Is Google Penalizing Sites with AI-Written Content

Neil Patel Says Yes

Nov 22, 2022

Marketing guru Neil Patel recently reported he analyzed 100 of his websites with AI-generated content, and many experienced significant traffic declines after the recent Google spam update. Actually, the company has 681 websites with AI-generated content, but only 100 averaged at least 3,000 visitors a month. That is his threshold for visitor counts that don’t exhibit too much daily traffic variance, according to Patel.

He said the purpose of these websites is to monitor Google search results for a variety of topics and specifically to test how it treats sites with AI-generated content compared to similar sites with some human and AI content and those with content solely created by humans.

The true test came a few weeks ago when Google introduced its latest Google Search Spam Updates. Google does have a policy around what it calls “Spammy automatically-generated content.” The formats that will get you a Google downranking include:

Text that makes no sense to the reader but contains search keywords
Text translated by an automated tool without human review or curation before publishing
Text generated through automated processes without regard for quality or user experience
Text generated using automated synonymizing, paraphrasing, or obfuscation techniques
Text generated from scraping feeds or search results
Stitching or combining content from different web pages without adding sufficient value

The bolded section above is our emphasis added. Patel says that of the 100 websites that his team evaluated, 53 included only unaltered AI-generated content. The other 47 included some editing, manually adding legitimate internal and external links, and video embeds.

Google Docks AI-Generated SEO?

You can see in the chart above the sites with unaltered AI-generated content saw average traffic declines of 17.29%. The sites with human-edited AI-generated content were hit with only a 6.38% decline. You can also see the keyword search position drop was more than double for the unaltered AI content sites. Patel does discount this a bit, saying the significance is less than it might seem since the keyword rankings were not particularly high for these sites. Either way, it seems like an open-and-shut case. The traffic drop alone should concern anyone using unaltered AI-generated content.

But there is more to the story.

Patel reports that only 14 of the 53 sites (26%) with unaltered AI-generated content were hit by the update. Their average decline was 51.65%. This is a more significant number since it eliminates the sites that weren’t affected and shows what can happen when a Google spam update ensnares your website.

Marketers are using AI-generated content to enhance their website performance. A big part of that is chasing SEO rankings. Jasper AI alone claims over 70,000 customers, but Patel’s unaltered content sites aren’t a realistic proxy for the vast majority of these customers. They are almost certainly actively altering the content. It’s fine to have an extreme case as a baseline. It is the human-edited sites with AI-generated content that are going to be a more interesting guide.

Of those 47 sites, only eight (17%) were impacted. As expected, human editing and curation will do a better job of hiding the “text generated through automated processes.” In addition, the traffic impact will be less but still very substantial.

“On average those 8 sites saw a 42.17% drop in traffic…They lost between 29.52% and 81.43% of their traffic,” says Patel.

I spend a lot of time dealing with data. A sample size of eight is not overwhelming evidence that should drive expectations. But let’s say that if you are using AI-generated content and adding curation, you may have as much as a 17% chance of getting penalized by Google. If you receive the penalty, your traffic could fall as much as 80%. We all would like a sample size of 1000 sites to offer better predictions. With that said, the smallest traffic drop was 29.52% of the tracked websites hit by the update. This should concern marketers that they may have a one-in-six chance of losing 30% of their monthly website traffic.

Synthedia is free. Share this post with a friend or a marketer. 😃

Patel also reported which content sectors saw a rise in SERP volatility after the Google spam update. News, Sports, Arts & Entertainment, Technology, Communities, and Shopping all showed a one-third or higher increase in SERP volatility. These are the categories of content that may be most at risk of penalization by Google. Other segments may be the target of future Google spam updates.

Lessons for Marketers

There is no free lunch. Using AI to generate content may supercharge your marketing content output. Then again, 22% of sites using at least some AI-generated content saw dramatic drops in traffic. That is a one-in-five chance of losing several SERP rating positions and a low-end drop of 30% in monthly traffic. You also have to consider that Google’s next spam update will likely catch more of the AI-generated sites. The new feature introduces new risks, according to Patel’s data.

Does this mean you shouldn’t use AI-generated content? No. It likely means you need to use it the same way you would with completely human-generated content. Make sure it is edited to have unique elements and that the content delivers value to readers. That will help generate backlinks and give Google a signal that your website and pages are legitimate and valuable to users.

As a final warning, you should consider the role of a blog in your marketing. If you sell a product, the blog is designed to help with SEO, customer journeys, lead capture, and sales. If your entire site gets downgraded because of the blog, and fewer prospective buyers see the products and lead capture incentives, that will severely undermine your business.

Impact on AI Content Generators

The final point to consider is how this data will undermine the business models of the many AI-based content generators. The impact of Google’s most recent spam update appears to be limited. That could change over time and ensnare more websites over time. It’s a market risk that few people have considered.

In addition, almost every text-to-text content generator uses some flavor of GPT-3. And there are thousands of marketers using the very same implementation of GPT-3. Each user prompt will vary somewhat, and each output of text generators will include some variance. Then again, at some point, we are going to start seeing so much AI-generated content for certain topics that it will begin to look the same. Maybe this is hardly different than the blatant copying of content that goes on, but it’s a new world, and the impacts may be significant.

If you are using or tracking the market for AI text-to-text generators, this is an important trend to monitor. Granted, these solutions do more than just generate SEO-optimized content. Blog content was an important early adoption driver for the industry, but it has also set the stage for wider use which could ensure the long-term stability of the solutions.