“AI-powered” has become one of the most overused phrases in software marketing — slapped on everything from a basic if-then form to genuinely sophisticated machine learning systems. Since the label alone tells you almost nothing anymore, it helps to know what to actually look for.
Real signal #1: does it get better with more data or use? Genuine machine learning systems improve their outputs as they’re exposed to more examples — that’s the entire point of “learning.” If a product’s behavior is identical on day one and day one thousand regardless of usage, it’s very likely running fixed rules dressed up as AI, not an actual model.
Real signal #2: does it handle inputs it wasn’t explicitly programmed for? A rule-based system (“if the user types X, respond with Y”) breaks or gives nonsense answers the moment an input falls outside its rule list. A real language or vision model generalizes — it produces a reasonable response even to phrasing or images it’s never seen exactly before. Try feeding a tool something slightly unusual and see if it degrades gracefully or breaks completely.
Real signal #3: does the company mention training data, model architecture, or evaluation results anywhere? This doesn’t need to be deeply technical, but legitimate AI products usually say something concrete — “built on GPT-4,” “trained on X dataset,” or at minimum describe what kind of model they use. Marketing copy that only ever says “powered by advanced AI” with zero specifics, across the entire site, is a real yellow flag.
Common “AI-washing” patterns worth recognizing:
- A basic keyword-matching chatbot marketed as an “AI assistant”
- A spreadsheet formula or simple statistical average rebranded as an “AI prediction engine”
- Standard photo filters (which have existed for decades) marketed as “AI enhancement”
- A basic recommendation list sorted by popularity, labeled an “AI-curated” feed
Why this matters beyond just curiosity: if you’re paying a premium for “AI” features, or trusting an “AI” tool with a decision that matters, it’s worth knowing whether there’s real adaptive intelligence behind it or just a rebranded set of fixed rules. The rules-based version isn’t necessarily bad — simple, predictable tools have real value — but you should know which one you’re actually using, and pricing that assumes cutting-edge AI shouldn’t apply to something that’s really a lookup table.
