Home · AI Search · GEO

GEO, run as a channel.

Generative Engine Optimisation is how a brand becomes the answer. Crayonmind helps your expertise get retrieved, cited, and recommended by ChatGPT, Gemini, Google's AI Overviews, and Perplexity - measured per engine, reported monthly, like any channel you already pay for.

No borrowed logos. No invented numbers. Proof gets built in public.

Buyers ask AI before they ask anyone. The brands that win are inside the answer before the question is finished.

Google switched AI Mode on in Bahasa Indonesia in September 2025, and AI Overviews already sit on top of everyday Indonesian searches. Most brands here have no idea what AI says about them - or that it says nothing at all. GEO is the discipline that closes that gap, on purpose, with a number attached.

The definition

GEO, defined

Generative Engine Optimisation (GEO) is the practice of making a brand retrievable, citable, and recommendable by AI answer engines such as ChatGPT, Gemini, Google AI Overviews, and Perplexity. SEO earns a ranking on a results page. GEO earns a place inside the answer itself.

The mechanics decide the method. Answer engines fetch your pages, break them into passages, and pull the passages that best match each buyer's question - then cite the sources they used. That means GEO work happens at the passage level: clear claims, real definitions, numbers with sources, structure a machine can lift out and quote. Each engine selects sources differently, so the work is measured engine by engine.

25-40%

more generative visibility for pages that add statistics, quotations, and cited sources.

Princeton GEO study, KDD 2024
38%

of pages cited by Google's AI Overviews rank in the top 10 for the query. Citation and ranking are different games.

Ahrefs, 2026
+206%

growth in ChatGPT referral traffic across 2025. The behaviour shift is measured, and it compounds.

Semrush clickstream study
82%

of articles cited by ChatGPT and Perplexity were human-written. Engines filter for substance.

Graphite, 2025

The disambiguation

GEO and SEO, side by side

SEOGEO
The surfaceA results page with ten blue linksThe answer itself - one response, few citations
The unit of workThe page and its keywordsThe passage and its claims
What winsRelevance and authority signalsExtractable substance: definitions, numbers with sources, structure
The metricRankings and clicksCitation share and recommendation share, per engine
The relationshipShared foundations. Google's AI Overviews are a GEO surface sitting on SEO ground - the same work strengthens both, and neither replaces the other.

You may also meet the name AEO - answer engine optimisation. It is the same discipline under a different label. Buyers say AI, practitioners say LLM; we say GEO and get on with the work.

The mechanics

How answer engines choose their sources

Engines read passages, never pages.

An answer engine splits your page into passages and judges each one alone. A section that only makes sense with the rest of the page around it dies in retrieval - which is why GEO writing puts one complete idea in every section, with the claim stated in the first sentence, no pronouns pointing backwards.

Retrieval is semantic.

Engines match meaning, never exact strings - and they quietly fan one question out into many hidden sub-questions: comparisons, alternatives, prices, timelines. The page that covers a topic's sub-intents as separate, clearly headed blocks gets found by the whole fan-out; the page built around one keyword gets found once.

Citation selection rewards substance.

The original Princeton GEO experiment measured it: adding statistics, quotations, and cited sources lifted generative visibility 25-40%. Lists and tables get lifted into answers far more often than prose walls. Substance is a ranking factor here in a way style never was.

Every engine has habits.

ChatGPT leans encyclopedic and filters for human-written substance. Perplexity leans on communities. AI Overviews favour structured, ranked, listed content - and mostly cite pages outside the top ten. One strategy per engine is the work; one report per engine is the proof.

We don't start from keywords. We start from the problem your customer is trying to solve, and build the content universe around it.

Keywords describe how people used to type. Problems describe why they ask. Every engagement starts from a tracked panel of real buyer questions in your category - the ones typed into ChatGPT at 11pm - and builds outward until your brand is the substance engines find when those questions come up. The full method lives in How we think - coming with launch.

