and MMM instructions dropped that to 20% and lifted TV and traditional to 38%. Gemini recommended just 6% TV while heavily favouring YouTube. Why the skew exists: roughly 100x more marketing-effectiveness data exists for digital in training sets than for TV, radio and out-of-home — and closed models can be tuned to favour owner properties. Human in the loop: Skillen pushes back on the "no human will touch a media plan" prediction, warning of model bias and "AI psychosis." Creators vs TV: Unilever's shift toward creators (from 30,000 to 300,000) raises consistency risks — if creators give different "reasons to believe," LLMs training on that content get confused. Justin counters that creator environments are low-attention and platform-dependent. The consistency imperative: brands need "singularity" across touchpoints so AI assistants form a coherent view at the moment of purchase. Key takeaways Hotspex research found 71% of consumers overrode an LLM's recommendation when they had a strong emotional brand connection — strong brands matter more, not less, in an agentic shopping world. 63% of media buyers now use AI to plan media (Futuri), and left unprompted those tools default heavily to digital — about 57% of a test budget into Google, Meta and Amazon. Adding IPA econometric models and MMM data to the prompt shifted the mix to 20% digital and 38% TV/traditional; Gemini allocated just 6% to TV while favouring YouTube. The skew exists because training data holds roughly 100x more effectiveness evidence for digital than for TV, radio and out-of-home. Skillen argues Google and Meta benefit commercially from weaker brands and will push to go brand-direct via their own AI. Creators are growing (~28% globally) but risk brand fragmentation; without consistent 'reasons to believe', AI assistants won't confidently recommend a product.