AI-generated content is quite popular among businesses and marketers. A lot of businesses use it without a clear understanding of its impact on SEO and organic traffic.
That’s the problem we used to face at Digital Marketer PK when convincing our clients to not go for AI content on their company’s blog – but not anymore.
We conducted a 13-month experiment to understand the impact of AI content on a site’s organic traffic and how search engines respond to human-edited AI-generated content.
Keep reading to discover the findings and 5 lessons we learned from the experiment.
The Backstory and Idea
Back in 2023, AI-generated content was the talk of the town (it still is). A lot of bloggers and businesses were publishing AI content at scale. At Digital Marketer Pk, we decided to run an experiment and see how AI content works.
As an inbound marketing agency for SaaS companies, we used to hear a lot about AI content from potential clients (and still do). Before we recommend our clients to use AI content for inbound marketing, we wanted to figure out if it works.
Back then, we were against AI content as it takes more time to create, edit, fact-check, and redo it than creating content from scratch. If you have to write people’s first content, it should be crafted by a subject matter expert.
It was hard to convince clients to stick with quality content when they used to see AI content everywhere.
We decided to see for ourselves how AI content performs in SERPs and how we can use it to help our clients with content marketing.
The experiment started.
We finalized a niche, did keyword research, and found 500 keywords and prompts for the AI content. We purchased a new domain in July 2023 and started publishing content.
The blog was in the business niche and it was more of an aggregator where we decided to collect and publish all the information about specific companies and list them for users.
Content Writing and Publication
We used a popular AI tool to create content using the template that we created in-house. This means all the articles we published had the same template and the same number of heading tags.
We tried multiple AI content generator tools and selected the best one. I have to admit that AI-content generating platforms aren’t too great. You can’t rely on their output content as it contains a lot of errors.
We generated all the content in a single month since we had done all the keyword research.
The content was edited heavily to fix factual errors and remove repetitions. I noticed that the content was unique, but it wasn’t original. It was more of scraped content gathered from different sources and reworded (that’s how AI content works).
Note: We didn’t add any commentary for the posts rather we only fixed errors, removed unnecessary content, and added updated information.
Content was drip-fed over a period of time. The first post went live on July 6, 2023, and the last post (#138) was published on July 17, 2024.
Here’s some more information about the articles and blog:
- The average word count per post is 1500+ words
- Unique featured image for every article
- The articles were properly formatted with HTML tags, outgoing links, internal links, and formatting
- Twenty-Twenty-Two WordPress theme used
- Essential pages added such as About, Contact, and legal pages
- We didn’t follow a fixed publishing schedule. The first batch of articles was published fairly quickly with multiple articles scheduled per day. Then we let them index and stopped publishing new content
- No social media accounts created
- All the new posts were submitted manually for indexing to Google Search Console and we used IndexNow plugin for Bing
- New posts were indexed almost immediately by Google while Bing took a lot of time
- No backlinks were built ever
- Basic on-page SEO done.
Traffic Statistics
Over the course of 13 months (July 1, 2023 to July 1, 2024), the blog received 41K active users with 35.7K organic visitors.
This was all-natural, organic traffic with zero word of mouth or any other form of online or offline marketing.
The rankings gradually improved as content was crawled and indexed by search engine crawlers. It took approximately 5 months when we started seeing some decent organic traffic in January 2024.
Google and Bing were the main contributors to organic traffic. Over 13 months, 36% of organic traffic was received from Google and Bing each while the remaining 28% was received from other search engines:
The content was indexed by all the search engine crawlers indicating that it wasn’t something spiders hate too much. The organic traffic continued to grow as we kept adding new posts.
Note: Yahoo, DuckDuckGo, and Ecosia use Bing search, so it’s collectively traffic from Bing. We reported it separately for better understanding.
Google March 2024 Core Update
Everything was moving in an upward direction until March 2024. Google’s March 2024 core update turned the tables and the organic traffic from Google dropped in the first week of March.
Meanwhile, the organic traffic from Bing remained consistent.
Google was the top contributor from July 23 to February 2024 with 51% of total organic traffic followed by Bing at 27%:
This changed drastically after the core update. The traffic from Google dropped to 19% while traffic from Bing jumped to 46%:
One reason why traffic from Bing increased during this period is that it was slow to crawl and index content.
We weren’t too surprised to see a hit from Google in March 2024. It was a scaled content abuse that punished AI-generated content that was created to manipulate search engine rankings instead of helping users. Scaled content creation through AI or human-generated content that’s not useful for readers is what Google strictly prohibits.
I’m pretty sure the blog was hit with scaled content abuse.
Google didn’t look at it from the perspective of how the content was created (as it’s mentioned in the policy). The algorithm looked at the helpfulness of the content (helpful content) and figured the content was written to manipulate search rankings.
Post-Update Traffic
The blog continues to receive traffic and gets over 2K active users per month organically. Traffic from Google declined significantly after July 2024 while traffic from Bing remained consistent (probably because we didn’t publish new content).
