Ever read a paragraph that feels perfectly polished yet strangely hollow? You're likely sensing an 'AI tell.' As Large Language Models (LLMs) flood our screens, a new game of cat-and-mouse has emerged. We aren't just looking for hallucinations anymore; we're looking for the linguistic fingerprints that scream 'bot.'
The Vocabulary of the Machine
AI tells often start with specific words. For a while, 'delve' was the ultimate red flag, but as models evolve, so do the patterns. Recent research suggests that newer models have their own favorites; for instance, some versions of Claude Opus have been noted for an unusual fondness for the word 'dependable' or the phrase 'this matters.' It's not just single words, but a general reliance on corporate buzzwords and a 'tidy' structure that feels too symmetrical to be human.
Patterns Over Words
Beyond the vocabulary, there are structural tells. AI tends to favor predictable rhythms and 'negative parallelism'—a specific way of balancing sentences that feels robotic. From the overuse of em dashes to a rigid adherence to five-paragraph essay formats, these patterns create a 'slop' effect. Databases like Sloptells are now tracking these shifts in real-time, noting how tells die out in one model only to reappear in another.
Whether you're a student, a journalist, or a casual reader, the ability to spot these tells is becoming a vital digital literacy skill. As AI continues to mimic human nuance, the gap between 'perfect' writing and 'authentic' writing may be the only thing left to guide us.
Sources
Media