The vocabulary

The terms your monthly report uses

Citation share

The percentage of tracked answers in your category that cite your brand as a source, measured per engine. The headline KPI of GEO - the AI-era equivalent of ranking position, except it measures being used, never just being listed.

Recommendation share

The percentage of tracked answers that go beyond citing you and tell the buyer to choose you. The top rung of the ladder, and the number that correlates with revenue.

Prompt panel

The fixed set of real buyer questions we track for your category - in Bahasa and English, phrased the way people actually ask. Fixed on purpose: a panel that changes every month can prove anything, and therefore proves nothing.

Share of answer

How much of a generated answer talks about you, once you appear in it. Two brands can both be cited while one owns three sentences and the other owns a clause.

Content universe

The full set of pages built around one customer problem - definitions, comparisons, processes, evidence - built outward until engines meet your substance wherever the problem is asked about. Our methodology's core unit, and always yours to keep.

What we build

Three layers, one engine

01 · Retrieval architecture

Your expertise, readable by machines

Most AI crawlers read raw HTML once and leave - they run no JavaScript. If your substance hides behind scripts, tabs, or images, it does not exist to them. We rebuild the foundations so engines can take you apart cleanly:

  • Server-rendered content - every claim in the initial HTML
  • Passage-level structure - one idea per section, liftable in isolation
  • Entity and schema layer - who you are, stated consistently, everywhere
02 · Content universe

Answers to the questions buyers actually ask

Around your tracked panel of buyer questions we build the pages engines love to quote: real definitions, honest comparisons, numbers with named sources, process pages that read like receipts. Written in the language your market searches in, by people who own every claim - AI helps produce, humans decide what is true.

This is where substance compounds: every answered question widens the universe, and every widened universe makes the next citation more likely.

03 · Measurement per engine

Citation share, reported monthly

Your panel of real buyer questions gets checked continuously across ChatGPT, Gemini, AI Overviews, AI Mode, and Perplexity. The monthly report reads like a channel report, because it is one:

  • Citation share - how often answers use you, per engine
  • Recommendation share - how often answers tell buyers to choose you
  • Movement and misses - what changed, what did not, and why
  • Next moves - the following month's work, in plain sentences

The ladder

Visible. Cited. Recommended.

Being findable is table stakes. Cited means the engine used you to answer. Recommended means the engine tells the buyer to choose you. Each step is measurable - so we measure it.

visible cited recommended

How it starts

From audit to running channel

01

Diagnose

The AI Readiness Audit: where you stand per engine, with receipts. Fixed fee, findings yours either way.

weeks 1-2
02

Fix retrieval

Architecture first: rendering, structure, entities, schema. Engines must be able to read you before they can cite you.

from the gap map
03

Build the universe

The first content wave around your buyer-question panel - written, reviewed, shipped.

scoped per category
04

Baseline the panel

Citation tracking goes live across engines. Month zero gets written down, so movement is provable.

before the run starts
05

Run monthly

Operate, widen, report. If a number does not move, you hear it from us first - with the next move attached.

stop any month

The first 90 days decide whether GEO is real at your company or another dashboard nobody opens.

Month one ends with the baseline: your citation share per engine, written down, ugly or not. Month two ships the retrieval fixes and the first content wave - the pages built around your buyers' actual questions, in the language they ask them. Month three delivers the first real channel report: movement against baseline, per engine, with the misses stated first and the next moves attached. From there it compounds monthly - every answered question widens the universe, every widened universe makes the next citation more likely. And if by then the numbers say your category resists GEO, the report says that too, in plain sentences. A channel you cannot falsify is a channel you cannot trust.

Pricing

Three stages, three shapes

DIAGNOSE

Fixed fee

The audit is a one-off diagnostic with a fixed scope and a fixed price, quoted in writing before we start. The findings are yours whether or not we ever work together again.

BUILD

Scoped project

Architecture and the first content wave, priced from the audit's gap map - with a definition of done you can hold us to.

RUN

Monthly

Operating the channel: widening the universe, tracking the panel, reporting the numbers. Month to month - stop any time and keep the engine.

