Does AI-Generated Content Rank as Well as Human-Written Content in Google and AI Search?

What was the study methodology?

The study analyzed 31,493 keywords across 14 industries from September 2025 through March 2026. Each keyword's top 20 Google results were scraped, yielding 629,860 URLs, which were then tested against a validated AI-content detection pipeline combining three classifiers.

Detection accuracy was established by benchmarking against a labeled corpus of 2,000 pages with known authorship (1,000 confirmed AI-generated, 1,000 confirmed human-written). The ensemble classifier achieved the following error rates:

To reduce false-positive contamination, any page flagged by only one of the three classifiers was manually reviewed by two human raters. Inter-rater agreement was 91.3%. Pages with split decisions were excluded from the dataset, removing 3.1% of total URLs.

How much of Google's results are AI-generated?

AI-generated content accounts for 14% of pages in the top 20 Google results across all 31,493 keywords studied. That figure drops to 9.1% when narrowed to the top 10, and to 4.8% in the top 3 positions. Google's ranking systems demonstrably favor human content at every tier of the results page.

Breaking the data down by content type reveals significant variation. Product pages had the lowest AI-content share at 6.2%, while generic informational content reached 22.4%. Industry-level differences were equally stark: finance and healthcare pages showed 7.8% AI content (likely due to YMYL scrutiny), while marketing and technology topics hit 19.6%.

Query intent also matters. Navigational queries contained 3.1% AI content. Transactional queries sat at 8.9%. Informational queries — the category where AI content creation is cheapest — reached 21.3%. This pattern suggests that AI content concentrates wherever the cost of production was historically highest relative to revenue potential.

How much do AI engines cite AI-generated content?

AI search engines cite AI-generated content at a higher rate than Google ranks it — 18% across platforms — but still substantially underweight it relative to its share of total published content. Since November 2024, AI-generated pages have constituted an estimated 37% of new content published on the indexed web.

Platform-level citation data, based on 500 queries tested per platform over 90 days:

Perplexity's higher AI-content citation rate correlates with its retrieval architecture: it indexes and surfaces a broader set of recent sources, including forums and UGC sites where AI content proliferates. ChatGPT's search mode applies stricter source filtering, resulting in a lower AI-content citation rate despite using Bing's index.

Why does AI content underperform despite volume?

AI content underperforms because it fails on three measurable dimensions: entity coverage, structural depth, and E-E-A-T signals. In the dataset, AI-generated pages averaged 1.2 named entities per 1,000 words compared to 4.7 for human-written pages — a 74% deficit in the specificity that search engines use to establish topical relevance.

Structural depth analysis shows that AI content averages 847 words per page with 3.2 H2 headings, while the human content ranking in positions 1-5 averages 2,140 words with 7.8 H2 headings. The AI-generated pages also show 62% less use of original data, citations to primary sources, or embedded media — all signals that correlate with higher E-E-A-T ratings in Google's quality rater guidelines.

The E-E-A-T gap is compounded by a lack of author entity signals. Only 11% of AI-generated pages included a byline linked to a verifiable author profile, compared to 68% of human-written pages in the top 10. For a detailed breakdown of how E-E-A-T functions in AI search contexts, see the SCALEBASE analysis of E-E-A-T in AI search.

• Entity density: AI content averages 1.2 named entities per 1,000 words vs. 4.7 for human content
• Word count: AI pages average 847 words vs. 2,140 for top-5 human pages
• Original data usage: 62% lower in AI content
• Author bylines linked to verifiable profiles: 11% (AI) vs. 68% (human)
• Schema markup adoption: 23% (AI) vs. 54% (human)

What does this mean for content strategy in 2026?

The data points to one conclusion: publishing volume without quality signals does not translate into rankings or citations. AI content is being produced at roughly 2.6x the rate of human content in the informational category, yet it captures only 14% of Google positions and 18% of AI citations. The efficiency argument for pure AI content collapses under these conversion rates.

Organizations producing content at scale should treat AI as a drafting tool, not a publishing pipeline. The top-performing content in this dataset was human-edited, entity-rich, and structurally deep. Sites that combine AI-assisted drafting with human expertise, original data, and proper topical authority building outperformed both pure-AI and pure-human approaches by 1.4x in average ranking position.

Actionable takeaways from the data: invest in entity signals (named experts, original research, cited sources), maintain structural depth (2,000+ words with clear H2 hierarchies for informational content), and audit existing AI-generated content for the quality gaps identified above. SCALEBASE clients who applied this framework to legacy AI content saw an average 23-position improvement within 60 days.

For implementation support on content auditing and quality optimization, see SCALEBASE SEO services.

What are the limitations of this study?

Four limitations constrain the generalizability of these findings. First, AI-content detection is imperfect: the 4.2% false positive rate means approximately 2,645 of the 629,860 URLs may be misclassified, and the 8.7% false negative rate means some AI content was counted as human. These error rates are within industry norms but introduce noise.

Second, the sample skews toward English-language, US-centric queries. Results may differ for non-English markets where AI content adoption rates and detection accuracy vary. Third, the 7-month study window (September 2025 to March 2026) captures a period of rapid change in both AI content production tools and search engine algorithms. Rankings observed in October 2025 may not hold by April 2026.

Fourth, the study does not distinguish between fully AI-generated content and AI-assisted content with substantial human editing. The detection classifiers flag content based on statistical patterns in the text, which means heavily edited AI drafts may register as human-written. This likely underestimates the true share of AI-involved content in search results.