How to get found by Google and recommended by ChatGPT

What is GEO?

Generative engine optimization (GEO) is the practice of making your business visible to AI engines, so that when someone asks ChatGPT who should build their company's AI agent, or asks Perplexity for a roofing contractor, your name comes up. Search engine optimization (SEO) earns you a ranking on a results page. GEO earns you a mention inside a generated answer, which is a much shorter list and gets chosen differently.

The two work together. Every major AI engine pulls from a search index before it writes anything: ChatGPT search leans on Bing, and Google's AI Overviews use Google's own index. If a page can't rank, it can't be retrieved, and if it can't be retrieved it can't be cited. GEO sits on top of sound SEO rather than replacing it.

What changed between SEO and GEO

The fundamentals carry over: crawlable pages, fast loads, semantic HTML, real content. What changes is who's reading and what wins.

• The reader is a model, not a scanner. Humans skim headlines and bounce. A language model reads the whole page and pulls out claims, so copy that's all mood and no facts gives it nothing to quote.
• Answers beat keywords. Ranking pages target query strings. Cited pages answer a question plainly in the first sentence under a question-shaped heading. The GEO research published so far consistently finds that direct answers, statistics, and quotable specifics beat marketing copy.
• Your entity matters more than your page. An AI engine recommending a business wants corroboration: the same name, services, and story on your site, your LinkedIn, your GitHub, and the directories it learned from. One nice website with no footprint anywhere else reads as unverifiable.
• There's no position eleven. Page two of Google still gets a trickle of clicks. A generated answer names two or three options and stops, so visibility is much closer to all or nothing.

What actually gets you cited

1. Answer-shaped content

Write pages that resolve one question per section, with the direct answer in the first sentence and the supporting detail after it. Journalists call this the inverted pyramid. Include the specifics an engine can lift: numbers, names, dates, concrete steps. "We build AI agents" is unquotable. "An AI agent engagement typically runs four to eight weeks from spec to production" is a sentence a model can hand to a user.

2. Structured data

JSON-LD structured data (schema.org) tells engines what your pages are instead of making them guess: your organization type, founders, services, FAQs, articles. It powers Google rich results directly, and it acts as the machine-readable statement of your entity that everything else corroborates. This page ships Article and FAQPage schema, and the site ships ProfessionalService schema. View source and you'll find all of it.

3. Off-site corroboration

Most of GEO happens away from your website. Models and their retrieval layers trust what multiple sources agree on, so the goal is a consistent entity everywhere your business appears: the same name and description on LinkedIn, GitHub, industry directories, and any profile that ranks for your name. For local businesses this extends to Google Business Profile and review sites. Mentions in places models demonstrably read, like Reddit, Hacker News, and established industry publications, carry extra weight.

4. Let the crawlers in

AI engines fetch pages with their own user agents: GPTBot and OAI-SearchBot from OpenAI, ClaudeBot from Anthropic, PerplexityBot, and Google-Extended. Blanket bot blocking, aggressive firewall rules, or content that only renders through client-side JavaScript can quietly remove you from every AI answer. Check that your key pages return full HTML to a plain fetch, and register with both Google Search Console and Bing Webmaster Tools. Bing's index feeds ChatGPT search and Copilot, so it matters more than its market share suggests.

5. llms.txt, with caveats

llms.txt is a proposed convention: a markdown file at your site root that summarizes who you are and links your key pages, written for language models the way robots.txt is written for crawlers. No engine has committed to reading it, and we won't pretend otherwise. But it takes fifteen minutes, some crawlers already request it, and it's the one place you fully control the summary a model might read first. Ours is at /llms.txt.

6. The boring fundamentals

Fast loads, one h1 per page, headings in order, alt text, canonical URLs, a sitemap, no duplicate hosts. None of it is glamorous, but it decides whether anything above ever gets read. Most sites we audit fail here first.

The checklist we run

Every site we build or audit goes through the same pass:

1. Title, meta description, and canonical URL on every page. One host, with no www/apex or staging duplicates.
2. Organization-level JSON-LD (business type, services, founders, sameAs profile links) plus page-level Article, FAQPage, or Product schema where it applies.
3. Sitemap submitted to Google Search Console and Bing Webmaster Tools.
4. robots.txt allowing GPTBot, ClaudeBot, PerplexityBot, and OAI-SearchBot. Key pages verified to serve full HTML without JavaScript.
5. At least one page per service that answers the question a real customer would ask, direct answer first.
6. Entity audit: name, description, and services consistent across the site, LinkedIn, GitHub, and every directory that ranks for the brand name.
7. llms.txt at the root, kept in sync with the site.
8. Core Web Vitals passing on mobile. Images sized and lazy-loaded.

Then comes the part nobody can shortcut: publishing real material on a steady schedule. Citations and backlinks follow work worth citing. Everything above just makes sure they can find it.