Detailed Analysis
A Reddit post from a small Amazon-affiliate site owner has surfaced a practical case study in how Claude Code is being used to build tools for "answer engine optimization" (AEO) — the practice of structuring web content so AI chatbots like ChatGPT quote it directly rather than merely ranking it in search results. The author, running a low-authority buying-guide site with most pages stuck on page two of Google, discovered that ChatGPT had become their top traffic referrer, ahead of Google Search entirely, with Perplexity and Google's AI Overviews rounding out the bulk of visits. Traditional organic search accounted for roughly 5% of traffic. This anecdote, while unverified and from a single small site, illustrates a broader shift being discussed across SEO and content communities: domain authority, the currency that has governed Google rankings for two decades, appears far less predictive of whether an AI system chooses to cite a page.
The core insight the author packaged into a free, MIT-licensed Claude Code skill called "aeo-blocks" is that AI retrieval systems reward structural clarity over site reputation. The skill automates generation of content patterns believed to get quoted by LLMs: front-loaded 40-60 word answer blocks, one question per H2 heading (since retrieval operates on page chunks rather than whole documents), comparison tables, sourced statistics, FAQ sections, matching JSON-LD structured data, and llms.txt files — an emerging informal standard for signaling a site's key pages to AI crawlers. This represents a notable use case for Claude Code beyond traditional software engineering: using an agentic coding tool as a content-engineering assistant that produces both the on-page markup and the reference files needed to make a site machine-legible to AI answer engines.
The more technically significant finding in the post concerns a failure mode specific to LLM-generated content at scale: fabricated links. The author reports that 82% of AI-generated product links in a spot-check were wrong — mismatched products, dead Amazon ASINs, or entirely invented URLs — and that an initial automated verification pass approved all of them anyway, since a single LLM reviewing its own hallucinated output tends to rubber-stamp errors rather than catch them. The fix required a second, independent verification pass explicitly designed to distrust the first pass and re-check every link against the live target page. This is a concrete, real-world instance of a well-documented AI reliability problem — hallucination compounding when a model is used both to generate and to self-verify — and it reinforces why practitioners increasingly recommend adversarial or multi-agent verification pipelines rather than single-pass LLM review, especially for content involving prices, links, or factual claims with commercial consequences.
Broader context matters here: this post is emblematic of a fast-growing niche where creators are building and monetizing tools (the author discloses a paid "verifier agent" pack alongside the free skill) around the premise that generative AI answer engines are reshaping how content gets discovered, sidestepping traditional SEO gatekeepers like domain authority and backlink profiles. Whether or not ChatGPT's citation behavior is as democratized as this single case suggests, the pattern reflects real anxiety and opportunity in the content/SEO industry as AI-mediated search consumption grows. It also underscores a recurring theme in Anthropic's ecosystem: Claude Code is increasingly used not just by professional software engineers but by solo operators and marketers building lightweight automation "skills" — a trend Anthropic has actively encouraged through Claude's extensibility features — while also surfacing the practical guardrails (like independent fact-verification loops) that responsible AI-content workflows now require to avoid shipping hallucinated links or claims at scale.
Read original article →