L'Oréal has signed a foundational partnership with OpenAI that allows the cosmetics group to provide up-to-date product information directly to OpenAI to be incorporated into the models that power ChatGPT. As a result, when users ask about L'Oréal brands, the assistant will draw on the company’s own notes and databases rather than relying solely on the mix of reviews, Reddit posts, and wiki-like pages that typically inform foundational model training.
OpenAI has stated that the arrangement will not tilt results in L'Oréal’s favor. The deal also brings OpenAI’s image model into L'Oréal’s CreAItech system, and places Maybelline’s virtual try-on functionality inside ChatGPT.
Practical implications
In recent years, a variety of industry tactics have emerged to influence what generative models return: generative engine optimization (GEO), prompt-stuffing, seeding Reddit threads, and reverse-engineering which sources the model ingests. Those approaches are workarounds that attempt to shape model outputs via external channels.
By contrast, L'Oréal’s approach hands the data directly to OpenAI, bypassing the need for such external manipulation. Integrating Maybelline’s try-on into ChatGPT illustrates how brands can embed their experiences and assets directly into the assistant rather than attempting to game the model’s input sources.
Key takeaway
The partnership highlights that many GEO-type techniques are symptomatic of not having a direct relationship with model providers. For brands, securing a direct feed or formal integration with the model can be a more straightforward strategy than using external optimization tactics. L'Oréal’s deal with OpenAI exemplifies how such direct data provisioning can change the dynamic between brands and large language models.
