An Easy Way to Improve Your Prompts

Claygent missions, descriptions, & "Few Shot" method

👋 Hey, Claymaker!

If you’re new here, I’m a GTM Engineer at Clay. My goal is to help you mastery Clay and unlock GTM creativity.

I recently returned from Clay’s offsite retreat in Cape Cod where we glamped in modernized airstream trailers. (Autocamp was dope if you’re ever looking for a unique experience on the Cape. )

Honestly, it was one of the most refreshing offsites in a long time. No work. Just team bonding, camp fires, board games, DJ sessions, and even a clown workshop put on by the Founder of House of Yes. As the new kids say these days… it was a “vibe”.

Speaking of vibes… there’s a ton of exciting things coming down the Clay pipeline.

---

Specific Prompts = Better, More Useful Outputs

Prompting can be extremely frustrating at first, and maybe for a while…. When you’re deep in the trenches of trial and error.

But you can get better at it. Much better.

And you’ll be able to automate manual research more effectively and return more structured, usable outputs that are in the style & tone you’re looking for.

Let’s dive into:

1. Example of an input & output of a vague prompt

2. Example of what happens when you focus the prompt with a better description and ‘few shot’ method.

But first…

Example of a Vague Prompt

(This was me not very long ago…)

In this example, I want to find the personas that that a company sells to. This is typically helpful for personalizing emails.

The input:

The outputs:

It’s not structure (for later use in personalized emails) and a bit all over the place.

“Few-Shot” Prompting

A shot is an example.

By providing multiple examples in your prompts, you’re showing AI exactly what you’re looking for.

According to PromptHub, the “few-shot” method is really helpful with the following use cases:

1. Specialized Domains: When working in specialized fields such as legal, medical, or technical domains, where gathering vast amounts of data can be difficult, few shot prompting allows for high-quality, domain-specific outputs without the need for extensive datasets.

2. Dynamic Content Creation: Ideal for tasks like content generation where consistent styles and tone are paramount.

3. Strict Output Structure Requirements: Few shot prompting is particularly helpful in showing the model how you’d like your outputs to be structured.

4. Customized User Experiences: In personalized applications, such as chatbots or recommendation systems, where the AI needs to quickly adjust to individual user preferences and inputs.

Now let’s take a look at what happens…

Improved input:

The examples are for the format of the output.

We still need to demonstrate examples of how the model should transform the inputs into outputs.

To do this we add examples further below in the same prompting window. (Note, you are just inputting the “Expected Response” and the Description and Website are automatically added from the tags you included above.

The outputs

While this can further be improved with some interations, we’re already getting more structured and much easier to use when pulling into personalized emails.

---

✌️That’s a wrap for this week. Hope that was helpful!

My goal is to help people improve their Clay skillset and share what I’m learning as a Clay GTM Engineer. Feel free to drop me any feedback, suggestions, or set up time: alex.lindahl@clay.com.