How Much of Your Content Can AI Write?

Time to read: 10 minutes

My name is Brian, and I’m the SEO director here at Sixth City Marketing. Today we’re going to talk about how much of your content AI can write.

I know you probably came here for a hard stance on this, and I’m going to take one.

Think of it as a slider. As the percentage of AI-written content goes up, the system you need behind that writing to support its quality also has to increase in sophistication. That’s the first thing to remember.

If you are just giving ChatGPT a topic, having it write, and then publishing it on your site, that is definitely not the way to go. And I’ll prove it with statements from Google and some other information.

The Two Ways People Ask This Question

Normally, when people ask this question, they mean two different things:

• One single asset: You’re working on a blog post, and you’re wondering, can I use AI to write it?
• The whole website: What percentage of my overall website can be AI-written?

It’s essentially the same question. If you’re over-relying on AI to write single assets, those eventually accumulate into a higher percentage of your overall website being AI-written. So we can approach the answer the same way, and it takes care of both questions.

Why You Can’t Just Prompt ChatGPT and Publish

Why can’t we just prompt ChatGPT with a topic, let it do its thing, and publish that? Because we keep watching that fail.

One of the reasons it fails is that when you give AI a topic, it goes out and finds a number of resources, which include other web pages. It mashes it all together, and it basically just spits out the average. And an average is not high performance.

That’s why it’s so important and effective for you to stand out – because you’re going to offer information that cannot be found anywhere else and is much more likely to be included in an AI output.

What Google Actually Says About AI Content

Google has a specific statement on AI content, and really any content. What Google is looking at here – and what it is not going to react to favorably – is content that is:

Now think about giving a topic to ChatGPT. That’s little to no effort. What it spits back at you is average, and that’s little to no originality. Therefore, you’re offering little to no value. So you’re not going to see the performance there.

More AI Means a More Sophisticated System

I want to come back to this idea: the more you rely on AI, the more sophisticated the system behind that writing needs to be.

Without that, you’re going to get average slop in return. You also need very careful supervision at multiple points in the larger pipeline – checking in with the sub-outputs coming through before the final output. Otherwise the AI is going to drift, it’s going to get way off course, and your result is going to be very poor.

What a Genuinely Sophisticated System Looks Like

Here’s what a genuinely sophisticated AI system looks like. I took this from Ryan Law at Ahrefs. He uses it for some informational content – and I’m being very specific there, because there are a lot of other content types he does not use the system on. So even with a sophisticated system, it’s a very limited scope of content he’s creating with it.

On the left, we have the example of you just giving ChatGPT a topic, asking it to write, and going ahead and publishing it. Not good. Hopefully that’s clear by now. Here’s what an actual system looks like:

The Inputs Layer

• Your full editorial process, documented for the AI
• Writing samples it can load in as context
• A human brief that sets the angle you’re going to take and some of the anecdotes that should be used in the final piece
• Proprietary data of all kinds
• Business priorities – really anything and everything to make sure the tool is working right for you

The Supervision Layer

Then we’ve got the supervision layer. You review at three points:

1. At the outline step
2. When the draft is produced
3. At the final output

For that final review you need someone who really understands search systems and audience copywriting, as well as the business’s proprietary data and goals, to make sure it’s what you’re going for.

The Only System You Actually Need

Here’s the thing though: the less you rely on AI, the easier life gets and the better the performance gets. This is the only system you need – one short chat a month with your agency.

We don’t need an overly elaborate system. We just need an expert, some collaboration, and an agency that specializes in AI search and gets the benefits from it while also achieving excellent results.

What the Right 25% Actually Looks Like

Let’s talk about what that 25% AI should look like, because I’m really specific about this. It has to be the right 25%, or you’re really not following this philosophy.

All of that is fine, and AI can really help improve your content just by doing these simple things. But the body content – your expertise, your stories, your opinions, your recommendations – we don’t want to flatten that out. That’s what’s really adding value to the web and to a user’s query, which is why you’re going to rank really well and appear in more AI outputs.

My Rule of Thumb

“Couldn’t I Just Dictate My Expertise and Let AI Write It?”

As a business owner, you might be saying, “Okay, confession – I’ve given ChatGPT a topic, had it write the whole thing, and published it before.” But after this, couldn’t I just dictate my expertise on a topic, give that to ChatGPT, and publish that?

You could. That is better than just giving it a topic and having the AI write. But it’s still vastly incomplete, and that’s why even that isn’t going to get you high performance. Let’s look at the full picture.

Your dictated draft sits in the middle. Here’s everything around it:

1. Deep keyword and prompt research – before anything is written
2. Audience research from all different angles – because we want to prepare to build the perfect piece of content
3. Your dictated expertise – this is where your expertise comes in
4. On-page optimization – once we’ve extracted it
5. Formatting for scannability and large language model interpretation
6. Very deliberate and targeted internal linking
7. Copy that converts

The bottom line: it’s really important to take in this full process when you’re considering questions like that.

What This Looks Like in Real Results

When you put a simple system in place, use the 75/25 split I’m recommending, and have a collaborative client who can feed us that expert information – these are the results you can get.

Year-over-year gains from that same campaign:

Source: Sixth City Marketing client campaign data, wildlife removal company, year-over-year.

That 86% improvement in site clicks matters because a lot of their competitors – and many other websites – are actually seeing a drop in traffic since Google’s AI Overviews are answering more queries than ever.

And the 10% improvement in time spent reading is a really important one. That’s something human expert content does. It engages people. It gives them that sense of you being there with them, giving that arm-over-the-shoulder advice in person. It’s about as close as you can get to that on the page.

The jump from position 3 to position 1 for their highest revenue query doesn’t sound like a lot, but the click and conversion difference you get from a jump like that is very significant.

The Bottom Line

I hope that clears things up and helps you understand why you can’t just give ChatGPT a topic, have it write your content, and watch your site perform well. It’s just not going to cut it.

There’s no reason to overcomplicate things. Work with an agency that has the right systems in place and understands how to use AI as leverage instead of something that’s going to knock you down in the rankings. You’ll get the efficiency gains, but you’ll also see your business grow.

If you want to know what your best opportunities are, reach out to us for a free site assessment, and we’ll be happy to talk to you about your most effective levers.