A2Basics

We built A2Basics, a searchable directory of the full Amazon Basics catalog, to answer a question most affiliate sites never ask: when the visitor isn’t a person but an AI assistant, does your site still get cited?

A2Basics tracks roughly 2,500 Amazon Basics products with live pricing, specifications, and direct buy links. On the surface it’s a classic SEO play. Underneath, it’s a testbed for GEO — Generative Engine Optimization — the discipline of making a site legible, trustworthy, and quotable to the answer engines (ChatGPT, Perplexity, Google’s AI Overviews) that increasingly sit between your content and your reader.

SEO and GEO: one foundation, a different reader

Traditional SEO optimizes for a ranked list of blue links: match the query, earn the click. GEO optimizes for a synthesized answer: the engine reads your page, decides whether it trusts it, and either quotes it — with attribution and a link — or never mentions you at all. There’s no page-two consolation prize. You’re in the answer, or you’re invisible.

The reassuring part is that both disciplines share a foundation. Clean canonical URLs, fast pages, structured data, a crawlable sitemap — Google has recommended these for years, and they’re exactly what a language model needs to parse a page with confidence. Nail the SEO fundamentals and you’ve done half the GEO work already.

The divergence is in what the reader does with the page. Google ranks it; an LLM has to be able to quote it. That shifts the emphasis from keywords toward verifiable facts, stable identifiers, and machine-readable structure.

What we built for the machines

• Structured data everywhere. Every product page ships schema.org JSON-LD — Product, Offer, Brand, BreadcrumbList, and FAQPage — so an engine extracts price, ASIN, and brand instead of guessing from prose.
• An llms.txt manifest. Following the llmstxt.org convention, the site publishes a concise, machine-first map of itself — including an explicit “common queries this site answers” section and a freshness timestamp — so an assistant can orient in a single fetch instead of crawling 2,500 pages.
• Stable, descriptive canonicals. URLs follow /products/{category}/{slug}-{asin}/, pairing a human-readable slug with the ASIN — the exact identifier an assistant quotes back on a “what’s the ASIN for…” prompt.
• Fresh, timestamped facts. Prices and availability re-sync from the Amazon Creators API on every scheduled build, and each figure carries the UTC time it was fetched — the provenance an engine wants before it will repeat a number.
• IndexNow on every deploy. A post-build step pings IndexNow so Bing and its partners see changes in minutes rather than waiting on a crawl.
• A machine-readable feed. Beyond HTML, the site emits a structured JSON feed of the catalog for direct consumption.

Measuring what actually works

You can’t optimize what you can’t see, so analytics splits referrals three ways — Google vs. ChatGPT vs. direct — and tracks a view_item → view_on_amazon_click funnel per category. That turns “is GEO working?” from a hunch into a number: which products assistants surface, and whether that traffic converts as well as search does.

To pressure-test it, we asked ChatGPT to audit the site’s own AI-discoverability. Its report reads like a GEO scorecard — crediting the machine-friendly identifiers and fresh timestamps, and flagging precisely where the site is still a catalog rather than a recommendation.

The frontier: from catalog to decision support

The next phase is the harder, more valuable half of GEO: being not just parseable but preferable. That means human-written TL;DRs, “best for” statements, and comparison tables — the quotable judgments an assistant reaches for when a user asks “which one should I actually get?” A page an engine can read is table stakes. A page an engine wants to cite is the whole game.