I published an article last year that was about 60% written by Claude and 40% written by me. It ranks number two for its target keyword. It generates around 1,200 organic visits per month. It has been cited in three AI Overview responses that I know of.
I published another article around the same time that was 100% written by ChatGPT with minimal editing. Same site, same general topic area, same internal linking strategy. It sits on page four. It gets maybe 15 visits per month. It has never been cited by an AI platform.
Same tools. Same site. Radically different outcomes. And I think the gap between those two results tells you almost everything you need to know about how AI content works on Google in 2026.
What Google actually says (and what they actually do)
Google's official position on AI-generated content has been consistent since February 2023 when they published their guidance. They do not penalize content for being AI-generated. They evaluate content based on quality, helpfulness, and whether it demonstrates experience, expertise, authoritativeness, and trustworthiness. The method of production, whether it is human, AI, or some combination, is not itself a ranking factor.
This position has not changed in 2026. Google does not have an "AI content penalty" in the way that many people seem to believe. There is no sitewide downgrade for using AI tools. There is no detection system that identifies AI text and automatically demotes it.
What Google does have, and what has gotten dramatically more aggressive over the past year, is a set of quality signals that happen to correlate strongly with mass-produced AI content. If you prompt an AI to "write a blog post about kitchen remodeling tips" and publish the output without meaningful human contribution, you are publishing something that is indistinguishable from what thousands of other people are doing with the same prompt. Google's algorithms spot that pattern because they detect sameness. Not AI-ness. Sameness.
This distinction matters a lot. Google is not looking for AI fingerprints in your text. They are looking for content that adds nothing new to what already exists on the web. The fact that most zero-effort AI content fails this test is a consequence of how AI works, not of any anti-AI bias in the algorithm.
Reputable sites like CNET, Bankrate, and several major publishers use AI in their content production workflows and continue to rank well. They use AI responsibly, with human oversight, editorial judgment, and genuine expertise layered on top. The tool is the same. The process is different.
The Semrush data: what 20,000 URLs tell us
Semrush conducted one of the more interesting studies on this topic, analyzing 20,000 blog URLs through a multi-layered research approach to determine how AI content performs in search results. The findings are nuanced in ways that defy simple takeaways.
Content that scores 8.5 out of 10 or higher on what Semrush calls "semantic completeness," meaning it thoroughly covers the topic from multiple angles, is 4.2 times more likely to be cited in AI Overviews. This matters because AI Overview citations correlate with higher organic rankings and cited pages earn 35% more organic clicks than pages that are not cited.
Here is the part that I think is most telling: 47% of AI Overview citations now come from pages ranking below position five in traditional organic results. That is a remarkable statistic. It means that a page does not need to be in the top three organic results to get cited by AI Overviews. It needs to be the most semantically complete answer to the question. And semantic completeness is exactly where the gap between pure AI content and AI-plus-human content shows up most clearly.
A pure AI article on kitchen remodeling tips will cover the obvious: budget planning, choosing materials, hiring contractors, timeline expectations. It will do this competently. Every point will be accurate. The writing will be clear. And it will be functionally identical to every other AI-generated article on the topic.
An AI-plus-human article on the same topic might include a paragraph about how the author's own kitchen renovation went $12,000 over budget because they did not account for the electrical panel upgrade their 1960s house needed. It might mention that in Portland specifically, permit processing times have doubled since 2024 and you need to start that process earlier than online guides suggest. It might note that the Semrush data on average renovation costs skews high because it over-represents coastal markets.
Those human contributions are what push the semantic completeness score past the threshold where AI systems start citing you. The AI draft provides the structure and coverage. The human provides the specificity, the experience, and the genuine insight that no language model can fabricate.
The "AI plus human expertise" framework
I have landed on a framework that I use for all content production now, and I am going to describe it in detail because I think the process matters as much as the outcome.
Step one is what I call expert research. Before I touch an AI tool, I spend time gathering original inputs. This means conducting interviews with subject matter experts, reviewing primary data sources, collecting personal anecdotes and case studies, and identifying angles that are not already saturated in existing content. This step takes 30 to 60 minutes per article and it is the most important step in the entire process.
Step two is AI drafting with expert inputs. I use the AI as a writing accelerator, not as a writing replacement. My prompts include the specific data points, expert quotes, case studies, and original angles from step one. The AI organizes and expands this material into a coherent draft. Critically, I am not asking the AI to research and write. I am asking it to write using research I have already done. This is a meaningful difference because the AI's output is now grounded in original material rather than in its training data, which is the same training data everyone else's AI is using.
Step three is human editing with a specific checklist. I go through the draft looking for five things. First, specificity: I replace every generic claim with a specific one. "Many businesses find that..." becomes "Three of my clients in the SaaS space found that..." Second, recency: I add current data, reference recent events, and remove anything that feels dateless. Third, experience signals: I add personal observations, I disagree with common advice where I genuinely disagree, I admit uncertainty where it exists. Fourth, local or niche specificity: I add details that are relevant to the specific audience and that a generic AI would not know. Fifth, citation quality: I verify every factual claim and add links to primary sources.
Step four is what I call the differentiation check. Before publishing, I search Google for the target keyword and read the top five results. Then I ask myself: does my article contain at least three pieces of information or perspective that none of those five results contain? If the answer is no, I go back and add more original material until it does. This single check has done more for my content performance than any other quality control measure.
The entire process takes me about three hours per article, compared to about five hours for fully manual writing and about 30 minutes for pure AI output. It is not the fastest approach. It is not the cheapest. But it is the approach that consistently produces content that ranks and earns citations.
