How To Use ChatGPT Smartly

Adding Context

The more concrete guideline you provide, the more likely ChatGPT can give you precise answers. This is what we called adding context in prompt engineering, which means you provide sufficient context in your conversation so that ChatGPT can give you very detailed answers with consuming more tokens.

For instance, let’s say we would like to know how to do online marketing for promoting your cake that you just finished developed after long time. Instead of saying

“Let me know where to start with online marketing for my new cake.”

You can make your questions more concrete with adding background, purpose and problem. For example,

“I am junior marketer and I need to promote my new cake. What can I do for getting brand recognition?”

To be more specific, I just added:

I am junior marketer: Background
I need to promote my new cake: Purpose
What can I do for getting brand recognition?: Problem

But Which Context?

But sometimes it might be challenging for you to think of which context you need to provide. For instance, if you are not a marketer, you wouldn’t know some marketing jargons (i.e. retention rate, bounce rate, CPR or so on).

In this case, you can make ChatGPT ask you back what you want to achieve and make it collect contexts by itself. For instance, if you want to make users stay at your website for a longer time (e.g. reducing bounce rate) but not know which context you need to provide, here are some my real examples that I used.

I: “I just created website for my newly developed cake but I can see customers do not stay in my website for long time. What would be the problem?”

ChatGPT: …

8️⃣ No clear call-to-action (CTA)

• Users don’t know what to do next (e.g., order, see flavors, request a quote). • CTA buttons should be clear, visible, and action-oriented.

I: “I am junior online marketer and I would like to improve my websites with better CTA buttons. What are some best practices to do so?”

ChatGPT: 1️⃣ Use clear, action-oriented text

• Avoid vague words like “Submit” or “Click Here.” …

I wanted to approach to more various marketing solutions and made ChatGPT to ask back what I meant or infer what I was talking about. In this case, ChatGPT tried to infer what I was trying to achieve and it provided some possible problems. At the end, a word called CTA was caught in my eyes and I asked back with using that word.

Avoiding Hallucination

Hallucination indicates falsy or misleading response that AI can give to you against expected answer that you were looking for¹. For instance, when you asked about “X” (former name of Twitter), ChatGPT explained about alphabet “X”.

SearchGPT

To avoid this, you can use SearchGPT so that you can make ChatGPT can give you an answer based on results on web search.

Delimiter

Delimiter indicates some variations for your answers. Normally you can use markdown grammers or JSON format (only if you are familiar with). Here are some examples of using delimiters.

# For Title

You can use "#" in front of it.

And you can also use "-" in front of each list items.

- One
- Two
- Three
- Four

Hallucination (Artificial Intelligence) ↩