Remember the ‘caveman prompt’? It was popular a few years ago. You'd type "talk like a caveman and keep it short," and suddenly ChatGPT would stop adding fluff to every response. It actually gave you an answer, and the prompt even kept it from responding with a follow up question.
That was the whole point back when models weren’t smart enough to trim the response alone. You'd ask one question that deserved a sentence or two in response, but you'd get paragraphs instead.
When ChatGPT first launched, every response came with the same generic, puffed-up polish. The caveman hack cut through it, which is why many of us used it daily. Now, three years later, I barely use it, because the problem it solved no longer exists.
The newer models actually understand plain English. They don't need to be tricked into being useful. So I wondered if the old caveman trick still deserved a spot in my toolkit.
I ran three tests. Here's what happened.
Test 1: Explaining something complicated
I asked ChatGPT: Why do airplanes stay in the air?
The first answer read like a textbook. Accurate, thorough, and about twice as long as it needed to be.
Then I asked for the caveman version when meant I prompted: Talk like a caveman and keep it short.
The answer: Air push wing. Wing go up. Plane stay sky.
Not bad, honestly. Six words and you get the gist. No one's confused about what's happening. But then I tried something different:
You're explaining this to a curious 12-year-old. Use everyday language, include one simple analogy, and keep it under 120 words.
That answer was a lot better.
It made sense. I could picture it. It compared a wing to holding your hand out of a car window, angled up, the air pushes it higher. It was actually interesting to read.
"Air push wing" gets the idea across, mostly. But it doesn't explain anything. A good answer does both, stays short and makes you understand.
Test 2: Writing an email
Next, I asked ChatGPT to write a refund request.
The caveman version was blunt. Too blunt. Something like "Want money back. Product bad. Give refund." I wouldn't send that to anyone I hoped to get a response from; the goal wasn't supposed sound like an actual caveman.
So I tried this: You're a customer service expert. Write a refund request that's under 150 words, polite but firm.
I barely had to tweak the email becuase it came back with the right tone — respectful but direct, with the order details laid out clearly. It sounded like something a real person would write.
That's the thing about newer ChatGPT. It doesn't just follow word limits. Give it a little context — a role, a tone, a situation — and it actually gets what you're going for. In 2023, you had to wrestle the model into shape. Now you just have to tell it what shape you want.
Test 3: Summarizing a long article
Here's where the old trick still works. I pasted a long article into ChatGPT and asked for a caveman-style summary. The result was rough and not something you'd put in a presentation. But, it got the job done because it felt like useful bulletpoints.
It stripped everything down to the bones, meaning there wasn't an intro or transitions, no unnecessary filler. It's now my go-to prompt when I'm staring at a 4,000-word report with a minute to spare. Or even news that I need to understand before a meeting with a tech giant.
What works better now
The real change isn't the prompt, it's how ChatGPT has evolved. In 2023, prompt engineering was about working around the model's quirks. You weren't really communicating with it, you were poking it in the right spots to get a decent output. Now, straightforward instructions just work.
The prompts I keep coming back to have a few things in common.
- Give ChatGPT a role. Instead of asking it to explain investing, tell it: You're a financial advisor explaining investing to someone opening their first retirement account. That one line changes the answer completely. You go from a Wikipedia overview to something that actually sounds like advice. The model picks different examples, different vocabulary, a different level of detail. It's the single easiest way to improve any prompt.
- Tell it who it's talking to. The caveman prompt was basically a shortcut for "be brief." It worked, but it was a blunt instrument. Today I'd rather say: Keep it under 150 words, avoid jargon, remove analogies, give me 3 bullet points.
- Specify the format. Want a checklist? Ask for one. Need a table? Say that. Step-by-step instructions? Tell it how you want the information laid out. This sounds obvious, but I spent months getting frustrated with wall-of-text answers before I realized I just never told ChatGPT what I actually wanted to look at.
- Keep the conversation going. I used to cram everything into one massive prompt, thinking I only had one shot. Now I ask a simple question first. Then I follow up.
The verdict
The caveman trick still has its place. When I need a lightning-fast summary of a dense document, it works. I used it last week on a 30-page policy brief and had the key points in about ten seconds.
For everything else, I'm leaving it in 2023. The best results now come from giving ChatGPT enough context to work with. Tell it who to be, who it's talking to, and what kind of answer you want.
Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds. Subscribe to Tom's Guide on YouTube and follow us on TikTok. Finally, you can visit our dedicated Tom's Guide Savings Squad hub for expert help on getting the best products for less.
More from Tom's Guide
