AI vs Human Content: 6-Month SERP Tracking Study 2026

The "does AI content rank?" debate has been emotional for two years. By April 2026 the question is empirical. We tracked 200 paired articles — 100 AI-generated, 100 human-written, matched on domain, publish week, internal-link template, topic specificity, and word count band — across 14 publishing domains for six months. The data does not say AI content is dead. It does not say human content is irreplaceable. It says the two perform very differently across the stages of an article's life on a SERP.

AI wins the first lap: faster indexing, higher initial position, cheaper to produce. Human wins the marathon: by month three the ranking trajectories invert; by month six there is a five-position median gap, in human's favor. The SERP-feature picture is harsher — featured snippets, People-Also-Ask answers, and AI Overview citations all skew strongly toward human content, with the AIO gap the most pronounced (4% AI vs 11% human cited).

What follows is the methodology, the headline numbers by stage of life, and a taxonomy of which content types AI handles well and which it handles poorly. The full dataset — pair IDs, domain categories, publish dates, weekly position snapshots, and SERP feature flags — is publicly downloadable as a methodology companion. Where we draw inferences beyond what the raw data supports, we say so.

For most of 2024 and 2025 the AI content question was litigated with anecdote. One agency would publish a case study showing AI content ranking #1 within a week. Another would publish a postmortem on a domain that lost 60% of organic traffic after an AI content sprint. Both were true. Neither was generalizable. The absence of paired controls — same domain, same week, same length, same internal-link structure — meant every claim was confounded by variables nobody bothered to hold constant.

We started this study in October 2025 specifically to remove those confounds. The pairing methodology is simple but expensive: every AI article in the dataset has a human counterpart on the same domain, published within seven days, in the same content category, within the same word-count band, with the same internal-link template applied at publish time. The treatment variable is the author intervention; everything else is controlled. The dataset is small enough (200 pairs) to allow individual-pair audit, large enough that the headline deltas are statistically significant at 95% CI.

We are not claiming this study is definitive. We are claiming it is the most carefully-controlled paired SERP-tracking dataset currently public for the AI-vs-human content question. Methodology disclosed; raw data downloadable; replication welcomed.

The dataset comprises 200 articles — 100 AI-generated, 100 human-written — published between October 1 and December 15, 2025, and tracked daily through April 15, 2026. Each AI article is paired with one human article on the same domain. Domains were selected to span four publisher categories: B2B SaaS company blogs (5 domains), digital agency blogs (3), news / publisher sites (3), and ecommerce / DTC blogs (3). All 14 domains had at least 6 months of indexing history before the study window opened, to avoid new-domain crawl artifacts.

Tracking instrumented daily SERP position (top 100 for the primary target keyword), weekly SERP-feature inventory (featured snippet, PAA inclusion, AI Overview citation), GA4 sessions / engagement metrics, and a backlink scrape via Ahrefs API at week 0, 4, 12, and 24. Position is reported as median across the daily samples within each week to dampen daily volatility. Confidence intervals on aggregate deltas are bootstrap 95% CIs computed across the 100 paired observations.

The first surprise in the data is the indexing-speed advantage. AI articles hit Google's index faster than their human pairs across all 14 domains, with median time-to-first-index of 14 hours for AI versus 26 hours for human. The 1.8× advantage is consistent across publisher types and shows up cleanly in the bootstrap CI (95% CI on the ratio: 1.6×–2.0×). It is the cleanest single result in the study.

The indexing-speed advantage has a short shelf life. By month three the ranking trajectories of the two cohorts cleanly diverge: human articles gain a median 6 positions from their week-1 baseline; AI articles drift down a median 3 positions. The gap opens further through months four, five, and six. By month six, human has a 5-position median advantage — fully reversed from the 4-position AI advantage at week 1.

The trajectory pattern is the most operationally important finding in the study. A two-week post-publish review will show AI outperforming human — that is exactly the time at which most content teams cut their performance review and lock in the AI decision. By the time the trajectories invert at month three, the decision has already shipped to the next quarter's plan. Anyone evaluating an AI content program on a 30-day window will reliably mis-read the data.

What changes between week 1 and month 3? Three things, in decreasing order of estimated importance: backlinks (human articles attract 4× more backlinks in the 6-month window), engagement signals (human articles run lower bounce and longer session duration, both of which Google now uses as ranking inputs through Chrome's anonymized telemetry), and SERP feature placement (human articles capture more snippets and PAA panels, which themselves cause CTR redistribution that reinforces position).

SERP feature capture is the place where the AI-vs-human gap is widest, most consistent, and most actionable. We tracked three feature classes weekly: featured snippets, People-Also-Ask answer panels, and AI Overview source citations. Across all three, human content captures more — and the AI Overview gap is the largest at roughly 3× the citation rate.

One sub-question we wanted to settle: when AI and human articles rank at the same position, do they draw the same click-through rate? The answer is no — AI articles draw 8% lower CTR position-adjusted, and the gap is driven by title quality plus a small but measurable trust effect on AI-detected bylines. Engagement-on-page is a larger gap: bounce rate runs 8 percentage points worse, and session duration is roughly 35% shorter.

The conversion gap is the one number that should change planning for any team running AI content into a measurable funnel. A 1.75× higher conversion rate on the human cohort means that even if AI ranks identically and produces identical session counts, the funnel output is 75% smaller. For B2B SaaS spending $40 cost-per- session-equivalent on paid acquisition, that is the difference between a profitable content channel and a vanity-metrics exercise.

The patterns that hurt AI content most across this dataset are recognizable: em-dash overuse, generic openers, the "delve" / "moreover" / "in today's fast-paced world" cluster, and — more costly than the prose tics — the absence of original data, named sources, or first-party POV. The patterns that help: clear comparison tables, structured H2/H3 hierarchy, explicit numerical depth in claims, and clean schema markup. None of those patterns require human authorship — they require an editor pass.

The aggregate numbers mask large variation by content type. AI performs near-parity with human on data-heavy formats (statistics roundups, comparison/vs articles), drops moderately on procedural content (how-to guides), and falls off a cliff on opinion and review content. The "AI uplift gap" — AI's blended SERP performance as a percentage of its human pair's — ranges from 92% on the strongest format to 38% on the weakest.

One observation worth flagging: the AI-content "good enough" threshold has crept upward through 2025 and 2026 as frontier models improved, but the relative ranking of formats has been stable since at least mid-2024. Statistics and comparison content have always been AI's strongest formats. Opinion and reviews have always been AI's weakest. Better models tightened the gap on the strong formats faster than on the weak formats. The ceiling on opinion/review content is being set by the absence of first-party experience, not by model quality — and no foreseeable model release closes that gap.