It Works Until It Doesn’t: AI Content Strategies That Backfire

Author: Lily Ray | Published: 2026-05-13 | Source: article

Summary

Lily Ray, who describes a decade of helping companies recover from Google algorithm updates, monitored more than 220 websites publicly identified (by themselves or by their vendors) as customers of AI content creation, automation, and scaling platforms. Many of those tools now also sell AI-search visibility (GEO / AEO (Getting Recommended by AI)). Her headline: “it works, until it doesn’t.”

Method and limits, in her words. Organic traffic estimates and organic page-count time series from Ahrefs, corroborated against the Sistrix Visibility Index; top-traffic URLs from Ahrefs’ top-pages export; figures as of May 2026. The sites came from the public customer-story pages of over a dozen AI content platforms. For many she narrowed to the subfolder where the AI-assisted content was published. Three disclaimers: these are third-party estimates, not first-party analytics; declines could reflect many factors (Google changes, the operators’ own site changes, competition, brand changes, acquisitions, seasonality, architecture), and she is “not asserting that any AI content tool directly caused any traffic outcome”, only describing a correlation; vendors and domains are deliberately unnamed.

Findings. Across the 220+ sites and subfolders: 54% lost 30% or more of peak organic traffic, 39% lost 50% or more, 22% lost 75% or more. The recurring trajectory: rapid organic page growth over six to twelve months, a traffic peak roughly three to six months after the content peak, then a steep decline that erases most of the gain and frequently drops below the prior baseline within the following year. Most drops came after the vendor case studies were published, which she says makes her wonder whether the case studies themselves contribute. Many brands substantially cut their footprint in 2025 and 2026, “often removing, redirecting, or 410’ing” the very pages featured as success stories; the case studies often remain live while the pages do not. She borrows Glenn Gabe’s label “Mount AI” for the shape, and says it appears across cybersecurity, travel, marketing, SaaS, healthcare, B2B services, crypto, and consumer goods, and across vendors.

Why, in her opinion. The templates are “highly formulaic and easily repeatable by competitors.” What begins as a genuine attempt at helpful content “ends up being an easily detectable footprint by Google when enough sites are publishing similar pages.” She ties this to the September 2023 helpful content update, the March 2024 core update (Google’s stated aim: reduce unhelpful, unoriginal content by 45%), and the Scaled Content Abuse policy formalized alongside it, all authorship-neutral. She notes some casualties of those updates were publishing original human-written content inside “excessive SEO frameworks.”

Eight recurring templates on declining sites (most use three or four; the most aggressive use all eight, usually in the hundreds or thousands of pages): comparison pages at scale; “what is X” glossaries, including AI-translated multilingual ones; “best X for Y” listicles; self-promotional listicles; competitor-alternatives pages; programmatic location and language scaling; FAQ farms; off-topic content at scale. She is explicit that these page types do work for ranking and for influencing LLM responses, and that scale is what makes them identifiable as spam.

A second pattern: late January 2026. Google confirmed nothing, but at least 40 sites she identified turned negative around January 20, 2026, with declines of 40% to 95% over January to April 2026, concentrated among sites scaling GEO-optimized self-promotional listicles. Impact often hit the whole blog or subfolder, and sometimes the full domain.

Safe use. The tools are not the problem; implementation is. Use them for research, organization, briefs, and pulling in proprietary company data, overseen by experienced SEO professionals, not “set it and forget it.” AI-assisted content should demonstrate E-E-A-T (Experience, Expertise, Authoritativeness, Trust), add original information beyond competing pages (Information Gain), and consider disclosing AI use. She closes with five pre-publish questions.

Key Claims

• Sample: 220+ domains (often a specific subfolder) from 12+ AI content vendors’ public customer stories; Ahrefs plus Sistrix; as of May 2026. (primary)
• 54% / 39% / 22% lost ≥30% / ≥50% / ≥75% of peak organic traffic. (primary; third-party estimates)
• Trajectory: 6–12 months of page growth, traffic peak 3–6 months after the content peak, steep decline within the following year, frequently below the prior baseline. (primary)
• No causal claim: correlation across sites sharing content patterns; many confounders listed. (primary)
• Case studies outlive the pages: brands removed, redirected, or 410’d featured pages in 2025–2026; she suggests this may explain some later traffic recovery. (primary, hedged)
• Footprint explanation: formulaic, competitor-repeatable templates become detectable once enough sites publish them. (her stated belief, not tested by the data)
• Eight templates all rank and earn AI citations; scale is what makes them detectable. (primary opinion)
• Programmatic location scaling is “one of the oldest tricks in the SEO book,” punished for at least 10 years; pages have “very little unique content” and often no real location behind them. (primary opinion)
• FAQ farms add low-quality baggage at scale; Google announced it was deprecating FAQ rich results “just last week” (early May 2026). (primary)
• Late-January 2026 unconfirmed update: 40+ sites, 40–95% declines, centered on self-promotional listicles, spreading to subfolders or whole domains. (primary observation; Google unconfirmed)
• What hurts SEO can hurt AI search, “largely because of RAG.” (primary opinion)
• Recommendation: E-E-A-T, information gain, AI-use transparency, human expert in the loop. (primary opinion)
• Who is still growing: generally brands whose content does not match the eight templates. (primary observation)

Notable quotes

“I am not asserting that any AI content tool directly caused any traffic outcome described in this piece. I am describing a correlation observed across many listed sites that share similar content patterns and organic traffic trajectories.”

