How to get cited by ChatGPT: the depth-optimization playbook
ChatGPT doesn't sweep wide: it picks few sources and leans hard on each. To get cited, you must pull off two things at once — be selected among a small set of pages, and be a source so dense in evidence that the model can't ignore it.
The short answer
ChatGPT cites few and deep. It references on average 6.88 sources per query — less than half Perplexity's 16.35 — but each source carries heavy weight: its mean influence score is 0.271, roughly four to five times that of Perplexity or Google. In other words, when ChatGPT cites you, you genuinely shape the answer.
The strategic consequence is twofold, and that's what makes ChatGPT demanding: you must be both selected AND absorbed. Getting into the small set of retained sources isn't enough; you also have to be dense enough in evidence that the model leans on you rather than its six other candidates. This is the inverse of the Perplexity strategy, where getting into the wide pool is the whole game.
What makes ChatGPT different from the other engines?
ChatGPT stands out through two measured behaviors.
| Behavior | ChatGPT | Perplexity | Google AI Overviews |
|---|---|---|---|
| Sources cited per query | 6.88 (sparse) | 16.35 (broad) | 12.06 |
| Influence per source | 0.271 (deep, 4–5×) | 0.065 | 0.058 |
| Reacts to framing | Expert role | "Cite your sources" | "Cite your sources" + English |
| Under multi-constraint | Compresses to 3.4 citations | Rises to 17.7 | — |
Source: 2026 breadth/depth study (602 prompts, 21,143 citations, 18,151 fetched pages).
Two particularities to remember.
1. It compresses under complexity. Faced with a multi-constraint query, ChatGPT reduces its source count — it drops to 3.4 citations, where Perplexity opens up to 17.7. It synthesizes more internally instead of retrieving broadly. The harder the question, the tighter the bottleneck, and the more you must be an unavoidable source to appear in it.
2. It reacts to expert framing, not to source requests. This is a counterintuitive point: where Google and Perplexity cite more when the user explicitly asks "cite your sources," ChatGPT cites more under an expert-role framing. Content that presents itself as an authoritative reference in its field has a specific advantage on this engine.
Lever #1: getting absorbed through evidence genres
Once selected, what determines your influence on the answer is the evidence genre your page contains. It's the most powerful factor identified in the research — and it's largely common to all engines, but it takes on its full meaning on ChatGPT, where each absorbed source weighs enormously.
Influence gain by evidence genre present in the page:
| Evidence genre | Influence gain |
|---|---|
| Code / technical examples | +76.9% |
| Numbers / statistics | +61.6% |
| Definitions | +57.3% |
| Comparisons | +55.3% |
| How-to / tutorials | +41.2% |
The most influential semantic roles are definition (0.153) and comparison (0.152); passages that merely reference a source without asserting anything are the weakest (0.053).
What to do concretely:
- Open your pages with a clear definition of the concept covered. It's both a high-gain genre and the most influential semantic role.
- Embed structured comparisons — "X vs Y" tables, side-by-side criteria. This very article contains several; that's no accident.
- Densify with sourced numbers. A quantified, attributed claim is worth far more than a generality.
- For technical content, show code and procedures. It's the highest-yield genre (+76.9%).
Lever #2: semantic alignment
After evidence genre, the second signal most correlated with influence is the semantic similarity between your content and the generated answer (r = 0.36), just behind the page-query relevance score (r = 0.43). Both beat raw page length.
The lesson: longer text doesn't help on its own. What helps is answering precisely, in the words and concepts the answer will mobilize. Dense, aligned content beats long, diffuse content.
This is consistent with the citation-failure taxonomy: most relevant uncited pages fail on semantic alignment, not writing quality. The right topic isn't enough — you need the right angle, the right entities, the right level of precision. (See our article on diagnostic GEO.)
Lever #3: authority framing
Since ChatGPT cites more under an expert-role framing, the positioning of your content matters.
- Sign and credit. Identified authors, visible qualifications, displayed expertise. E-E-A-T signals (Experience, Expertise, Authoritativeness, Trustworthiness) are recognized by the model.
- Adopt an assertive tone. Cut hedging language ("maybe," "it seems"). Backed assertions beat cautious phrasings.
- Lean on earned media. Like all generative engines, ChatGPT strongly favors third-party content — press, Wikipedia, expert reviews — over brand-owned content. An authority presence beyond your own domain strengthens your credibility in the model's eyes. This is what Traaker's Earned Media axis measures.
What helps less than you think
Wrapping content in an FAQ isn't enough. The question-and-answer format is a neutral-to-negative signal for absorption (−5.74% in the breadth/depth study) when there's no evidence density behind it. Form doesn't replace substance: a hollow FAQ doesn't get cited.
Raw length isn't a lever. The correlations show it clearly — semantic alignment and evidence genre beat word count. Lengthening a page without densifying its evidence is wasted effort.
Over-optimizing for breadth is a misread on ChatGPT. Given its low citation count and high influence per source, the "be everywhere" strategy that wins on Perplexity is secondary here. Save broad coverage for Perplexity — see our Perplexity guide. On ChatGPT, depth decides.
The ChatGPT action plan in 6 points
- Open with a definition. High-gain genre (+57.3%) and most influential semantic role.
- Densify with evidence: sourced numbers, structured comparisons, code and procedures for technical content.
- Align semantically — match each page to the expected answer; precision before length.
- Frame with authority: credited authors, assertive tone, visible E-E-A-T signals.
- Build your earned media to strengthen credibility beyond your own domain.
- Target complex queries by being unavoidable — ChatGPT compresses to 3.4 sources under constraint, so selection there is even more contested.
Measure before you optimize
On ChatGPT, the trap is the inverse of Perplexity's: being found is useless if you're not dense enough to be absorbed. Traaker measures your visibility engine by engine: which queries ChatGPT cites you on, how much your pages actually shape its answers, and where your evidence genres are too weak to carry weight.
The ChatGPT strategy fits in one sentence: be selected, be dense, be unavoidable. Depth first, breadth second.
Methodological note. The figures cited come from observational studies (correlations, not proven causal relationships). They are the best public evidence available to date — read them as strong tendencies, not guarantees. Engine behavior evolves; measure yours continuously.