8 ChatGPT tricks most people still aren’t using

Most people end up using ChatGPT like a slightly smarter search engine. That used to be me when LLMs kicked off a few years back. I’d ask a question, get an answer, maybe ask a follow-up, and continue on from there. There’s nothing wrong with that, and it works well enough for simple lookups even today. However, it barely scratches the surface of what the tool is actually capable of. The fact of the matter is, that the gap between a casual user and an expert user who genuinely gets things done with ChatGPT isn’t technical know-how. Rather, it’s a skill issue. It comes down to a handful of core habits that completely transform how you interact with the model and allow you to extract the absolute maximum from its extensive capabilities.

I’ve spent a lot of time pushing ChatGPT well past the basics, across grammar fixes, research, technical research, and day-to-day decision-making guided by data. While you will find plenty of online sources offering hyper-specific prompts, these eight specific tricks are the ones that have made the single biggest impact in how I approach ChatGPT and other large language models. Looking for Gemini tricks? We’ve got that covered, too. But these little tricks are the ones that have made all the difference in getting me better results from ChatGPT.

AI hallucinates and makes assumptions. Consequently, it is surprising to me that almost nobody takes advantage of this simple trick, even though it dramatically improves the quality of the responses you receive.

The default behavior of any AI model is to make assumptions. You ask for a marketing email, and it writes one. It picks a tone, a length, a structure, and a target audience, all without asking you a single thing. Most of the time, the resulting copy is perfectly fine but extremely generic, precisely because the system had to guess every variable it lacked. The more open-ended the task, the worse this gets.

The fix is simple. Before giving ChatGPT a task, tell it to ask you questions first. Something like: “I need help writing a cold outreach email. Before you write anything, ask me whatever you need to know to do this well.”

That’s all it takes. What follows this prompt is usually a brief, targeted questionnaire covering your target audience, core goals, tone preferences, and any specific context that might alter the final output. Once it has that, what it produces is almost always much more precise and closer to what you actually wanted.

Where this really pays off is in creative and strategic work where there is usually no room for assumptions. When you are briefing ChatGPT to assist with a brand campaign, a product launch plan, or even a highly personal piece of content like a letter or a speech, the assumptions it makes regarding tone, audience, and underlying intent matter enormously.

A prompt that starts with “ask me what you need to know” instead of “write me a speech” forces the model to surface those variables before they become problems buried in a draft that you then have to rework. I would argue that this is an incredibly useful exercise in its own right because it forces you to clearly articulate your actual goals, which frequently turns out to be half the battle in the first place.

If you use ChatGPT for work, you’ve almost certainly repeated yourself across conversations. Your company name, your role, recurring project details, and technical specs you always need to reference. These repetitive, manual steps quickly add up to a significant amount of lost time, yet the problem is entirely avoidable.

A far better approach is to kick off any working session by dropping a comprehensive block of foundational information right at the start. Think of it as briefing a new assistant before handing them a task. A network engineer, or self-hosting enthusiast, might paste in a list of IP addresses, server names, and environment details. A content creator might share a brand voice guide, a target audience description, and a set of topics that are off-limits. A product manager might drop in a project brief, a list of stakeholders, and key constraints. You catch my drift.

Once that’s in the conversation, ChatGPT will reference it consistently without you having to repeat it. You can even formalize this with a Custom GPT and a persistent system prompt, but even within a single session, this habit alone meaningfully reduces the back-and-forth. The more contextual details it possesses right from the baseline, the less it has to guess, which ensures the output remains highly coherent across an extended conversation. Think of it less like prompting and more like onboarding.

If your daily ChatGPT workflows constantly revolve around the same core topics, you should navigate to the Settings menu, select Memory, and add these details to the “More About You” section so that the AI can seamlessly leverage that background knowledge across every future chat.

We’ve all used ChatGPT to reframe a piece of text for a social media post or LinkedIn. “Write me a LinkedIn post about this interesting news I just read” is a perfectly valid prompt. Unfortunately, it is also a surefire recipe for generating generic results that read exactly like what everyone else is posting.

The reason is that without a reference point, ChatGPT defaults to whatever the statistical average of LinkedIn posts looks like. That usually means a generic hook that starts with a question, three punchy lines in the middle, a call to action at the end, and a handful of relevant hashtags. It’s not wrong, but it’s thoroughly unremarkable. If you have spent any real amount of time scrolling through LinkedIn, this formulaic writing style becomes instantly unmistakable, signaling right away that it was generated by an AI.

