When Open AI released Chat GPT on November 30, 2022, it became one of the fastest products to reach one million users (5 days compared to Instagram’s 2.5 months and Netflix’s 3.5 years). Since GPT-3 went into public beta in July 2020, we have asked ourselves whether Google would punish AI content and whether it would perform well. Now, we might have an answer.
In my analysis of over 260 pages across bankrate.com, cnet.com and creditcards.com, I found AI content performs as well as human-generated content when done properly. We now have a precedent that will lead to many publishers embracing AI due to significant savings in cost and time. But - as always - there is a bit more nuance to the story.
How I found and analyzed AI content - the methodology
Three big sites have disclosed to create content with the support of AI through a byline or statement (hat tip to Tony Hill for finding it first). After buzz in the SEO community, CNET’s editor in Chief, Connie Guglielmo, published a statement in which she explained that the editorial team was trying AI out to see if there was a pragmatic use case and that every article is fact-checked and reviewed by an expert. This is important, and we’ll come back to it later.
To find and analyze the AI content, I scraped articles linked from the company author profile page and used Screaming Frog to extract the publish date and other data. To date, bankrate.com published 163 articles with the support of AI, cnet.com 75, and creditcards.com 23. With Ahrefs, I was able to bulk-check the backlink profile, (estimated) traffic to each URL and traffic trend over time.
How to make AI content work - examples from Bankrate, Creditcards.com and Cnet
AI content works on these domains for specific reasons. It’s doubtful that a new domain without experts would see the same results, even though we don’t have an a/b test yet.
Each site has a hand full of articles that don’t drive traffic, but mostly due to the fact that the piece was published a few days ago. Google needs to crawl and evaluate content before it can be expected to drive meaningful traffic, no matter whether it’s written by an AI or a human.
Also, keep in mind that several factors decide how quickly and how well content ranks in the search results: keyword competitiveness, authority of the domain, content quality, internal linking, XML sitemaps, and more. The oldest AI-generated article in the analysis was published on Bankrate in early May 2022, the youngest ones were published in the last few days.
The key to success for AI content
Based on my analysis, AI content performs well when three criteria are met:
- The content is reviewed and edited
- The domain is reputable
- The content targets questions with straight answers
Every AI piece on the three domains was at least edited, most of it was edited and reviewed. It’s likely Google rewards subject matter expert edits and reviews, though it’s questionable whether the author name has a positive effect or the factual correctness and wording. But at the very least, expert edits and reviews nurture confidence and trust. The most important point about AI content for SEO might just be that it needs to be fact-checked, corrected, streamlined and sometimes slightly rewritten before publishing. AI content tools save a lot of time and money, but they're not good enough yet to publish the raw output.
All three domains have been around for ages and accumulated strong backlink profiles. Cnet.com launched in 1994, creditcards.com and bankrate.com in 1995. Each site is a company that employs full-time experts following journalistic principles and guidelines.
I looked at the relationship between the # of referring domains and traffic but found none (see screenshots below), links to the exact URL don’t seem to be the deciding factor. Of the 261 analyzed URLs, some had a lot of links but didn’t drive a lot of traffic, and some had few links but drove a lot of traffic. However, the link profile of the whole domain could have a big impact.
Lastly, and this is the second-most important point to understand, the content was created for topics with clear constraints. I call content created for questions like “what does 0 percent apr mean” or “Should you tip your contractor?” functional content. The boundaries of the answer are clear and easy to create for generative AI.
In the example of creditcards.com, you can see very well that articles tend to rank better over time. I looked at the 23 AI-generated articles in two batches since Ahrefs only allows you to enter 11 URLs at a time.
Creditcards.com first batch
Creditcards.com second batch
A sample of AI-generated content on cnet.com shows even clearer how articles need time to reach their potential (see screenshot below).
Cnet article sample
Conclusion: AI content works for SEO when done right
According to Ahrefs’ traffic estimation, Bankrate drives ~125K organic visits every month through AI content, Cnet ~20K and Creditcards.com ~2K. In my professional opinion, we can take this as proof that AI-generated content performs as well as human-generated content when properly reviewed and edited.
Contrast the results above with a site like apnews.com, which has published about 1,700 articles with AI since 2014 and has not seen a lot of organic traffic as a result. The AI pieces get 700-1,000 monthly visits, with the majority going to 20 articles.
Even though AP news has been around since 1991, the content is extremely short and has no expert byline.
There is a lot of fear of Google punishing AI content. And, to be fair, Google has reverted its stance on AI content. In April, they changed the wording in their spam guidelines from “automatically generated content” to “automatically generated content intended to manipulate search rankings”.
However, we can now confidently say no Google update has targeted and punished (good) AI content specifically.
More importantly, this sets the precedent for a new, much more efficient way to create content: AI plus editors and reviewers. Most publishers will follow suit because the economics are just too attractive. As a result, average freelance writers will struggle, but good editors and subject-matter experts will become more valuable.
Longterm, it’s possible that AI content accelerates the “SEO flywheel effect”. Sites that perform well in Search tend to get more backlinks, which in return solidifies their performance. It’s harder for new sites to come up in competitive verticals, hence the flywheel. If AI content performs better on domains with high authority, established sites have an even bigger competitive advantage.
The impact of GPT-4, which is scheduled to arrive this year, will be interesting to watch. To prepare, you might want to give my other pieces on AI a read: