It's 23:00, you need a laptop, and decision fatigue has fully taken over. You type “best laptop under €1000” into ChatGPT because it's right there and thinking is hard. Enter. Back comes a tidy list of five laptops, mildly pleased with itself. None of them know you exist. None of them know you need something light enough for a backpack, fast enough for video calls, and cheap enough that you won't cry if your kid drops it. You asked a very capable assistant the same flat question you'd type into a search bar, and got back exactly that: an answer to nobody in particular.
ChatGPT didn't fail you here. You used it like Google. Google is a librarian: hand it three keywords and it silently slides ten links across the counter, then leaves you to do the reading. ChatGPT is supposed to be more like a friend who actually knows you, the one who asks “wait, is this for work or the gym?” before answering. Skip that part, mutter a few keywords at it instead, and it has nothing to reason with. So it guesses, politely, and calls it an answer.
Old habits, new tool
This isn't a new failure mode. It's twenty-year-old muscle memory. Google spent two decades training you to think in fragments: type three words, skim ten blue links, move on. That reflex didn't retire when ChatGPT showed up. If anything, it's making a comeback. Semrush recently tracked ChatGPT prompts over time and found the share written like a Google search nearly doubled in four months, from under one in five to more than one in three, as the tool reaches a wider, more casual crowd. Old habits aren't fading. They're just moving house.
Nielsen Norman Group watched this play out with real people in a research lab. One participant, shopping for a football goal for his son, used Google and got his answer in ten minutes flat. Later in the same session, researchers pointed him at Gemini for a plumbing problem he'd normally have Googled. He fumbled the first attempt, then got real, specific help once he actually described the problem instead of typing a search term. Afterward, he admitted he probably should have started with the chatbot in the first place.
Doc Digital SEM's research on query behavior found that a Google query still averages two to four words, while a ChatGPT prompt, once someone's using it right, tends to run ten to thirty. Same person, same need, a completely different amount of information handed over, and a completely different quality of answer coming back.
What actually fixes it
You don't need a framework for this. You need three things a Google search never asks for. What you actually want to do with the answer, not just the topic. What matters most to you, since “best” means something different for everyone. And any constraint that would otherwise rule out a perfectly good answer. That's it. Not a longer prompt for the sake of length, just enough for the model to reason with instead of guess.
As Ethan Mollick, the Wharton professor who studies how people actually use AI, puts it: most people still treat AI like Google, asking it factual questions, but AI isn't Google, it won't hand back the same reliable answer twice. His fix isn't a smarter intern, it's treating AI like a new coworker: patient, capable, and just as lost without the context you'd give any person new to the job.
The real question was never which AI is best. It's whether you're giving it enough to actually reason with, and that's a more interesting question to sit with. It starts with something as simple as context, the first letter CORDON asks for and weighs on your behalf.
CORDON exists for exactly this gap: a few short questions about your situation, your goal, and what would rule an answer out, turned into one prompt that actually gives the AI something to reason with. Free, no account needed.