Australian startup Springboards launched an AI model called Flint built specifically to break the sameness problem plaguing generative marketing tools: ask most large language models for a tagline and they converge on the same handful of predictable answers. On the independent Novelty Bench, Flint scored 7 out of 10, generating seven functionally distinct responses across ten prompts, against an average of 2.88 for leading LLMs. “Frontier models were getting smarter while outputs grew eerily similar. We built Flint, the model we needed ourselves,” said Pip Bingemann, Springboards’ co-founder and CEO.
The proof point is already in market. L’Oreal used Flint to brainstorm a new campaign aimed at male shoppers for its CeraVe skincare brand in Australia and New Zealand, a region where CeraVe had never before tailored creative to men. Springboards brought local market insight into the sessions alongside the model itself, and the team used Flint’s outputs as a starting point for refinement rather than finished creative, generating options like “Pave your own path” and “Slay the day, one step at a time” before the human team narrowed and edited. That distinction matters: Flint was deployed as a brainstorming partner with a human still making the final call, not as a replacement for the creative department.
The original insight is that “novelty” is becoming a measurable, marketed product attribute in generative marketing tools, the same way “accuracy” became one for early AI copywriting products. As more brands run the same handful of frontier LLMs through similar prompts, outputs are converging, and a startup building a benchmark specifically for divergence signals the market now treats creative sameness as a solvable, sellable problem. Marketers weighing where AI belongs in campaign planning versus creative production should note the split Springboards’ own customer demonstrates: strategy can lean on automation, but distinct brand voice needs a model built to resist consensus. It is also a reminder that AI adoption alone was never the differentiator; which model, and how it is used, is.
Source: Springboards
