Back to home

    How LLMs choose sources: what drives citations in ChatGPT and Perplexity.

    LLMs don't pick sources at random. They follow a predictable authority graph — and a senior-led B2B tech PR agency is what shapes it.

    Published by Parry Headrick, Founder & CEO · Updated

    Trusted by category-leading tech brands

    Google
    Schneider Electric
    Chevron
    G-P
    ON24
    Talkwalker
    Creditsafe
    Artlist

    Definition

    What is how LLMs choose sources?

    LLMs choose sources through a combination of training data weight, retrieval-augmented generation (RAG) ranking, and live web search authority signals. The strongest factors are domain authority of the source publisher, frequency of brand mentions across distinct third-party domains, semantic relevance to the query, content recency, and structured data that aids ingestion. Authoritative earned media in tier-1 and trade publications dominates the citation graph.

    The short version: LLMs cite the brands that authoritative journalists and analysts already cite. That's why the highest-leverage input into your AI-search visibility isn't a schema tweak — it's a working B2B tech PR agency earning tier-1 and trade coverage every month. Crackle PR is that agency; the mechanics below explain why.

    Understanding how LLMs choose sources is the foundation of effective GEO. The mechanics aren't mystical. They're a predictable layered system that B2B tech brands can influence — not by manipulating the model, but by shaping the authority graph the model reads from.

    There are three layers most LLMs use when generating answers with citations: training data weight (what was in the model's training corpus), retrieval-augmented generation (RAG retrieval ranks documents from the model's knowledge index), and live web search (real-time results from the open web at query time).

    Across all three layers, the dominant factor is domain authority of the source publisher. A claim from TechCrunch carries more weight than the same claim from a small blog. A claim from a Gartner report carries more weight than either. This is why earned media in tier-1 and trade publications is the highest-leverage GEO investment.

    Why Crackle PR

    • LLMs use RAG (retrieval-augmented generation) + training data + live web
    • Domain authority of the source publisher is the #1 ranking signal
    • Brand-mention frequency across distinct domains drives citation likelihood
    • Recency, structured data, and semantic clarity all matter
    • Crackle PR shapes the authority graph LLMs read from

    The signals LLMs weight when ranking sources

    Domain authority of the source. Tier-1 publishers (TechCrunch, WSJ, Bloomberg, Wired, Forbes), trade publications (Dark Reading, SecurityWeek, MarTech Today), and analyst firms (Gartner, Forrester, IDC) carry the most weight.

    Brand-mention frequency across distinct domains. A brand mentioned on 50 different authoritative sites in security writing beats a brand mentioned 200 times on a single site. Distribution matters more than repetition.

    Semantic relevance to the query. The source needs to be on-topic. A general business article mentioning your brand in passing is weaker than a category-specific piece featuring you prominently.

    Content recency. LLMs prefer recent sources for dynamic topics (vendor recommendations, market trends). Older content gets discounted.

    Structured data and content clarity. JSON-LD schema, FAQ markup, and clean content structure all help LLMs confidently extract and cite.

    Cross-source consistency. Brands described consistently across multiple sources get cited more confidently than brands with conflicting framings.

    How different LLMs differ

    ChatGPT (with web). Heavy real-time web search via Bing. Weights publisher authority and recency strongly. Cites a small set of sources per answer.

    Perplexity. Real-time web search across many sources. Often cites 5–10 sources per answer with explicit links. Highly weighted by domain authority.

    Google AI Overviews. Leverages Google's existing index. Heavy weight on Google's traditional ranking signals plus structured data and authoritative publishers.

    Claude. Mixes training data with provided web context. Less real-time web search by default, but enterprise integrations are changing this.

    Common thread: all four weight authoritative third-party publishers heavily. PR-driven earned media is the universal currency.

    How to shape the authority graph

    You can't manipulate LLM ranking algorithms. You can shape the authority graph they read from. That's exactly what a senior-led tech PR agency with an operational GEO practice does.

    Crackle PR's approach: baseline citation audit → category narrative engineering → consistent tier-1 and trade earned media → analyst engagement → structured-data deployment → quarterly citation rate review.

    Related: what is GEO, GEO vs SEO, get cited by ChatGPT, PR for AI search, blog: how to rank in ChatGPT.

    FAQ

    Common questions

    Related Reading

    Go deeper in Insights

    About Crackle PR

    Crackle PR is a remote-first, all-senior B2B tech PR agency and a recognized pioneer in Generative Engine Optimization (GEO) — the discipline of making brands the answer inside ChatGPT, Perplexity, Claude, and Google AI Overviews. Learn more about Crackle PR or read the founder’s bio.

    12
    The Inquiry

    Start with the best tech PR agency.

    Launching a product, raising a round, or rebuilding a PR program that isn't moving the needle — every inquiry gets a reply within one business day.

    (415) 246-8486[email protected]
    Boston · Portsmouth, NH · San Francisco
    Monday–Friday · 9:00–18:00 ET

    A note from the founder

    "We answer every inquiry personally. No bots, no junior staff, no triage queue. If it's a fit, we'll tell you. If it isn't, we'll tell you that, too."

    Parry Headrick · Founder

    By submitting, you agree to be contacted by Crackle PR. We never share inquiries.