What mass-produced AI content looks like to Google
I want to paint a picture of what is getting hammered in search results right now, because understanding the failure pattern helps you avoid it.
There are sites, and I am sure you have seen them, that published hundreds or thousands of AI-generated articles in the past year. Many of them targeted long-tail keywords with content that was factually correct, grammatically sound, and completely devoid of any original thought. They followed a template: introduction explaining the topic, five to seven subheadings covering obvious subtopics, a conclusion restating the main points. Each article read like a Wikipedia summary rewritten by a competent but incurious editor.
Google's December 2025 core update hit many of these sites hard. Not all of them, and not equally, but the pattern was clear. Sites that relied on volume over originality saw significant declines. The sites that survived were the ones where AI content was augmented with genuine expertise, original data, or unique perspectives.
The reason, I believe, is not that Google detected AI text. It is that Google's quality algorithms, which have been iterating toward identifying helpful versus unhelpful content for years, finally got good enough to distinguish between content that exists to rank and content that exists to inform. AI happened to make it very easy to produce large quantities of the former, which created a target-rich environment for Google's quality systems.
I know a site owner who published 400 AI-generated articles in four months. His organic traffic peaked at about 85,000 monthly visits, then dropped to 12,000 after the core update. He told me Google is biased against AI content. I pulled up ten of his articles at random and every single one could have been generated by anyone with a ChatGPT account and the same keyword list. There was no experience. No data. No perspective. No reason for Google to prefer his version over the identical content on fifty other sites.
Compare that to a client of mine who publishes eight AI-assisted articles per month, each one incorporating original survey data from their industry, quotes from practitioners, and specific case studies from their consulting work. Their organic traffic has grown 40% year over year and they have picked up AI Overview citations for 23 of their priority keywords. Same tool. Different process. Different results.
The experience gap that AI cannot close
I think the single most important concept for anyone publishing AI content in 2026 is what I call the experience gap. This is the difference between information that exists in an AI's training data, which is available to everyone, and information that exists in your unique experience, which is available only to you.
When you prompt an AI to write about a topic, it synthesizes information from its training data. That information is, by definition, not original. Every fact, every framework, every piece of advice in the output already exists somewhere on the web. The AI is recombining existing knowledge, not creating new knowledge.
Your unique experience is the opposite. When you describe a specific client engagement that went sideways, or share data from a survey you conducted, or explain why the conventional wisdom about your industry is wrong based on what you have seen firsthand, you are producing information that does not exist anywhere else. No AI can replicate it because no AI has lived your professional life.
This is why I am so insistent on the "expert research first" step in my framework. The AI is a writing tool. The originality comes from you. And originality, measured through Google's lens as unique information, first-hand experience, and genuine expertise, is what separates content that ranks from content that exists.
I have noticed that the most successful AI-assisted content tends to follow a 70/30 pattern. About 70% of the content is foundational information that the AI handles well: definitions, background context, general best practices, step-by-step processes. The other 30% is original material that only a human with relevant experience could provide: specific examples, original data, contrarian opinions backed by evidence, warnings about common mistakes based on firsthand observation.
That 30% does disproportionate work. It is what makes the content unique. It is what earns citations. It is what gets linked to. And it is what signals to Google's algorithms that this content adds something to the web rather than merely restating what already exists.
The practical question: should you use AI for content?
Yes. Obviously yes. I do not think there is a credible argument for avoiding AI tools in content production in 2026. The efficiency gains are too significant, the quality of AI writing has gotten too good, and the competitive disadvantage of fully manual content production is too large.
But there is a version of "using AI for content" that works and a version that does not. The version that works treats AI as a collaborator that amplifies human expertise. The version that does not treats AI as a replacement for human expertise.
The 73% of marketers who use a combination of AI and human writing are outperforming both the holdouts who refuse to use AI and the operators who over-rely on it. Specifically, 69% of marketers refine AI drafts with human editing, 48% build on initial AI drafts rather than publishing them as-is, and 55% conduct original research to strengthen their AI-assisted content.
I think the market will continue to reward this hybrid approach for as long as AI systems are trained on existing web content. Because as long as AI can only recombine what already exists, the people and organizations that contribute genuinely new information, new data, new perspectives, new experiences to the web will maintain an advantage. AI makes them faster and more productive. It does not replace the need for them to actually know things and have opinions about what they know.
The day that changes, everything I have written here becomes obsolete. But I do not think that day is 2026.
A note on detection and disclosure
I get asked constantly whether to disclose AI usage in content. My position is pragmatic rather than ideological. Google has said they do not require disclosure of AI involvement in content creation. Readers have mixed feelings about it. Some do not care. Some have strong negative reactions.
I do not disclose AI involvement on a per-article basis because I think it creates a misleading impression. If I write an article using AI as a drafting tool but contribute all the original research, analysis, and expertise myself, labeling it "AI-generated" implies a degree of automation that does not reflect the actual process. It would be like a novelist disclosing that they used spell check.
What I am transparent about, including right now in this article, is my general process. I use AI tools in my content workflow. I also bring original research, personal experience, and editorial judgment to everything I publish. The content is mine. The tool is AI. I am comfortable with that, and my content performance suggests Google is comfortable with it too.
Your mileage may vary on disclosure. If you are in a regulated industry, if your audience has strong feelings about AI, or if your content claims to represent personal experience that was actually AI-generated, the calculus changes. Use judgment. The goal is honest, useful content that serves your readers, regardless of what tools you used to create it.