Let me be straightforward here. I’d still recommend writing the post yourself, but if you’re pressed for time and need a little helping hand, the better move is to show it an example before you ask. Paste in a post you wrote previously that you were happy with, or one you admire from someone else, and say, “Write something in this style.” You can be even more specific by saying something like, “Match the sentence length, the level of formality, and the way this builds to a point.” Large language models like ChatGPT are the best at pattern-matching tone, style, and syntax when they are provided with a concrete structural baseline to work from.

This applies to almost every content task. Writing samples, code snippets, report formats, slide structures, even social posts. If you can show it one example of what something “good” looks like according to you, the output quality jumps considerably.

Interestingly, it also works in reverse. If you paste in an example of what you do not want and tell it why, that negative reference is just as instructive. Instructing ChatGPT to avoid anything that sounds like a specific negative sample is often the absolute fastest way to establish a rigid style boundary that would otherwise take multiple paragraphs to explain.

Just keep in mind that within the ongoing context of that specific chat, ChatGPT will consistently replicate that exact style, which can accidentally introduce the very same stylistic ruts you are attempting to avoid. Consider switching it up from time to time.

Most people who have tried ChatGPT’s Voice Mode used it once and thought, “Okay, it’s like Google Assistant but better,” and went back to typing. In my opinion, approaching the feature that way completely undersells its true potential.

While it might be a supercharged Google Assistant, similar to Gemini, Voice Mode changes the nature of the interaction. Typing creates a lot of friction that pushes you toward concise, well-formed questions. Speaking completely removes that technical friction, allowing you to think out loud, ramble through complex problems, openly contradict your own points, and let a conversation evolve naturally. That’s actually extremely useful for discussions when researching a topic that doesn’t need, or might not have, a hyper-formed opinion.

I use Voice Mode most often for two things. The first is brainstorming, specifically the early phase of it where the goal isn’t to produce something but to make sense of a set of competing ideas. Talking through a problem while ChatGPT asks clarifying questions and reflects things back is genuinely different from the typed equivalent. Because there is far less pressure to deliver a perfectly coherent thought from the very first sentence, this relaxed approach can often surface unique angles you never would have discovered through a more rigid, deliberate process.

The second is working through on-the-go research or even learning. Activating Voice Mode to ask questions about an ongoing research topic is a favorite pastime of mine, and I do it frequently while out for a walk or driving. Talking through those out loud, with something that can ask useful questions, push back gently, and help you stress-test your own reasoning, is a different experience from just consuming media. Lately, I’ve been using voice mode to research local festivals for an upcoming trip to Japan.

If you take a moment to feed it a specific article or research document beforehand, the collaborative experience becomes even more powerful. Once you’ve used Voice Mode for something genuinely complicated, going back to typing for that kind of work feels like a step backward.

If there’s a task you find yourself prompting ChatGPT for more than once a week, it deserves its own Custom GPT. The setup takes a few minutes the first time, and it pays for itself almost immediately.

Custom GPTs let you bake in a persistent set of instructions, a tone, a role, constraints, and background context that ChatGPT carries into every conversation without you having to re-establish it.

I hate to admit it, but despite building a career in journalism and writing that spans over a decade and a half, I still catch myself mismanaging Oxford commas and em dashes from time to time, much to the ongoing grief of my editors. Sorry, Mitja.

I used to pay for Grammarly to catch most of my mistakes. Now, I rely on a Custom GPT configured with explicit instructions not to alter my original voice, any copy that I’ve written, or writing style, except to strictly flag clear grammatical errors. Once I’ve written something, I dump it there, and it gives me a list of grammar errors that I need to fix. That’s it.

The point isn’t just saving time — though it does that — it also helps with consistency. When a tool has been specifically configured for one job, the results are more reliable and require less correction. In my case, I have complete confidence that the AI is not going to alter my copy; I refuse to let my writing be embellished by AI, but tedious manual tasks like scanning text line by line to verify commas and dashes can easily be automated.

If your work requires it, you can also give your Custom GPT access to files, like a product database or a brand guidelines document, so it always has the right reference material available. Once you’ve built one for a recurring workflow, it’s hard to imagine going back to starting from a blank prompt every time.

This one seems obvious once you’ve done it, but a surprising number of people still describe problems in text when they could just show them.

