GEO explained: how to rank inside ChatGPT, Perplexity and Google AI Overviews

What GEO actually is

A year ago, "rank on Google" meant winning a blue link. Today, a meaningful share of high-intent queries never reach a search results page at all — they're answered inside ChatGPT, Perplexity, Google's AI Overviews, Claude, Gemini, or You.com. The answer is composed on the fly, and a small number of sources get cited in it.

Generative Engine Optimisation is the work of becoming one of those cited sources. It's not a hack and it's not a new keyword tool. It's an editorial and structural discipline: making sure the models know what you do, trust you to say it, and find passages clean enough to lift.

Why classical SEO is necessary but no longer sufficient

Every major answer engine grounds its responses on a public corpus that overlaps heavily with what Google indexes. So yes — if you can't be crawled, you can't be cited. Technical SEO remains the price of entry.

But ranking #1 on Google is not the same as being cited by an LLM. We routinely see brands that own the SERP for a query and yet are nowhere in the AI answer for the same prompt. The reverse is also true: small brands with weak link profiles getting quoted in ChatGPT because their content is unusually well-structured for extraction.

The shift is from "rank the page" to "earn the passage."

The three levers that actually move citation share

After working with a dozen brands across SaaS, DTC, and B2B services over the last 12 months, we keep coming back to the same three:

1. Entity strength

LLMs build an internal representation of your brand. If that representation is thin or ambiguous, you'll be confused with a competitor or omitted entirely. Investments that compound:

• A complete, accurate Wikidata entry with sameAs links to your domain, LinkedIn, GitHub, Crunchbase, and any other authoritative profile.
• A consistent Organization JSON-LD block on every page, with founder, foundingDate, sameAs, and description.
• An AboutPage that reads like a Wikipedia article — third-person, factual, dense with proper nouns.

2. Source-worthy reference content

Models prefer to quote pages that read like references. That means:

• Original data — surveys, benchmarks, internal numbers you can publish.
• Definitions: short, dictionary-style passages near the top of an article.
• Lists, comparison tables, and step-by-step procedures the model can lift as a unit.
• Bylines from real, credentialed humans with their own entity footprint.

The AI answer engine isn't looking for the best page. It's looking for the page it can quote with the least risk of being wrong.

3. Schema engineered for extraction

Article, HowTo, FAQ, Product, Review, and Organization — implemented properly across your templates — are how you tell the parsing layer what each passage means. Mistakes here are silent: nothing breaks, but you simply don't get pulled.

A quick GEO audit checklist

Run this against your most commercially important page this week:

If a page fails three or more of these, it's not citation-ready — and no amount of link building will fix it.

How we measure GEO

Citation share, not ranking. For each priority prompt:

1. Run it against ChatGPT, Perplexity, Google AIO, Claude, Gemini, and You.com.
2. Capture which domains are cited, in what position, with what sentiment.
3. Repeat weekly. Track drift.

Tools like Profound, Peec.ai, and Otterly make this easier; for clients with high prompt volume we build a custom tracker on top of the engines' own APIs.

What to do this week

1. Pick your top 10 commercial prompts (not keywords — the actual questions a buyer would ask an LLM).
2. Run them through three engines and record which competitors get cited.
3. Score your top-cited competitor's page against the checklist above.
4. Pick the two interventions you're most behind on. Ship them this sprint.

GEO is still early enough that disciplined execution beats clever tactics. The brands that take it seriously now are buying citation share at clearance prices.