Stop using AI like Google

A lot of people try AI once, decide it is overrated, and go back to how they worked before. Watch what they actually did and the reason is usually the same. They typed a short question, the way they would into a search engine, got a generic answer back, and concluded the tool was not all that. The verdict was fair on the evidence. The evidence was just from using a powerful tool in completely the wrong way.

For twenty years we have all been trained by search engines to do one thing: type a few keywords, scan a list of links, click one, find the answer ourselves. That habit is deep, and it is exactly the wrong instinct for AI. Conor Grennan, who teaches people to work with these tools, coined a phrase for this, the search-engine mindset, and it is the single most common reason capable users get disappointing results. Treat the model like Google and you waste almost everything it can do.

The Google reflex

The search reflex has three moves baked in. Keep it short, because search engines reward brevity. Expect one shot, because you type once and get a page of results. And do the real work yourself, because the engine only points you at sources. All three are sensible for search. All three throw away what makes AI useful.

When you type a five-word question into a model and judge it on the one-paragraph answer, you are using a tool built for extended, contextual, back-and-forth work as if it were a lookup service. Grennan's point, set out in his WorkLab conversation on moving past the search-engine mindset, is that using AI merely as a search engine wastes most of what it can do. The limit is not in the tool. It is in the habit you brought to it.

Why one-shot prompts disappoint

A search engine is meant to give you the same good answer to a short query every time. AI is different in a way that matters: the quality of what you get back depends heavily on how much you put in. A short, contextless prompt gives the model almost nothing to work with, so it produces the most generic plausible answer, which is exactly the bland output that makes people shrug. Feed it the situation, the constraints, an example of what good looks like, and the same model produces something specific and useful.

So the one-shot prompt is not a fair test. It is the AI equivalent of judging a new colleague by the quality of their work after a ten-second briefing shouted across the office. Nat Eliason, who writes about getting real value from these tools, is blunt that most people who dismiss AI simply "lack the skills and imagination to use the tools to their fullest." That is not a put-down. It is a description of a gap that closes quickly once you change the habit.

Conversational, iterative use

The shift is from asking to working. Instead of typing a question and grading the answer, you start a conversation and steer it. You give the model the context, see what it produces, tell it what is wrong, and push it toward what you actually need. The value is in the loop, not the first reply.

This is closer to working with a fast, capable assistant than to looking something up. You would never expect a good answer from a colleague after a five-word brief, and you would not accept their first draft as final without comment. You would give them the background, react to what they produced, and refine it together. That is exactly the posture that gets good work from AI, and it is the opposite of the search-and-scan reflex.

The shift also changes what counts as a good question. In search, a good query is short and precise. In a conversation with AI, a good opening is generous: you say what you are trying to achieve, why, for whom, and what would make the answer useful to you. That feels like over-explaining, and it is the single biggest difference between people who get bland results and people who get sharp ones. The model cannot read the situation in your head. The more of it you put on the page, the better the work that comes back.

A search engine answers your question. AI does the work with you, if you let it into the loop.

Three habits to break the reflex

The reflex is strong, so it helps to replace it with three deliberate habits.

Give it the work, not a keyword. Instead of a short question, describe the actual task: what you are trying to produce, for whom, with what constraints. Paste in the relevant material. The longer, richer brief feels like more effort, and it is the difference between generic and genuinely useful.

Expect to iterate. Treat the first answer as a draft to react to, never the finished product. Tell the model what missed, what to change, what to keep. Two or three rounds of steering will get you somewhere a single prompt never could.

Show it what good looks like. If you have an example of the kind of output you want, a past report, an email in your voice, a format you like, give it to the model. AI is very good at matching a pattern it can see and poor at guessing one it cannot.

The honest limits

Breaking the search habit does not mean swinging to blind trust. The model can still be confidently wrong, and a fluent, well-shaped answer is not the same as a correct one. So the new habit comes with a standing rule: still verify anything that matters, especially facts, figures and names. The tool is neither a search engine nor an oracle. It is a fast, fallible collaborator whose work you check.

There are also tasks where the old reflex is genuinely fine. For a quick, factual lookup you can verify in a second, a short query is the right tool, and a long conversation is overkill. The point is not to turn every question into a dialogue. It is to stop judging AI by its performance on the one job it is worst suited to, and to use the conversational mode for the work where it shines.

What to do this week

Take a task you would normally do yourself and run it as a conversation instead of a search. Brief the model properly, react to what it gives you, push back two or three times, and show it an example of what you want. Notice how different the result is from the one-line-prompt version you tried before. That single change in habit is the thing that turns AI from underwhelming to genuinely useful, and it costs nothing but a few extra sentences.

Helping people and teams build that working habit, rather than abandoning the tools after a disappointing first go, is much of what the AI Lessons for Leaders sessions do. If AI has underwhelmed you so far, book a personal lesson and we will change how you work with it.

Sources and further reading

• Conor Grennan on moving past the search-engine mindset, Microsoft WorkLab, 2024. Where the "search-engine mindset" idea comes from, and why it leaves people underwhelmed.
• Nat Eliason, "The Window of Opportunity is Here", 17 April 2024. His point that most people who write AI off simply have not learned to use it well yet.