If your PC is throwing an error, screenshot the error and upload it. If you’re staring at a wall of logs trying to figure out what went wrong, paste in the relevant section or screenshot it. If you are trying to diagnose why your code is not behaving exactly as expected, simply provide a visual of the output right next to the source code rather than struggling to describe what you see. In every case, showing is faster and more accurate than describing, and the analysis you get back reflects the actual information rather than your summary of it.

You’re still getting the same results as you would if you copy-pasted, let’s say, log files. However, you are much less likely to make mistakes since ChatGPT can just as easily parse the text from a screenshot.

There are other use cases too, of course. If you happen to be comparing two complex pricing plans, simply snap a screenshot of the comparison grid and ask the AI which tier aligns best with your specific needs. If you’re looking at a contract clause you don’t fully understand, a screenshot gets you a plain-English explanation in seconds. People spend a lot of time translating visual information into words when they could simply hand over the image.

Open-ended prompts get open-ended answers. That’s true for most things in life, and especially true for LLMs. It can be useful when you want to go down a rabbit hole, but for most practical tasks, it’s not what you want. Constraints don’t limit the output. They help you get more focused output.

Before asking for anything, I tend to spend a few minutes defining the guardrails for what I want. Length is the most obvious one. If I’m asking ChatGPT to rework my sentence for a social media post, giving it a character count constraint makes sense. However, structural constraints can go much further than simple lengths. You can explicitly isolate industry jargon to avoid, lock down a hyper-specific target audience, or force a precise formatting layout, such as a bulleted list instead of a dense paragraph.

The more specific the constraints, the less work you have to do after-the-fact editing the output into shape. It sounds counterintuitive to spend more time on the prompt when you’re trying to save time, but the results speak for themselves. A well-constrained prompt usually produces something usable in one pass. An open-ended one often produces something you then spend more time reworking. It’s an obvious decision.

ChatGPT tends to be extremely good at using a piece of information and reworking it to meet the constraints of other media forms or platforms. So, if you’ve written a long-form article, recorded a podcast, delivered a presentation, or produced any long-form piece of content, you already have the raw material for a dozen smaller ones. It is an incredibly efficient, low-key workflow that many professionals already use to generate multiple forms of content from a single source material. However, you’d be surprised to know that it can be used the other way around, too.

Let’s start with repurposing. Give ChatGPT a full article and ask for a LinkedIn post, a short email newsletter version, three potential angles for a short video script, and a summary that could work as a caption. Each of those is a different format, a different length, a different tone of voice, and a different audience, but they all come from the same source material. You’ve already provided it the source of truth.

The LLM is just reshaping it. What might have otherwise demanded an entire afternoon of tedious reformatting and editing can now be accomplished in under fifteen minutes, and the individual outputs remain highly consistent because they are structurally anchored to the same core ideas. Bonus tip: try creating a Custom GPT for this action to speed up the process even further.

The reverse workflow is just as valuable and a lot more underused. If you’re not sure what angle to take on a piece before you write it, describe the topic to ChatGPT and ask for five different ways you could approach it, each with a different key argument or reader takeaway. No, you’re not outsourcing critical thinking to an AI. Instead, you’re using the AI model as a sounding board to quickly surface options you might not have considered, so you can make a more deliberate choice before committing to a direction.

I’ve found this particularly useful for pieces where I know what I want to write about but not what I actually want to say about it. Seeing three or four alternative framings laid out in plain terms very quickly clarifies which one is actually the right one. More importantly, this approach helps weed out my disorganized train-of-thought or stream-of-consciousness ideas that frequently contain hidden logical fallacies I had completely missed during initial brainstorming. And once you’ve chosen, you go back to doing the writing yourself, just with a much clearer sense of where you’re headed.

If you came here expecting easy hacks to optimize ChatGPT, you might be surprised to learn that they all share a common thread. They all involve giving ChatGPT more to work with and being more deliberate about how you engage with it. More context, more structure, more examples, and more constraints help the model give you better results. Sounds like more work, but it really isn’t.

Most people using it at a basic level are getting basic and generic-sounding results, not because the tool is limited, but because they’re treating it like an old-school search engine where you give it a question and expect the output. The truth is that the only way to extract high-quality output from a large language model is to treat it like a trusted collaborator — or, better yet, a highly capable intern who has immediate access to every piece of context required to execute the assignment perfectly. You, however, still need to give it direction. Give it that direction, and the results follow.