Traffic from Google dropped to 2% whereas Bing drives 55% of all organic traffic. Google doesn’t like the site anymore:
The domain and posts are indexed in Google but nothing ranks anywhere in SERPs. Some posts were de-indexed, but Google crawls and indexes them once submitted for crawling manually. As of October 2024, we have 129 indexed pages in Google Search Console (out of 142 indexable URLs including pages and posts):
The sitemap is crawled and indexed regularly without any issues:
Interestingly, we haven’t touched the blog since July 2024. The blog receives 50+/- crawl requests per day which indicates that the URLs aren’t deindexed.
We can argue that fixing content issues might improve organic traffic and ranking. This, however, needs to be tested.
Nobody knows for sure what’ll happen if we transform content completely.
Findings
Here’s a list of the primary findings and learnings from the experiment:
1. AI Content is Scraped Content
You can argue otherwise.
Scraped content refers to any content that’s stolen from another website without permission. This type of content violates Google policy.
AI content generators use ChatGPT, Claude, or similar tools that don’t generate their own content, rather they find content from their databases (or online) and reword it.
In the first decade of the current century, content spinning was a popular way to write original content. Writers used to rewrite other people’s content in their own words known as content spinning. Then we had tools that were used to spin content at scale which was tackled by the Google Panda update in 2011.
AI content is a sophisticated form of spun content.
Unfortunately, most of the leading AI content generators use the same technique. The tool we used had its own database of content and it rewrites existing content in its original form.
The type of content Google and your readers want should be much more than unique. It needs to be helpful, deliver value, and must offer a solution that’s not just obvious.
The problem with AI content is that it states the obvious. It lacks the ability to create arguments and counter-arguments like a human.
This leads to the second biggest issue with AI content.
2. Your Niche Plays a Major Role
AI can only generate content on popular topics. It can’t generate content for a focused niche. That’s the problem we faced.
We had to choose a broad industry. We tried to narrow it down to a niche, but AI tools were unable to generate content for them.
We were targeting long-form content and had to stick with popular topics to create long-form content.
This is the reason we opted for the business category and covered broad topics about popular companies, their revenue, operations, net worth, shares, board members, etc. These details are easiest to get through an AI tool.
If you need content on an unpopular, new, or trending topic, you won’t get it.
Broad and popular topics have high competition in SERPs. This makes it quite hard for newcomers to rank in SERPs with AI-generated content.
3. AI Content Needs Rigorous Editing
We spent, on average, 1.5 hours editing per AI-generated article that covered all of the following:
- Tweaking the first few lines of each paragraph as they usually repeat the heading and offer no value
- Checking if the content in each heading and new section is actually about the corresponding heading
- Fact-checking what’s reported. This took the most time. AI can make up things. If you ask AI to cite the source, it just cites any random article on the web that has nothing to do with the topic
- Fixing typos and errors
- Finding repetitions and fixing them. When you are churning out a long-form article, AI keeps repeating itself. This was the most boring part as we had to remove a lot of content that was repeated multiple times
- Rewriting content takes a lot of time. It’s nearly impossible to copy-paste content from an AI tool on your blog. You have to add some personality and rewrite most of it
- Content formatting and on-page SEO took a lot of time.
After doing all of this, your content still looks AI-generated.
This is a reason why I say that editing AI-generated content is harder than writing a new article from scratch.
4. If You Scale It, You Can Generate Organic Traffic
It’s a fact and a reason why a lot of AI sites are still out there.
While you might not be able to rank in Google with AI content, Bing ranks all types of content.
If you target Bing and publish AI content at scale, you’ll be able to drive massive traffic from Bing and its partner sites. Our blog is still getting traffic from Bing and all the pages are ranked (most of them on the first page).
This is without any active link building.
I’m not sure how it will impact your business in the long run, but you can enjoy traffic in the short term.
What this means is that AI content isn’t bad, you just need to know how to use it for content marketing. You need to target the right channel and the right search engine to generate organic traffic and it might do wonders for you.
5. It’s Risky
At the end of the day, AI content on your site is risky.
Even after major edits, you can’t be 100% sure if Google won’t punish your site.
One thing we know is that it was a site-wide action by Google. If you publish a mix of AI and human generated content, your entire site will be hit – not a few pages.
This is because scale content abuse isn’t about how you generate content – it’s what type of content you publish. Once your domain is flagged, you’ll have a hard time getting it fixed.
AI content is inherently low-quality content that lacks value. It can be used for a social post and other meaningful purposes, but you can’t use it on a site and expect people to love it.
It is also not suitable for branding. It has the potential to negatively impact brand reputation and credibility.
Should You Use AI-Generated Blog Posts on Your Blog?
No.
Anyone who has been into online marketing pre Google Panda update knows very well what spun and scraped content is and how good Google is at catching it.
AI content is one level better than that, but it’s still the same old content with no value.
If you want to get help from AI, use it for other purposes such as headline generation, idea generation for blog posts, outline development, meta description writing, title generator, and social post creation. You can do a lot with AI content and use it in effective and productive ways.
Using it for blogging is not recommended.
You might get away with it and might be able to trick Google, but it’s not a long-term, sustainable growth model for any business.
