Home · AI Search · AI Citations
AI citations are the new distribution: when ChatGPT, Gemini, AI Overviews, or Perplexity answer your buyers, the sources they cite win the customer. Crayonmind helps your brand become one of those sources - on every engine that matters in Indonesia.
No borrowed logos. No invented numbers. Proof gets built in public.
Every engine answers differently. A citation strategy that treats them as one engine loses on all of them.
ChatGPT leans encyclopedic and filters hard for human substance. Perplexity leans on communities. Google's AI Overviews favour structured, ranked, listed content. The overlap between what they cite is small - Profound measured roughly one domain in ten shared between ChatGPT and Perplexity. So the work is per engine, and so is the report.
The definition
An AI citation is a source reference inside an AI-generated answer - the link or brand name the engine shows when it used your content to respond. It carries the click and something bigger: the trust transfer. The engine just vouched for you, in front of your buyer, at the exact moment of the question.
Citations sit in the middle of a measurable ladder: mentioned means your name appears in the answer, cited means your content was used as a source, recommended means the engine tells the buyer to choose you. Citation share - how often answers in your category cite you - is the KPI this whole page is about.
A citation is worth more than the click it carries. It is the engine vouching for you, at the moment of the question.
Most AI answers end without a click at all - the answer absorbed the journey. When a click does happen, it arrives pre-sold: Semrush's clickstream work found AI-referred visitors converting better than search visitors, because they show up already recommended. Citation share is how you count both effects - the influence without the click, and the click that closes.
The engines
Grounded via Bing's index, leans encyclopedic, filters for human substance - 82% of its cited articles were human-written (Graphite, 2025).
Favours structured and listed content - 63% of citations point at list-format pages, and most cited pages sit outside the top ten (Ahrefs, 2026).
Fans one question out into many, retrieves passages with their neighbours - and it speaks Bahasa Indonesia, since September 2025.
Google's assistant surface; entity clarity and consistent facts about your brand decide what it repeats.
Cites generously and leans on communities and discussions - being talked about matters as much as publishing.
Microsoft's surface, grounded in Bing - schema and clean structure count double here.
Engine habits shift with every model release - which is the argument for tracked panels over one-time audits, and for reading the logs instead of the headlines.
The vocabulary
A source reference inside an AI-generated answer - the link or brand name shown when the engine used your content to respond.
The percentage of tracked answers in your category citing you, per engine. The headline KPI: it measures being used, never just being findable.
Your name appearing in answer text without a source credit. Weaker than a citation, still measurable - and off-site mentions are how engines corroborate you exist.
The engine telling the buyer to choose you. The top rung, tracked as its own share - because cited-but-never-recommended is a real and fixable position.
The fixed set of real buyer questions checked continuously across engines. Fixed on purpose: a moving panel can prove anything, and therefore proves nothing.
The evidence
more generative visibility for pages adding statistics, quotations, and cited sources - the original GEO experiment.
Princeton GEO study, KDD 2024more AI citations earned by original research than by commentary. Publishing your own data is the strongest magnet there is.
Kevin Indig, 2025half of AI-cited content was published or updated within the last quarter - engines weight freshness hard.
Ahrefs, 2025domains overlap between ChatGPT and Perplexity citations. One-engine strategies miss most of the surface.
Profound, 2025Citation work starts from a panel of real buyer questions in your category - in Bahasa, in English, phrased the way people actually ask. The panel gets checked continuously across every engine, so "are we cited" stops being an anecdote and becomes a time series.
Month zero gets written down before any work ships. Movement you can see is the entire point.
Monthly, per engine, in plain sentences: citation share, recommendation share, which answers moved, which didn't, and the next month's work. Bad months get reported first - with the fix attached, because a KPI you only hear about when it rises is marketing, and this is measurement.
The mechanics
Engines split pages into passages and match them against the question's meaning - including sub-questions the buyer never typed. Sections that answer one thing completely, first sentence first, are the retrievable unit.
Statistics, quotations, and cited sources measurably lift the odds of being used - the Princeton experiment put it at 25-40%. Tables and numbered lists get lifted most. Original data outperforms everything.
Engines cross-check who you are against directories, press, communities, and profiles. Brand mentions now correlate with citations far more strongly than backlinks - being talked about is a citation strategy.
The levers
Existing pages restructured so each section opens with a liftable claim - the cheapest wins usually live here.
Owning the "what is X" layer of your category - the pages engines reach for first when a question needs grounding.
Data nobody else publishes, stated plainly and dated. The strongest single magnet the research has measured.
Directories, press, communities, video - the corroboration layer engines check before they trust a source.
Quarterly reviews with visible dates - half of cited content is under a quarter old, and stale pages age out of answers.
The same facts about you everywhere machines look. Contradictions read as risk, and engines route around risk.
Ninety days in, citation work either has a rising line to show - or an honest explanation and a changed plan.
Month one writes the baseline down, per engine, however it looks. Month two ships retrieval fixes and the first content built for the panel. Month three reads the movement: which answers changed, which engines moved first, what earned the change. The panel stays fixed the whole way - that is what makes the line real instead of curated.
How it starts
The audit: your citation share today, per engine, with the transcripts to prove it.
weeks 1-2Engines must read you before they can cite you - rendering, structure, entities.
from the gap mapRetrieval-ready content plus mentions where each engine actually looks.
scoped per categoryCitation share monthly, movement and misses, next moves in plain sentences.
stop any monthIn Indonesia, the citation sets are still forming. First movers get written into the answers.
AI Mode has answered in Bahasa Indonesia since September 2025, AI Overviews sit on everyday Indonesian queries, and ChatGPT is the country's most-used AI app - yet almost nobody here tracks or works their citations deliberately. The brands that become citable in this first wave anchor the answers later entrants must displace. That window is the reason this page exists now.
Nobody controls what an engine cites - anyone promising a citation is selling you a hope. What we control: the panel, the work the evidence supports, and a monthly number you can hold us to. That is the deal, stated upfront.
Questions, answered
A source reference inside an AI-generated answer - the link or brand name the engine shows when it used your content to respond. It carries the click and the trust transfer: the engine vouched for you at the moment of the question.
A mention is your name in the answer text. A citation is your content credited as a source. A recommendation is the engine telling the buyer to choose you. A ladder - and each rung is separately measurable.
No - engines decide their own sources, and anyone guaranteeing placement is guessing with your money. We baseline, do the work the evidence supports, and report the movement monthly. You judge us on the time series.
Google's surfaces first - AI Overviews and AI Mode run in Bahasa Indonesia on top of the dominant search engine - with ChatGPT right behind as the country's most-used AI application. Each tracked separately, because each cites differently.
AI citations are the outcome. GEO is the methodology that earns them. This page exists because buyers ask for the outcome by name - the work underneath is one discipline.
Nearby