TL;DR — 30 seconds
- Lily Ray (Amsive, 10+ years in SEO recovery) published an analysis of 220+ sites listed as success stories by 12+ AI content platforms.
- 54% lost 30%+ of peak organic traffic. 22% lost 75%+.
- Typical pattern: rapid page volume growth → traffic peak 3-6 months later → steep decline often below baseline.
- 8 high-risk templates identified — comparison pages, glossaries, self-promotional listicles, and FAQ farms are most exposed.
- Solution: use AI for research and briefs, never as your sole writer. E-E-A-T + information gain + human review = real protection.
The promise was simple: publish hundreds of AI pages and watch traffic explode. For many sites, it worked — for a few months. Then Google updated its algorithms, and a large portion of that traffic collapsed.
Lily Ray, Senior Director of Search Insights at Amsive and one of the most-followed SEO analysts in the world, has published a study on this exact phenomenon. She examined real site portfolios — not theoretical cases — and the results carry important lessons for any business that invested in AI content in 2024-2025.
Primary source: Lily Ray, "It Works Until It Doesn't: AI Content Strategies That Backfire", May 2026. Secondary source: Google Spam Policies — Scaled Content Abuse.
Methodology: 220 Sites, 12 Platforms, 2 Years of Data
Ray didn't analyze random sites. She targeted sites publicly listed as success stories by more than 12 AI content creation platforms — the same "our clients testify" pages these tools display to convert new customers.
For each of the 220+ domains, she cross-referenced:
- Ahrefs — organic traffic estimates and indexed page count time series
- Sistrix Visibility Index — to corroborate trends and detect relative declines
The analysis focused on the specific subfolders where AI content was published (not the entire domain), making results more precise. Coverage spans 2024 through May 2026.
Methodological note: These are third-party estimates (Ahrefs, Sistrix), not first-party Search Console data. Declines may reflect multiple simultaneous factors. No domain or vendor names are explicitly cited. Ray herself notes: correlation observed, not causation proven.
Key Numbers: 54%, 39%, 22%
The chronological pattern is striking and repeated across most analyzed sites:
- Phase 1 — Growth: intensive publication of AI pages over 6-12 months
- Phase 2 — Peak: organic traffic climbs for 3-6 months after the content peak
- Phase 3 — Collapse: gradual decline, often falling below the original baseline
What Ray highlights with precision: most traffic drops occurred AFTER the success stories were published. Platforms featured these testimonials during the growth phase, before the decline began. Several sites have since removed, redirected, or 410'd the very pages featured in those testimonials.
The 8 High-Risk AI Content Templates
The study identifies 8 content types that concentrate the majority of traffic declines. If your strategy relies on any of them at scale, this is a warning signal.
Scaled comparison pages
Format: /product-A-vs-product-B. Generated en masse to cover all competitor combinations. Low differentiation between pages, easily replicable by any competitor.
Programmatic "What is X" glossaries
Format: /what-is-[term]. Often duplicated and auto-translated into multiple languages without real adaptation. Google treats these as thin content if they offer nothing beyond a generic definition.
"Best X for Y" listicles at scale
Format: /best-[tool]-for-[use-case]. Extremely common in SaaS and affiliate. Problem: without genuine first-hand experience with the reviewed tools, they fail on the "Experience" criterion of E-E-A-T.
Self-promotional listicles
The publisher ranks itself first among its competitors — without genuine comparative data. A manifest editorial bias that Google detects as a bad-faith signal.
"Alternatives to [competitor]" pages
One page per named competitor. Generates short-term traffic on branded searches, but the intent is transparent to both Google and the user.
Programmatic geographic/language variations
Same content duplicated across 50 cities or 20 countries with a simple name replacement. If unique content per page doesn't exceed 50%, this is scaled content abuse under Google's official policies.
FAQ farms (one question per URL)
Each question on a topic gets its own URL with FAQ schema markup, to maximize featured snippets. Google has already removed FAQ Rich Results from SERPs — a devaluation signal for this format.
Off-topic content at scale
Publishing content on topics unrelated to core business activity to capture generic traffic. Disconnected from the domain's topical authority, this content is flagged as filler.
Does your blog use some of these templates? Cicero Studio runs a full GEO + SEO audit to identify high-risk pages and high-value content opportunities.
Free audit →The Unconfirmed January 2026 Update
Ray identifies a period of turbulence around January 20, 2026. Google officially confirmed no core update at this date — but data from 40+ sites tells the same story: traffic declines of 40% to 95% over a few weeks, persisting through April 2026.
Content types most affected during this window:
- Self-promotional listicles (where the publisher positions itself as the best market solution)
- GEO-optimized content at scale — designed to be cited by generative AI but lacking genuine expertise
This turbulence fits into a longer timeline: the Google Helpful Content Update of September 2023 targeted content "created for search engines rather than people." The March 2024 Core Update formalized the reduction of "45% of low-quality content." The Scaled Content Abuse policy is now explicitly written into Google's official spam guidelines: generating many pages to manipulate rankings is explicitly classified as spam.