Every number lands in writing before work starts. A 20-minute scoping call puts your price on one page - no surprises, no anchoring games.

A fit, honestly

GEO is for some brands. Here is the line.

A good fit

  • Your buyers research on ChatGPT, Gemini, or AI Overviews before they ever reach your site
  • You have real expertise or a real product - GEO amplifies substance, it cannot replace it
  • You want a number every month, never a vibe
  • Your category gets asked about in Bahasa, English, or both - we build for the language your market uses

A poor fit

  • You need results this month - engines refresh on their own clock, and honest GEO respects that
  • There is nothing true underneath - no substance to make retrievable
  • You want rankings alone - that conversation is AI SEO, and we are happy to have it

Said plainly

Four claims we will keep

GEO adds to SEO.

AI Overviews sit on search ground. The same substance strengthens both surfaces; nothing here undoes your SEO.

You own everything.

Content, schema, dashboards, findings - yours from day one, portable if you leave.

No lock-in.

The run stage is monthly. Stop any month and the engine keeps working for you.

Bad months get reported first.

If a number does not move, you hear it from us before you find it yourself - with the fix attached.

Proof, built in public

Live GEO experiments and Indonesia AI-search benchmark reports will land on this site as they happen - real pages, real citations, real logs. Until then, judge us by how we think: ask us anything about your category and we will show you, live.

What we need from you is small, and non-negotiable: an owner, an hour a week, and the truth.

One person on your side who can answer questions and unblock access. One hour a week while the engine gets built - your expertise is the raw material, and nobody can substitute for it. And honesty about your category: what you are genuinely good at, where competitors beat you, which claims you can stand behind. GEO amplifies what is true; it has nothing to work with otherwise.

The market

GEO in Indonesia is early. That is the point.

The Bahasa Indonesia answer layer is forming right now. Google's AI Mode has spoken Bahasa since September 2025, AI Overviews sit on everyday Indonesian queries, and ChatGPT is among the country's most-used apps - while almost no Indonesian brand does deliberate GEO work.

Early matters mechanically, never just rhetorically: the answers being generated about your category today are anchored by whoever is citable today. Brands that enter the citation set in the first wave become the precedent later entrants have to displace. And because buyers here ask in colloquial Bahasa, in English, and in a mix of both, we build and track both languages as native pairs - never translations.

Our own stake in the ground: Crayonmind publishes Indonesia AI-search benchmarks - the numbers nobody here publishes - as public reports. The same method we sell is the method we use on ourselves, in the open.

Questions, answered

What buyers ask us about GEO

What is GEO and how is it different from SEO?

Generative Engine Optimisation makes a brand retrievable, citable, and recommendable by AI answer engines. SEO earns a ranking on a results page; GEO earns a place inside the answer itself. They share foundations - clean architecture, real authority - and the same work strengthens both.

What does GEO cost?

Fixed, and in writing before we start. The audit is a one-off fixed fee, the build is a scoped project, the run is monthly. A 20-minute scoping call puts your number on one page - and it stays that number.

How long until my brand shows up in AI answers?

Each engine refreshes on its own clock, so we refuse to promise a date. What we do instead: baseline your citation share in week one, then report movement monthly against the tracked panel - so you see it move from the first report, engine by engine.

Which engines do you cover?

ChatGPT, Google Gemini, Google AI Overviews and AI Mode, Perplexity, and Microsoft Copilot - tracked separately, because each one selects and cites sources differently.

Does AI-generated content work for GEO?

Engines filter for substance: Graphite's 2025 study found 82% of articles cited by ChatGPT and Perplexity were human-written. We use AI in production and keep humans owning every claim that ships - which is also how we would want to be cited.

Why start with the audit?

Because the gap map decides everything after it. The audit establishes where you stand per engine, ranks what is missing by impact, and sequences the plan - a fixed-fee diagnostic whose findings are yours either way.

Nearby

Where to go from here

The answers about your category
are being written today.

Start with the audit Start a conversation