Prompt Engineering with Perplexity and Grok 4: AI Optimized Articles

The Goal:

First, let’s state what my goal was for this custom instruction prompt.

The main goal is to successfully generate a full outline based off of one keyword phrase. Along with the generation of related semantic phrases to include in the article; Such that the contents are optimized for discoverability within search and generative engines.

How We Get There - Initial Steps

How I started: basic prompting, detailing and giving context for what it is we are doing and what we are trying to accomplish.

Good, but not exactly what we are looking for. So I re-prompt again, using Grok 4, giving more specific context.

More or less, this is what you want in your instructions prompts; an example format for the result the LLM should give you is what will confine the model to only give us what we really want, while cutting out the noise.

Heres the rest of the result:

Prompt Test

Now that we have a nice looking instruction prompt, along with an example template, lets test this out with a random keyword phrase.

Lets use deep research for the best results.

As you can see, we have a full outline of headers to choose from, along with subheadings to help us structure the article in accordance with ai overview and ai summary optimization.

The related semantic phrases can also be used within the writers main bodies of content to boost discovery.

There are some optimized article headliner titles if you choose to use them, along with a few other tips that a writer can use with this information.

Try It Yourself

If you'd like to try it yourself, and optimize the prompt for yourself, you can use it at this perplexity spaces link.

From that point, If you would like a specific amount of headlines to have pre-formatted/want to limit the amount of headlines generated, simply edit the instructions area here:

Your experience and quality may vary. Based on the prompt optimization, and deep research being enabled, it is likely best to keep deep research on. At least for the initial prompt. And then any optimizations you may make, switch the model to Grok 4.

I hope I was able to help you learn something new in this emerging space of prompt engineering.

Thank you for reading.