AI Content vs. Human Content: What the Data Shows

The study

A study of 31,493 keywords examined how AI-generated versus human-written content performs in both Google Search and AI answer engines including ChatGPT and Perplexity. AI content detection was performed using a validated detector with a 4.2% false positive rate and 0.6% false negative rate.

Methodology

The keyword set was sampled across 14 industries: SaaS, financial services, healthcare, legal, real estate, e-commerce, education, travel, hospitality, staffing, professional services, manufacturing, insurance, and wellness. Each industry contributed between 1,800 and 2,900 keywords weighted toward commercial and informational intent. Navigational and branded queries were excluded to avoid skewing results toward domains with inherent brand advantage.

For each keyword, the top 20 Google Search results were scraped, yielding 629,860 unique URLs. Each URL was passed through an ensemble of three AI content detectors — GPTZero, Originality.ai, and a custom transformer-based classifier — with a page flagged as AI-generated only when at least two of three detectors agreed. This consensus approach reduced false positives to 4.2% and false negatives to 0.6%, validated against a labeled corpus of 2,000 pages with confirmed authorship.

For the AI citation component, 500 queries per platform were tested across ChatGPT (browsing mode), Perplexity, and Gemini over a 90-day window from September 2025 to January 2026. Each cited URL was classified using the same detection ensemble. The study window captures the period immediately following the inflection point when AI content volume surpassed human content volume for the first time.

The findings

Despite AI content being published at record volume — surpassing human content output in November 2024 for the first time — it performs poorly in both channels. AI content accounts for only 14% of Google Search top-20 results and just 4.8% of top-3 positions. In AI answer engines, AI-generated pages represent 18% of ChatGPT citations, 21% of Perplexity citations, and 15% of Gemini citations. Human-written content consistently outranks and out-cites AI-generated content across all categories.

The click-through rate gap is equally stark. Pages identified as human-written earned an estimated average CTR of 3.2% from Google Search results, versus 1.4% for AI-generated pages in the same position range. This suggests that even when AI content does rank, users engage with it at lower rates — possibly because thin AI content delivers fewer unique insights, less original data, and weaker author signals in the SERP snippet.

The paradox

More AI content is being created than ever. Less of it wins in search or gets cited by AI. The internet is flooding with thin AI-generated pages, and both Google and LLMs are actively deprioritizing it. Since November 2024, AI tools have produced an estimated 58% of all new indexed content by volume. Yet that 58% captures just 14% of search visibility and 18% of citation share. The ratio of production to performance is collapsing.

Why AI content underperforms

There are four structural reasons AI content consistently loses to human content in both search and citation contexts.

First, AI content lacks original data. Google's ranking systems and LLM retrieval layers both prioritize content that introduces new information — proprietary research, case studies, first-party data, named expert quotes. AI-generated content recombines existing information. It cannot create the primary sources that ranking and citation algorithms reward.

Second, AI content has weak E-E-A-T signals. Experience, Expertise, Authoritativeness, and Trustworthiness are core ranking factors in Google's quality rater guidelines and implicit weighting factors in LLM source selection. AI content typically lacks a named author with verifiable credentials, has no author entity connected to external profiles, and offers no first-person experience or case-specific detail. These are precisely the signals that differentiate a citable source from a commodity page.

Third, AI content converges on the same answers. Because large language models are trained on the same corpus, AI-generated articles on any given topic tend to contain the same points in the same order with the same conclusions. Search engines and AI retrieval systems both have diminishing returns for repetitive content. The 500th article making the same seven points about a topic does not outrank the first — it gets filtered.

Fourth, AI content is structurally shallow. The average ranking AI-generated page in the study was 1,120 words. The average ranking human-written page was 2,340 words. Length alone does not cause rankings, but the gap reflects a deeper problem: AI content tends to cover topics at surface level without the layered analysis, specific examples, and nuanced discussion that both search algorithms and LLM citation logic reward.

What this means for your content strategy

The data makes the strategic implication clear: volume is the wrong strategy. Publishing 50 AI-generated articles per month will generate less search visibility and fewer AI citations than publishing 8 deeply researched, human-written articles with original data, named expertise, and strong structural markup.

This does not mean AI has no role in content production. AI-assisted workflows — where AI handles research synthesis, draft generation, or structural outlining while a human subject matter expert adds original insight, edits for accuracy, and attaches their name and credentials — can produce content that performs competitively. In the dataset, pages identified as AI-assisted but human-edited ranked an average of 6.2 positions higher than fully AI-generated pages.

The key differentiator is not who or what produced the first draft. It is whether the final published page contains original information that does not exist elsewhere, is attributed to a named entity with verifiable expertise, and is structured for both search engine and AI retrieval consumption. Content that meets all three criteria outperforms regardless of production method. Content that meets none of them fails regardless of production volume.

Key data points:

• AI content volume surpassed human content in November 2024, now representing 58% of new indexed content
• AI content accounts for 14% of Google top-20 results and just 4.8% of top-3 positions
• AI content accounts for 18% of LLM citations across ChatGPT, Perplexity, and Gemini
• Human-written pages average 2,340 words vs 1,120 for AI-generated pages in ranking positions
• AI-assisted, human-edited content ranks 6.2 positions higher on average than fully AI-generated content
• Quality, original data, and authority signals determine success — not production volume