Why citable: The limit on everything else in the piece. Quote this whenever the percentages are used.

“I am not saying these page types don’t work for ranking in search and influencing LLM responses. Indeed they do! That’s why so many sites are implementing them.”

Why citable: The honest version of the warning. The risk is scale and sameness, not the page type.

“Could a competitor publish a near-identical version of this page tomorrow using the same prompt?”

Why citable: One of her five pre-publish questions; the most portable test in the piece.

“Is there any first-party data, expertise, or original perspective on this page that isn’t available on the first ten results already ranking for the query?”

Why citable: Her working definition of Information Gain.

“The packaging is new, but the pattern is not.”

Why citable: Links AI-era scaling to the 2023–2024 update cycle in one line.

Connections

Entities mentioned: Lily Ray, Google, Ahrefs, Glenn Gabe (NEW) · Sistrix (corroborating dataset; no page), Amsive (her March 2024 article; no page) Concepts referenced: Scaled Content Abuse, Information Gain, Programmatic SEO, GEO / AEO (Getting Recommended by AI), E-E-A-T (Experience, Expertise, Authoritativeness, Trust), Schema Markup (Structured Data), Duplicate Content (for Service-Area Pages)

Contradictions / Tensions

• Corrects this wiki’s reading of They Lied About AI Content (This Proves It). That page recorded the “footprint” explanation as Caleb Ulku’s own inference. It is Ray’s: she writes that scaled templates become “an easily detectable footprint by Google when enough sites are publishing similar pages.” It is still her belief, not something the data tests.
• Caleb’s relay is accurate on the numbers (54/39/22, 6–12 months, 3–6 months, eight templates, most sites using three or four) and on her disclaimers. His “fewer pages, tighter topics” is a paraphrase of her “quality, originality, and topical focus over scale.” “These tools work when an experienced person is holding them” is his paraphrase; her wording is that the tools “should be used and overseen by experienced SEO professionals.”
• Caleb’s local-SEO layer is not in Ray. She never mentions Google Business Profile, the Core 30, People Also Ask, or local service businesses. Her location-scaling examples are state-by-state, country-by-country, and multilingual pages. The local application is Caleb’s extension.
• Corroborates Caleb on FAQ rich results: Ray, writing 2026-05-13, says Google announced the deprecation “just last week.” Still no official Google source for it in this wiki.
• Consistent with official policy: her description of Scaled Content Abuse matches Spam policies for Google web search. Her E-E-A-T recommendation is compatible with E-E-A-T (Experience, Expertise, Authoritativeness, Trust)’s “not a ranking factor” finding; she recommends demonstrating it, not that it is scored.
• vs. James Dooley / Joy Hawkins on Duplicate Content (for Service-Area Pages): they report copied service-area pages outranking rewrites; Ray reports thin templated location pages among the patterns on declining sites. Not a direct contradiction (different site types, and Ray claims no causation), but it adds weight to the risk side.
• vs. GEO / AEO (Getting Recommended by AI) tactics in this wiki: the single-question, answer-first, schema-at-the-bottom page built for AI extraction is exactly her “FAQ farm.” Pages built to be cited are, in her data, among those that later fall.
• Limits: a sample drawn from vendor success stories is not a random sample of AI-assisted sites; estimates are third-party; she herself lists the confounders; the late-January 2026 update is unconfirmed by Google.

Notes

Her five pre-publish questions, verbatim themes: does the page exist for a customer or for a search engine/LLM; could a competitor publish a near-identical page tomorrow with the same prompt; would you be comfortable if Google, a journalist, or customers saw the full URL list in the subfolder; is the article biased and transparent about it; is there first-party data, expertise, or perspective not already in the top ten results.

Follow-up sources she references but this wiki does not hold: her February 2026 Substack piece on self-promotional listicles (the ~January 21, 2026 drop), her March 2024 Amsive article advising against scaled FAQ pages, Glenn Gabe’s “Mount AI” writing, and Google’s FAQ rich results deprecation announcement (early May 2026).

Charts were not captured in the text extract; claims about chart contents are limited to her captions.