Five Phrases That Make ChatGPT Think Instead of Guess

Ask ChatGPT "make a promotion plan for a beauty salon", and you get a list that would suit any salon in the world: social media, discounts, reviews. The model didn't make a mistake. It honestly answered a question that is missing half of its data, and filled the gaps with the most averaged option.

I build content systems for experts, and most breakdowns at the start come down to one thing: the prompt isn't weak, it just has no places where the model is obliged to stop and ask. Here are five phrases I add to a prompt when I need an answer for a specific task, not for an imaginary average person.

"First ask me three questions"

With this phrase, ChatGPT first asks about the client's budget in its answer instead of making it up itself.

The difference is noticeable: without this phrase, the model gives the plan from the paragraph above, with social media and discounts. With the phrase, it first asks what the salon's budget is and where clients come from now, and only then builds a plan around real numbers.

"Prove me wrong"

I add this phrase when I already have my own idea and want not praise but a stress test. With this wording the answer stops being agreement and lists where the thesis may not work in practice.

In my practice this works better than the direct question "what are the downsides of this": the wording "prove me wrong" gives the model a concrete task to argue, instead of politely listing risks separated by commas.

"Show three options: safe, medium, and bold"

Without this phrase, ChatGPT almost always gives one averaged answer that neither takes a risk nor impresses. The phrase makes it lay the solution out by level of risk and show the spread, while the choice stays with you.

It's handy for a product name, and in general for any task where one right answer doesn't exist and you need a comparison, not a ready solution.

"Explain it as if I'm ten"

This isn't about going easy on yourself. It's a check on whether the answer gives a truly simple explanation, not complicated words in place of substance. If after such a request the explanation didn't get simpler, the topic has been explained formally, through terms, not through the point. In that case it's worth asking again or giving the model more context on the topic.

I apply this phrase to any new technology or term before taking it into content for an audience. If I can't explain it in simple words myself, the audience definitely won't understand it on the first pass.

"What's missing to answer accurately"

A direct question to the model about gaps in its own data. Instead of guessing for you, it names the specific missing fact, for example, what the result was last time.

This phrase is especially useful before a big task: for example, a quarterly plan, where one wrong input at the start spoils the whole result at the end.

How to turn this into a habit

You don't have to memorize all five phrases. It's enough to take one habit: before sending a long prompt, ask yourself whether it's too general, and if so, add one of the phrases above. Over a few weeks it becomes a reflex, like the habit of rereading a message before sending it.

I break down similar working techniques and specific combinations of AI tools in the channel «Растём с ИИ» ("Growing with AI"). A related piece: how to choose an AI tool for a specific task. If you need to build a whole prompt system for your practice, you can start with a free consultation.