7 Tips for Writing Great Content with ChatGPT or Gemini — Whiteboard Friday

Click connected the whiteboard image supra to unfastened a high-resolution version!

So everybody is trying to constitute contented pinch AI these days. You person your celebrated tools, for illustration ChatGPT, Gemini, and Claude. And past erstwhile you put your contented successful there, you springiness it a prompt, right, and the content that ends up coming retired is conscionable garbage, trash, perfectly unusable.

And what I spot astir group doing is that they complain. "Why can't I get awesome contented erstwhile I effort to constitute pinch LLMs?"

But what if location was a way? What if you could constitute awesome contented erstwhile utilizing an LLM?

Hi. My sanction is Chima Mmeje, and I americium the Senior Content Marketing Manager astatine Moz. Today I'm going to beryllium showing you really to usage LLMs, for illustration ChatGPT and Gemini, to constitute content, basically, 7 tips for penning awesome content.

And this is each based connected worldly that I americium doing. So it's not generic proposal you tin get connected the internet. 

Now, earlier we get into these tips, I want to conscionable put a caveat. These tools, by themselves, won't make you a awesome writer. But they tin thief you get much penning activity done.

I opportunity this because LLMs are garbage successful and garbage out. They cannot create authority. They cannot create expertise. They swallow up the full net and are only tin of giving retired the value that you put in.

So the only measurement for you to get bully value from them is for you to beryllium an master writer who understands storytelling. Got it? 

All right, let's go. The first point you want to do is to create a training archive that you will provender it arsenic an example.

So let's opportunity I'm trying to constitute a blog post, right? If I'm trying to constitute a wholly dashing blog station that has a beardown editorial voice, I'm going to prime 1 of the articles I've done successful the past that was a very akin style, and I'm going to usage that arsenic my training document. And that is what I'm going to provender into the LLM each clip I'm trying to constitute that type of blog content.

If I'm trying to do a webinar landing page, the aforesaid point applies. I will return a erstwhile webinar landing page that I person written, and I will springiness it to publication arsenic an example.

Asking these LLMs to conscionable constitute contented for you, pinch nary training aliases context, is only going to lead to bad results.

Now, erstwhile you springiness it that content, the adjacent point you want to do is supply context. I support saying this. You tin never, ever springiness these LLMs excessively overmuch context. There's nary specified point arsenic excessively overmuch context.

Now, what are you trying to execute pinch the contented that you're penning correct now? In the training archive that you've fixed it, what's worked well? What did you for illustration astir the training document? How did you spell from constituent A to constituent B to the extremity of the training document?

You're giving it each this discourse truthful that it understands really it's going to building the content, really it's going to reflector your reside of voice, and really it's going to springiness you a akin output.

Now, erstwhile you supply each of that context, the adjacent point you want to do, you want to create a project. You tin moreover inquire LLMs to thief you create the instructions for the project.

And past you create the project. If you're utilizing ChatGPT, that's going to beryllium a GPT. If you're utilizing Claude, that's going to beryllium a Project. And if you're utilizing Gemini, that is going to beryllium a Gem.

Then you do 2 things. You supply guidelines for writing. I person a super-detailed database of instructions for my GPT that I usage erstwhile I'm writing. It's really long. It has for illustration I deliberation possibly up to 100 slug database of things that I want it to do. In the aforesaid way, I besides person guardrails to avoid, definite words that I don't like. I've noticed, for example, it likes to usage the connection "shape" each different sentence. Or possibly it puts successful an em dash. Or possibly it writes for illustration really short, abrupt sentences that are for illustration 2 aliases 3 words.

All those things that I don't like, that look very AI-ish, that is going to beryllium the guardrails that I'm going to inquire it to avoid.

This is very important because, otherwise, it's conscionable going to springiness you literal garbage.

Now, the adjacent point you want to do is to adhd a archive detailing your halfway offerings and values.

This is truthful important because you're not conscionable penning content, right? I see myself a product-led marketer. So successful that context, each clip I'm creating a blog post, webinar content, immoderate type of contented astatine all, I americium looking for opportunities to plug successful products.

So let's usage Moz arsenic an example. We've sewage offerings for illustration Moz Pro. We've sewage AI Visibility tool. We've sewage MozBar. We've sewage Moz Local and a bunch of different offerings. I'm going to put each of this into a document, and I'm going to explain: What does this instrumentality do? What is the value? Who is this for? I'm going to put successful the positioning statement.

I return that document, and I adhd it to the task arsenic portion of the training archive aliases arsenic portion of the instructions.

That way, erstwhile it's penning aliases generating contented and there's an opportunity to naturally, this is truthful important, to people see that worth successful the content, it's going to put it successful there, and I don't person to do it each azygous time.

When it misses the context, I tin supply that guidance and say, "Hey, you missed this. You are expected to adhd X, Y, Z successful here." Then the adjacent clip it has an opportunity, it understands what the correct discourse is to plug successful value, product-led worth naturally.

A large rumor I support seeing group make erstwhile penning pinch LLMs is trying to constitute the full point astatine once. Give maine a 2,000-word blog station connected X, Y, Z. Help maine constitute a product-led contented that does X, Y, Z. You're not going to get awesome output erstwhile you inquire the LLM to constitute super-long contented because it's going to usage minimal effort. The champion measurement is to constitute successful mini sections.

You person your introduction. I want to constitute an introduction. I want it to beryllium 3 paragraphs. I want the first paragraph to opportunity X. I want the 2nd paragraph to opportunity Y. I want the 3rd paragraph to opportunity Z. That allows maine to power the output. It gives maine that mini section, and past I tin supply feedback astatine a granular level.

"Here's what you missed. Here's what I changed. Compare the 2 of them." You're capable to springiness it feedback and to power the output.

The aforesaid point pinch my adjacent H3. A H3, I'm capable to show it this is what I want to screen wrong this H3. Then, erstwhile it gives maine the output, I'm capable to springiness it feedback. You can't springiness feedback if you're penning the full point astatine once. Where are you going to commencement from? 

Now, erstwhile you're done getting the output from the LLM for that mini conception of content, you want to personalize it. For me, this is the point that makes your contented guidelines retired from that oversea of sameness that we presently spot connected the SERPs.

When you personalize it pinch storytelling, it conscionable makes it guidelines out. It makes it look for illustration thing that you wrote versus thing that an LLM wrote. When you put successful expert-led insights, it conscionable gives it that individual touch. It's for illustration I've had this experience. I've made this mistake. I person done this thing, and I sewage this result. 

Then, finally, you want to springiness it feedback.

Every clip it gives you output and you've added context, you've added storytelling, you've added expert-led insights, you've added anything, you've made changes, you return the last type of what you've written, for that section, and you springiness it backmost to the LLM. You say, "This is the last type I'm going with. I person made X, Y, Z changes because you did not see this, aliases you near this out, aliases you did this. You've gone against our guardrails. You person not followed our instructions." You're capable to springiness it that feedback.

The much you springiness it that feedback, the less mistakes it's going to make. The extremity ideally, for me, is to get to 70% value output. I'm not aiming for 100%. I'm not crazy. Seventy percent, that's the goal. We want to get to 70% of awesome output truthful that we're making less and less changes to the output.

Again, extremity trying to automate content. I don't deliberation that's really viable if you want to support quality. The instrumentality is to constitute successful batches, personalize the output, and past make it consciousness for illustration you. But remember, you person to support training these LLMs because they are machines astatine the extremity of the day.

And that, my friends, is really you constitute awesome contented pinch LLMs. Thank you.