Custom GPT for Writing Ad Copy

I use ChatGPT daily to help me brainstorm ideas for ad copy, but I have a few particularly frequent use cases. In order to avoid re-explaining what I need every time, I wanted to make a Custom GPT that has the context for different ad types I would like to write.

I’ll explain one example for context:

There is an ad type on Google Ads called a Responsive Search Ad (RSA) that consists of up to 15 headlines and 4 descriptions. Google will mix and match those headlines and descriptions to be as relevant to a search as possible. There is a character limit for these headlines and descriptions (30 and 90 respectively). I will typically ask it to give me ~20-30 headlines for a given context and use the ones that make the most sense.

I would like to build a Custom GPT where I provide the necessary context and restrictions in the prompt or attached files, then provide users instructions or a template for what to enter, and the Custom GPT will help them.

The issue I’m running into is the Custom GPT doesn’t seem to be following my directions, more so than if I use ‘standard’ ChatGPT and explain what I need like normal.

For example, if explicitly say to follow a character limit, it will straight up ignore that sometimes. I understand that it isn’t perfect with character count and might go a character or two long sometimes, it just feels very inconsistent and I feel like I could be writing better prompts. When it gives a bad response, I can respond and say to double check the length, in which case it will make them shorter, but I feel like that defeats the purpose of making a Custom GPT.

I have tried a few things to solve this issue. I tried adding “double check” instructions to the end of the Custom GPT prompt, but this did not seem to help. I would still need to tell it to double check. I tried ALL CAPS for the length limit. Nothing. I tried moving those instructions towards the top of the prompt. Nothing.

I think I have two possible options: 1) Use ‘standard’ ChatGPT and build a prompt to enter before requesting ad copy or 2) Go to the forums and ask for advice.

So with that in mind, here are my questions:

  1. What are some best practices for building CustomGPTs? Should I be strictly using the builder? Is it good to write my own prompts for it? How can I tweak the builder’s prompts to better fit my needs?

  2. Have you seen any existing Custom GPTs for writing search ad copy specifically? What about for Google Ads? Social platforms? I have tried looking around, but haven’t had much luck. I know there is the “Copywriter GPT” by adrianlab, but it seems too generic. I would like to give specific instructions for MY most common use cases.

  3. I am using this as a learning opportunity to become more familiar with CustomGPTs and prompt writing/prompt engineering in general. Do you have any resources for general best practices for using ChatGPT? I have a good idea of the basics, what are some pro tips? Is there a CustomGPT for writing better prompts?

  1. What are some funny/hidden gem techniques? One of my favorites is the “No yapping” technique where adding that to the end of your prompt will greatly improve the conciseness of its answers with minimal input, and it actually works pretty well.

Thanks

1 Like

Hi @whyvalue and welcome to the community!

I know that the PR firm MyNewsdesk uses ChatGPT for creating PR copy - maybe you can reach out to them and bounce some ideas?

Whether GPTs or ChatGPT, for creating any kind of copy, what I’ve seen as the best tactic so far is a “continuously refined multi-turn conversations” method.

Refinement in the sense that you start off from e.g. “what is the primary thing we are trying to sell”, then into “who is our target audience”, followed by “what is the sub-segment of our audience” then into stylistic language, etc etc.

Multi-turn in the sense that for each one of those phases, i.e. “defining target audience”, you spend lots of back-and-forth regenerating content. What helps here is where you guide ChatGPT, e.g. you say “ok, I liked your use of the words X and Y, but I felt it was a little bit too bland, maybe we can make it more punchy”. So you are doing a reward-based guidance where you give it some positive but tell it to adjust.

In terms of how constructing the GPT - I used both PDFs and also just pure prompt - normally I create a prompt using Markdown format, since GPTs love that stuff.

Character limit - unfortunately the tokenization is not character based, but statistical in nature, and a token is on average a 1/3rd of a word. There are cumulative affects, so getting it to follow character limit is near impossible.

What helps is simpler guidance such as “more concise”, “make it shorter”, “use more punchy words” etc.