February 2026
Most word games use a fixed list of pre-selected words that rotate on a schedule. Everyone plays the same puzzle on the same day. FivLet takes a completely different approach — every single game is generated fresh by artificial intelligence, meaning no two players ever see the exact same puzzle. Here's how it works.
FivLet is built on Google's Gemini AI models, accessed through Firebase Genkit — Google's framework for building AI-powered applications. When you start a new game, FivLet sends a request to the Gemini model asking it to select an appropriate 5-letter word based on your chosen difficulty level. The model draws on its vast training data to pick a word that fits the criteria — common enough to be fair in Standard mode, or genuinely challenging in Difficult mode.
Once a word is selected, a second AI request is made specifically to generate a clue for that word. This is where FivLet gets interesting. The clue isn't a simple dictionary definition — the AI crafts a hint that gives you useful context without giving the answer away too easily. It might describe what the word relates to, how it's used in everyday language, or give a creative contextual hint. The clue is generated specifically for the word chosen, making each puzzle feel cohesive and intentional.
FivLet offers two difficulty modes that influence what the AI selects. In Standard mode, the AI is prompted to choose words from everyday vocabulary — words most people with a reasonable English vocabulary would know. In Difficult mode, the prompt shifts to request more obscure, complex, or less commonly used words. This gives experienced players a genuine challenge while keeping the game accessible for casual players in Standard mode.
The AI-generated approach means FivLet never runs out of puzzles. There's no finite word list, no repeated puzzles, and no spoilers from other players who finished the daily puzzle before you. Every game is yours alone. It also means the game can evolve — as AI models improve, the quality and creativity of words and clues improves alongside them.
One challenge with AI-generated content is consistency. Occasionally an AI might generate a clue that's too vague, or select a word that's obscure even for Difficult mode. FivLet includes validation logic to catch these edge cases — if a clue or word doesn't meet quality standards, the game will flag it and prompt you to start a new game rather than serving you a bad puzzle. We continuously refine the prompts and validation to minimize these occurrences.
AI-powered word puzzles represent the future of word gaming — infinite variety, personalized difficulty, and puzzles that feel fresh every time. FivLet is proud to be at the forefront of that evolution.