AI-generated content and E-E-A-T: nine months of observations.

Google's March 2024 core update targeted "scaled content abuse" explicitly. The September 2024 update refined it. By mid-2025 we have nine months of post-update data across 14 client domains that ship AI-assisted content of various kinds. Here is the pattern.

Setup

All 14 domains use LLM assistance in content production. The gradient runs from "LLM drafts, human edits heavily" at one end to "LLM generates, minimal human review" at the other. We categorised each domain's content operation on a 1 to 5 scale (1 = minimal AI, 5 = near-fully AI) and tracked organic search traffic from March 2024 baseline through June 2025.

Findings

Scale 1-2 (light AI assistance): No distinguishable traffic impact. Ranked and performed like non-AI content. Topic expertise of the human editor dominated.

Scale 3 (balanced): Mixed. Roughly half the domains grew organic traffic, half declined. The delta correlated strongly with publishing velocity. Teams that used AI to ship more of what they were already good at grew. Teams that used AI to fill topical gaps they had no expertise in declined.

Scale 4-5 (heavy AI): All four domains in this bucket lost 40 to 80% of organic traffic between March 2024 and June 2025. Two of the four had recoveries attempts (content audit, removal of thin pages, re-write with human depth). Recovery was slow: 6+ months to stabilise, not to fully recover.

What Google appears to be detecting

Based on public Google guidance plus our forensics:

• Topical expertise signals: does the author or brand have demonstrable depth? Wikipedia-style patches of knowledge without original data or perspective get demoted.

• Entity consistency: pages that name, link to, and quote real experts, research, and primary sources perform better.

• User behaviour: dwell time, return rate, and subsequent branded search all appear to matter.

What we are telling clients

1. Use AI for what it is good at: drafting, restructuring, summarising, first-pass editing. Not for generating original expertise you do not have.

2. Original data, real names, and citations are the biggest quality signal we can observe. Produce more of these.

3. Thin category pages that exist to rank are the category most at risk. Audit and prune.

Sources: internal tracking across 14 domains, March 2024 to June 2025; Google's public core-update documentation; Search Engine Land post-update coverage.