5 custom ChatGPT instructions I use to get better AI results – faster
Follow ZDNET: Add us as a preferred source on Google.
ZDNET's key takeaways
- Turn ChatGPT into a consistent tool with a few tight constraints.
- Use instructions to control tone, pacing, and structured formatting.
- Watch the downside: global rules can silently filter answers later.
Did you know you can teach ChatGPT how to respond to certain requests? Not only can you give ChatGPT instructions, but they'll stick (mostly) for every session.
This feature is called Custom Instructions. It lives in the Personalization tab of ChatGPT's settings. In a minute, I'll show you a set of really powerful directives that can help make you super productive.
Also: I'm a ChatGPT power user: Here are 7 useful settings that are turned off by default
But first, I'll demonstrate how they work in a way that satisfies my inner 12 year old. For this first example, I used the custom instruction, "After each response you give (all responses, including simple procedural responses), display the statement, "Sir, yes sir. How may I best provide AI to you today?"
As you can see, once the AI finished responding to the prompt, it presented me with the ego-boosting waste of cycles I asked it to perform.
This, for the record, is how we use achievements that represent the culmination of all human technological evolution since the beginning of time and the billions of dollars of investment.
'Cause, you know. For the lulz.
(Disclosure: Ziff Davis, ZDNET's parent company, filed an April 2025 lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.)
1. Giving each response an ID
In all seriousness, because you can instruct the AI to add text to every response, you can use the single most useful custom instruction I've found. I have this phrase in my custom setup:
For each response you give (all responses, including simple procedural responses), begin by providing a unique index ID (in the form "ID: XXX" where XXX is a sequential integer beginning with 001) so that I can later refer to any response by number, so that you know which response I'm talking about in our conversations. When responding to text-based questions, please number each answer. However, when responding to spoken questions, do not number the answers.
This is great, because I can now refer back to a statement with an ID number. Here's an example.
So, I can tell the AI something like, "Back in ID 031 you said blah-blah. But could it do blah-blah?" Of course, in a real session, the blah-blah would have some level of meaning.
I find the ID number reduces confusion and helps both me and the AI zero in on exactly what is being discussed. This works particularly well for lengthy typed conversations.
By the way, the reason I don't want it using this feature with spoken questions is because it gets really cumbersome. It halts the flow of the conversation. It doesn't really help when using audio, especially on mobile devices.
2. Removing emojis
Emojis have their place in chats with humans. I particularly like using them to end a back-and-forth conversation with something like a thumbs-up emoji.
But I found it very annoying when ChatGPT peppered its responses with emojis. Even when I used the Personalization options to set Emojis to Less, I still got emojis in responses. So, at the beginning of my custom instructions, I told ChatGPT this.
Never use emojis unless specifically requested, even if you're asked to write at a child's level. No emojis. Never.
It turns out that ChatGPT didn't fully follow these instructions, so at the very end of the custom instructions, I added this.
Again, never, ever use emojis unless specifically requested, even if you're asked to write at a child's level. No emojis. Never.
That worked. ChatGPT doesn't annoy me with emojis anymore.
3. Why it works
This next example is a productivity booster that also provides an opportunity for learning. I often use ChatGPT to help me generate prompts for other AIs, particularly image generators and coding agents.
I have this custom instruction always locked and loaded.
When generating prompts for other AI tools, include a "Why this works" explanation.
This is a particularly powerful teaching tool. The explanations not only provide additional clarification for the current project, they also help me be more effective when I'm creating future prompts on my own. They sometimes show me additional tactics to use when creating prompts and explain why those tactics are effective.
For example, the reason I included a second no-emoji prompt above is because one of ChatGPT's "why it works" prompts suggested that repetitiveness often works to reinforce a directive.
4. Wait until I'm ready
While some of my prompts are fairly short one-liners, I also often give ChatGPT fairly long prompts. I'll type in a bunch of paragraphs or paste in a big chunk of information, often before I give the AI a specific action item to complete.
Unfortunately, ChatGPT has a tendency to take in whatever it's given and evaluate it before it has everything I want it to know. This meant that I would regularly get back a ChatGPT spew well before I was done prompting. So, I used a custom instruction to fix that.
Whenever you receive a large amount of information (defined as more than two paragraphs, more than 30 words, or more than five lines), do not evaluate it. Instead, indicate that you are waiting for my go signal before proceeding. This also applies to uploaded documents.
That prompt itself is 45 words. What this does is tell ChatGPT to wait. Now, when I paste in a longer prompt, it will just respond with a simple statement saying it's waiting for further instructions.
Once I've completed my thought, carefully constructed and entered the prompt, and am ready for the AI to give it a think, then I just tell it, "Go."
For the record, "Make it so" also works.
5. Clean up your output
I'm a bit of a stickler for how I want information presented to me. The structure of the presentation impacts my ability to absorb and integrate it. I'm constantly incorporating and processing new information as part of my job. So getting structured responses, in the form that works best for my soft organic brain, is valuable to me.
I use this custom instruction, which helps break information into helpful and process-focused chunks.
Use headings for responses longer than five lines. Use numbered lists for sequences and bullet lists for collections. Use tables for comparisons by default. Avoid tables that will be too wide for the page to be read. Instead of wide tables, use sets of bullets.
I don't always get back what I want, but this instruction saves a ton of time and helps me integrate ChatGPT responses quickly and productively.
Issues and challenges
Some things don't work, or the AI doesn't really pay attention. Here's one I've taken a bunch of runs at, but it never seems to work.
Have you ever noticed that when you ask ChatGPT how to do something, it likes to tell you everything all in one brain dump? That's helpful when reading, but if you're actually trying to follow the steps, it's completely unmanageable. To solve that, I've tried including this custom instruction in my settings.
For multi-step processes, begin with a concise overview (one-line summaries of each step). Then fully execute only the first step. Pause and wait for my signal before continuing. Never present detailed instructions for more than one step at a time.
Unfortunately, while it does present some steps at the beginning of a response, it often spams five to 10 smaller steps. It doesn't seem to understand breaking down the instructions it presents into single-action responses.
The problem is a second issue with custom instructions. You're only allowed 1,500 characters. So while a much more in-depth prompt could probably get ChatGPT to behave the way I want for this directive, the limit on characters means I can't clarify as much as I want.
Your custom instructions will need to be fairly short, which gives ChatGPT more wiggle room. But they do help, so there's that.
Finally, be careful with what instructions you provide. These apply globally to all your chat prompts. For example, if you use a custom instruction like, "Do not recommend paid tools unless they materially improve outcomes," and a year later you're shopping for photo editors and want a list of the best ones, you may never see paid options. ChatGPT will not, unless specifically requested, tell you it has restricted its answers based on your custom instruction.
What about you? Have you experimented with ChatGPT's custom instructions to shape how it responds? Which constraints or formatting rules have made the biggest difference in your productivity? Have you tried forcing step-by-step execution, banning emojis, or requiring structured output? Did it behave the way you expected? What creative or unusual instructions have you discovered that meaningfully improved your workflow? Share your experiences, tips, and lessons learned in the comments below.
You can follow my day-to-day project updates on social media. Be sure to subscribe to my weekly update newsletter, and follow me on Twitter/X at @DavidGewirtz, on Facebook at Facebook.com/DavidGewirtz, on Instagram at Instagram.com/DavidGewirtz, on Bluesky at @DavidGewirtz.com, and on YouTube at YouTube.com/DavidGewirtzTV.
Post Comment