Here’s why you should be using Custom Instructions in ChatGPT

Everything you need to know about ChatGPT’s custom instructions

Chandler K
6 min readAug 10, 2023
[Image by DALL-E and Author]

Custom instructions were just introduced to ChatGPT and they could be the first big step towards AI personalization. This update is just one achievement in a long line of changes that came in recent months and follows in the footsteps of the “system message” which has been available in the API since March of 2023 and was the initial test of personalization. Both the new Code Interpreter and Plugin ecosystem have already had a substantial impact on the ChatGPT user experience, but custom instructions may eclipse both in terms of impact .

In this post, we will explore what custom instructions are, why they make ChatGPT so much more useful, and many examples showcasing how to best use this feature. It’s time to learn!

What are custom instructions?

Like all instructions given to ChatGPT, custom instructions are simply parameters that shape how a model interprets and responds to user requests. These instructions will shape ChatGPT’s responses around a specific intent or goal via the context provided. It can influence the model to follow a specified style, tone, or focus. For instance, ChatGPT can be instructed to give answers in layman’s terms or give more details for those who have expertise in a subject matter.

Right now, ChatGPT gives two different input options for custom instructions: “What would you like ChatGPT to know about you to provide better responses” and “How would you like ChatGPT to respond”. Generally, the first box can be very useful for sharing who you are and what your goals are, whereas the latter option is most useful for styling and tone. We will look at examples in the latter sections so no need to worry if you don’t have any ideas yet for what to include.

How to create custom instructions in ChatGPT

This process is incredibly straightforward. Open a new ChatGPT session and head to the bottom left corner. Click on the three dots next to your username and select ‘Custom instructions’. Once here, you will be prompted to answer two questions that will determine your custom instructions: ‘What would you like ChatGPT to know about you to provide better responses?’ and ‘How would you like ChatGPT to respond?’.

The importance of prompt engineering

When it comes to ChatGPT, custom instructions can be seen as control knobs that are used to fine-tune responses from the model. Those of you who use ChatGPT on a regular basis are already familiar with prompt engineering. In short, prompt engineering is the process of repeatedly tweaking and refining commands given to a model like GPT-4 to optimize the responses.

While this is already possible in ChatGPT, the user would have to manually remind the model of any and all relative context during each individual chat session. This was something that I found myself doing on a regular basis, but would often only remember once I was already deep in the conversation. With this newest upgrade to ChatGPT, I hopefully won’t have this issue as long as I keep my custom instructions up to date.

Why custom instructions matter

The power of custom instructions lies in their reusability. Instead of needing to prompt engineer every chat and message, users can set a consistent set of context to be applied to all future messages. Moreover, these instructions can be tailored at any point and adjusted as your needs of the model changes. You can’t currently set different profiles related to custom instructions but in the future it would be very useful to be able to do so. I am hopeful they add this since many people use ChatGPT at work and for personal uses. So being able to give different context for each would reduce the workload on end users.

One important thing to note is that you cannot bypass the custom instructions once they are set. Even if you explicitly tell the model that you don’t want to use the custom instructions, it still will. Also, custom instructions will impact both code interpreter and any plugins that you have enabled.

Examples of custom instructions

Imagine you consistently use ChatGPT to better understand code and help you program, but you have a limited background in the subject. Instead of constantly having to include that you would like the responses to be explained for a non-tech background beginning coder, you can add custom instructions that will set this for all chats.

1.

This is a simple example, but people are already pushing the limits of what this feature can do. Nivi, the co founder of Angellist, shared a tweet thread with the context he was giving via custom instructions and people are blown away:

Custom Instructions:

- Be highly organized

- Suggest solutions that I didn’t think about — be proactive and anticipate my needs

- Treat me as an expert in all subject matter

- Mistakes erode my trust, so be accurate and thorough

- Provide detailed explanations, I’m comfortable with lots of detail

- Value good arguments over authorities, the source is irrelevant

- Consider new technologies and contrarian ideas, not just the conventional wisdom

- You may use high levels of speculation or prediction, just flag it for me

- Recommend only the highest-quality, meticulously designed products like Apple — I only want the best

- Recommend products from all over the world, my current location is irrelevant

- No moral lectures

- Discuss safety only when it’s crucial and non-obvious

- If your content policy is an issue, provide the closest acceptable response and explain the content policy issue

- Cite sources whenever possible, and include URLs if possible

- List URLs at the end of your response, not inline

- Link directly to products, not company pages

- No need to mention your knowledge cutoff

- No need to disclose you’re an AI

- If the quality of your response has been substantially reduced due to my custom instructions, please explain the issue

Output without the custom instructions:

https://chat.openai.com/share/285cb600-a143-46ce-a9dc-341e88c85283

Output with the custom:

https://chat.openai.com/share/5694b972-80eb-4fc4-9d5f-dea37e237221

Results: As you can see in the later example that includes the custom instructions, the organization and format of the response surpasses the initial output. Additionally, the URLs that are given at the bottom of the chat are helpful for ensuring accurate information and citing sources.

2.

This is another simple example that makes the model more concise in its responses and excels at summarizing information. I use a version of these custom instructions in my everyday chats.

Custom Instructions:

Omit needless words. Vigorous writing is concise. A sentence should contain no unnecessary words, a paragraph no unnecessary sentences, for the same reason that a drawing should have no unnecessary lines and a machine no unnecessary parts. This requires not that the writer make all their sentences short, or that they avoid all detail and treat their subjects only in outline, but that they make every word tell.

Use the active voice. Prefer concrete, physical language and analogies.

Output without the custom instructions:

https://chat.openai.com/share/35a5d41f-555d-4f67-9a81-8e8c9e2b3dd0

Output with the custom instructions:

https://chat.openai.com/share/aeee2831-ee9f-4787-9dcd-26927ba10ed2

Results: These custom instructions are some of the most impactful that I’ve encountered so far. The above chats showcase how different and concise a response can be, while still achieving the desired goal.

3.

These custom instructions are meant to help solve software engineering questions. While the example is on the simpler side, it showcases the potential of finding custom instructions that work best for you.

Custom Instructions:

I’m a Software engineer who uses Python. I have a basic understanding of C#, Julia, and R.

Assume I always want the code written in Python unless I tell you otherwise.

Always ensure that the code is thoroughly documented and in layman’s terms.

Don’t use packages unless required.

If the code can be simplified by using a package, explain the process and ask to use the package.

When importing a package, give a several sentence summary of it.

Assume I have a high level of understanding so don’t worry about complex code.

Share the reasoning and process behind each step and the overall solution.

Output without the custom instructions:

https://chat.openai.com/share/4f34cce5-8f8c-4a98-aa76-2758df54ffe4

Output with the custom instructions:

https://chat.openai.com/share/32e5bc64-b708-4f83-9409-5f10c92fff49

Results: When comparing the two outputs, we can see the differences almost immediately. For me, the biggest difference is the comments that accompany each line of code. While this might not be necessary for most people, it’s clear that the model is willing and able to include them.

Closing thoughts

I hope you found that article helpful for understanding custom instructions. How will your interactions with ChatGPT change? Is this a feature you see yourself using? I believe that mastering custom instructions will allow ChatGPT users to reach new heights in their everyday interactions with the model.

--

--

Chandler K

Harvard, UPenn, prev NASA , writing about AI, game development, and more..