What is Generative Engine Optimization (GEO)?

Generative Engine Optimization (GEO) is the practice of structuring content so that it is easily parsed, cited, and recommended by AI answer engines like ChatGPT, Perplexity, and Google AI Overviews.

How does the seo-geo-optimizer npm package work?

It is a 14-phase agentic workflow that runs inside AI IDEs (Cursor, Windsurf, Copilot). It auto-detects your environment and installs rules to guide the AI through technical SEO, schema generation, and content formatting.

Why is traditional SEO insufficient for AI search?

Traditional SEO focuses on keyword density to rank as a blue link. AI search engines (LLMs) prioritize clarity, direct answers, and structured data to synthesize responses and cite sources.

Is seo-geo-optimizer an agentic AI or machine learning tool?

It is an agentic AI tool. It takes multi-step actions across 14 phases, modifies your workspace files, and validates its own output, unlike machine learning which only makes predictions.

SEO-GEO Optimizer

TL;DR

·Built a 14-phase agentic npm skill that auto-installs SEO/GEO rules into Cursor, Windsurf, and Copilot.
·Solved AI hallucination: prevents IDEs from giving outdated SEO advice (like optimizing for deprecated FID).
·411 organic downloads with zero marketing; featured on PrimeSkills AI Store.

Overview

SEO-GEO Optimizer is a developer-first library for the AI-search era, built as a 14-phase agentic workflow with human-in-the-loop validation gates rather than a single-pass script. It treats LLMs as first-class crawlers, not accidents.

Problem Statement

SEO best-practice is optimised for click-through; GEO (Generative Engine Optimization) is optimised for citation. Most sites are optimised for neither.

Market Need

I searched for my own portfolio on Perplexity and it returned someone else's work. Not a paraphrased version - a different person entirely. That was the immediate problem. The deeper one: I had no reliable way to make a site legible to something that wasn't a traditional crawler.

Target Users

Indie devs and small teams shipping content-heavy sites - docs, blogs, marketing pages - who want AI citations without hiring an SEO agency.

Research

Spent time with pages that were getting cited by AI engines and pages that weren't, looking for the difference. The pattern that appeared most consistently: cited pages gave the answer in the first paragraph. Not a teaser - the actual answer. The rest of the page could elaborate.

Insights

·AI engines reward clarity, not keyword density. This sounds obvious until you realise most content is optimised for the latter.
·Structured data is the floor now, not the differentiator. What the data says is still the question.
·I still don't fully understand why some pages get cited and others don't, even with identical markup. Answer-density matters, but I haven't proven the mechanism cleanly.

Opportunity

Package the 80% of GEO best-practice as sane defaults, and expose the 20% as opt-in knobs.

Scope

·Auto JSON-LD for Article / FAQ / HowTo / Product
·llms.txt generator
·Answer-block components for React frameworks
·14-phase agentic execution pipeline - a 40-question discovery intake through technical SEO, schema, content strategy, and GEO optimisation, with a human-in-the-loop validation gate between phases so nothing commits without review

Out of Scope

·Full analytics dashboard
·Hosted SaaS tier
·Manual keyword research tools

Assumptions

The bet was that developers would install an npm package rather than read a blog post and implement manually. That's not obvious - the blog-and-copy approach has always won for SEO tooling. I believed the developer-native packaging would convert better for this audience specifically.

Success Criteria

·Target: weekly downloads trending upward within 90 days - not a specific number, but evidence that word was spreading without paid distribution.
·Target: at least one case study of a user site cited by an AI engine within 60 days of launch.
·Kill condition: if install-to-first-citation takes longer than a month for the median user, the defaults aren't doing enough.

Expected Behaviour

npm install → wrap layout → done. First AI citation within a month.

North Star Metric

Sites Cited by AI Engines After Install - not downloads, not stars. Citations are the proof of value.

Supporting Metrics

·Weekly Downloads
·GitHub Issues Closed / Opened
·Time-to-First-Structured-Page

Prioritization

Everything was scored against 'does this show up in a real answer engine result?' Citations were the only currency.

Wireframes

The API surface was the design. Every export was named for the answer it enabled, not the mechanism underneath it.

PRD

One-page RFCs in-repo per major surface. Kept close to the code so they didn't drift from what was actually built.

Prototype

Dogfooded on this portfolio and my own blog before any public release. Iterated the API until it fit a real site in under 30 minutes. If I couldn't use it myself in half an hour, the defaults weren't sane enough.

Final Solution

A single package, three exports, and a 14-phase agentic pipeline underneath - each phase gated by a human-in-the-loop validation step - covering Article, FAQ, and HowTo pages out of the box.

Reflection

The best PM work here was aggressive scope-cutting. I said no to a hosted SaaS tier three times. Each time it would have delayed the first real user shipping by months.

Lessons Learned

·Developer tools are UX products. The API is the interface, and the quickstart is the first impression.
·For emerging categories, ship early and let the docs grow with the use cases. A perfect spec for a category that doesn't exist yet is a waste of time.