Large language models come factory-set to be polite, neutral and sanitised. Technically, this works flawlessly – but in practice, it leads to a noticeable lack of distinctiveness. 📉
When a B2B company stands for a clear-cut approach, pragmatic advice on an equal footing or a high level of technical precision, this identity often collapses entirely at the search bar: the AI simply doesn’t speak the brand’s language. ❌
The solution? It certainly doesn’t lie in a simple addition such as ‘Be relaxed and competent’ in the system prompt.
A genuine brand personality is a comprehensive architectural package:
🎭 Persona & behavioural models: Fixed role profiles, no-gos and concrete ‘few-shot’ examples instead of vague adjectives.
📚 Corporate language in the RAG: The structure and metadata of the vector database determine whether the response sounds like it comes from a senior consultant or an advertising brochure.
🛡️ Safety first, style second: When the brand persona applies – and in which areas (legal, compliance, sensitive complaints) a deterministic, factual fallback must be applied without fail.
🧪 Evaluation & UX: How brand compliance and tone can be measured automatically before the first user even logs into the portal.
Our colleague Michael Kettel has summarised exactly what this blueprint looks like in practice for enterprise portals (e.g. based on TYPO3) in a brand-new white paper. 💡
📄 Corporate Prompt Engineering
How to give an enterprise AI search a bespoke brand personality (5 pages, practical & to the point)