Kagi Translate's AI answers the question "What would horny Margaret Thatcher say?"
Users discovered they can type any style into Kagi's search bar, from 'LinkedIn Speak' to 'tiny little kitten'.
Kagi, known as a paid competitor to Google Search, launched its AI-powered Kagi Translate tool in 2024. While it officially supports 244 languages, users recently discovered they can bypass the dropdown menu and type any desired 'language' or style—such as 'rude man with a Boston accent,' 'Reddit Speak,' or 'horny Margaret Thatcher'—directly into the search bar. The tool's underlying large language models (LLMs) then attempt to synthesize text in that specific voice, leading to a wave of creative and humorous experiments across social media and Hacker News.
The viral moment represents a throwback to the early, playful days of ChatGPT, where marveling at an AI's ability to mimic distinct styles was a novelty. Kagi's social media team has leaned into the trend, even promoting its 'LinkedIn Speak' translation to help users 'fit right into that crowd.' However, the phenomenon also exposes the inherent risks of giving users open-ended access to powerful, generalized LLMs without strict guardrails, as the same capability that generates silly cat speak can be used for media criticism, political jokes, or other potentially problematic outputs.
- Kagi Translate uses a combination of LLMs and allows users to type any output 'language' (e.g., 'Gen Z slang,' 'McKinsey consultant') into its search bar.
- The feature was quietly discovered over a year ago on Hacker News but went viral in March 2026 for styles like 'LinkedIn Speak' and famous personas.
- The trend highlights the creative, pattern-synthesizing power of LLMs but also underscores the risks of unconstrained user access to generalized AI models.
Why It Matters
It demonstrates both the creative potential and the inherent safety challenges of deploying powerful, open-ended LLMs to the public.