As Ray puts it precisely: "The playbooks being sold as 'AI-first SEO' or 'GEO-optimized content at scale' look remarkably similar to the playbooks that got sites flattened by the Helpful Content Update and the March 2024 Core Update. The packaging is new, but the pattern is not."
How to Use AI Content Without SEO Risk
The study does not conclude that AI is unusable. It identifies the boundary between healthy and risky usage.
Where AI excels in content production
- Document research and source synthesis
- Writing structured editorial briefs
- Organizing and clustering raw data
- First drafts to be substantially reworked by a human expert
What creates penalty risk
- Publishing at scale without expert human review
- Creating pages whose only purpose is to exist for search engines
- Reproducing templates any competitor can replicate with the same prompt in 10 minutes
- Concealing editorial bias (self-promotion) behind a "neutral" format
Before publishing each content batch, Ray suggests 5 questions:
- Does a real customer need this page, or only Google?
- Can a competitor replicate this content using the exact same AI prompt?
- Would you be comfortable if Google, a journalist, or a customer saw your full URL list?
- Is the editorial bias transparent to users?
- Does this content contain proprietary data or a perspective absent from the current top-10 results?
Our Analysis: What This Means for Your Strategy
This study confirms what we've observed for several months: AI content volume is not a strategy, it's a calculated risk. A short-term bet that can accelerate a traffic surge, but exposes you to an equally rapid collapse as soon as Google refines its detection.
What this data means concretely for a business or agency:
- Audit your current portfolio before continuing to publish. If you have hundreds of pages matching the 8 templates identified, the risk is real.
- Prioritize depth over volume. One 1,800-word article with verifiable field expertise is worth more than 20 mass-generated 400-word articles. The 2025-2026 Core Updates consistently penalized volume without depth.
- Invest in GEO signals, not just SEO. AI engines (ChatGPT, Perplexity, Gemini) cite content with sourced expertise, proprietary data, and citable structure. Their use of low-quality content is even more limited than Google's.
Ray's study isn't a condemnation of AI content. It's a reminder that AI is a production tool, not an editorial strategy. As Lily Ray herself concludes: "None of this means AI content tools are unusable. The trouble starts when the goal becomes volume — or when the people closest to the content stop reviewing what is going out the door."
What this article doesn't cover
- Names of platforms or domains analyzed (not published by Ray to protect brands)
- Cases where AI content at scale has worked durably — they exist, but don't match the 8 identified templates
- Impact on e-commerce sites (the study focuses on editorial and SaaS sites)
- Actual Google Search Console metrics (Ahrefs/Sistrix estimates only)
Frequently Asked Questions
Lily Ray's (Amsive) study, published in May 2026, analyzed 220+ sites publicly listed as success stories by 12+ AI content platforms. Using Ahrefs and Sistrix data: 54% of sites lost 30%+ of peak organic traffic, 39% lost 50%+, and 22% lost 75%+. The typical pattern: rapid page growth over 6-12 months, traffic peak 3-6 months later, then a steep decline often below baseline. Source: Lily Ray Substack, May 2026.
The 8 high-risk templates: (1) scaled product comparison pages (A vs. B), (2) programmatic "What is X" glossaries (including auto-translations), (3) "Best X for Y" listicles at scale, (4) self-promotional listicles where the publisher ranks itself #1, (5) "Alternatives to [competitor]" landing pages, (6) programmatic location/language variations with minimal unique content, (7) FAQ farms (one question per URL), (8) off-topic content at scale.
Lily Ray identifies a turbulence period around January 20, 2026 — not officially confirmed by Google. 40+ sites experienced traffic declines of 40-95% between January and April 2026. Primary targets appear to be self-promotional listicles and GEO-optimized content at scale without genuine expertise.
Use AI for research, briefs, and data organization — not as your sole writer. Published content must demonstrate E-E-A-T signals (real Experience, Expertise, Authoritativeness, Trustworthiness), provide genuine information gain beyond existing top-10 results, and be reviewed by a human expert before publishing. Transparency about AI use is recommended by Google.
Google does not penalize AI content per se, but enforces its "Scaled Content Abuse" spam policy against sites generating many pages to manipulate rankings — regardless of the tool used. The criterion is value delivered to the user, not whether AI was involved. AI content demonstrating genuine E-E-A-T and information gain is not targeted.
Growth and SEO & GEO content strategy specialist. I founded Cicero Studio to help businesses build durable organic visibility — on Google and in AI-generated answers. Cicero Studio serves clients from €250 to €1,800/month with GEO audits, editorial production, and automated semantic linking.
LinkedIn