Here's the question almost nobody asks before working on their AI visibility: which sources do LLMs actually trust?

The answer isn't "quality sites" or "authoritative domains". It's more specific — and more actionable. Each AI engine has a documented hierarchy of preferred source types, which varies by sector, by query type, and by language.

Understanding this hierarchy means knowing where to invest your effort. This article breaks it down by engine, with the concrete implications for your strategy.

The general hierarchy: what the data shows

An analysis of 680 million citations across the main AI engines reveals a consistent structure:

Tier 1 — High-authority general media: Wikipedia, major press (BBC, Reuters, Le Monde, Les Échos), government and institutional sites. These sources are cited across all engines and all query types. Getting into them is difficult but highly durable.

Tier 2 — Sector-specific authority sources: professional rankings (Gartner, Forrester, G2, Capterra, Clutch), specialised press in a domain, professional associations, certified directories. These are the most actionable tier for B2B companies — harder to achieve than tier 3, more impactful than tier 4.

Tier 3 — Social and community platforms: LinkedIn, Reddit, YouTube, Stack Overflow, sector forums. These vary significantly by engine — LinkedIn is tier 1 for Copilot, tier 2 for ChatGPT, less important for Claude.

Tier 4 — Brand-owned sources: company websites, corporate blogs. These matter for establishing your entity but are insufficient on their own — LLMs need third-party confirmation.

The key insight: most B2B companies invest almost entirely in tier 4. The AI citation game is won or lost in tiers 2 and 3.

Engine-by-engine breakdown

ChatGPT — training memory + Bing for real-time

ChatGPT draws from two source pools with different characteristics.

From training data (queries without web search):

  • Wikipedia entries — strong entity anchor signal
  • Major press coverage — particularly mentions in articles rather than standalone brand sites
  • Sector rankings published by established media (Gartner Magic Quadrant, Forbes lists, Les Échos rankings)
  • Forum discussions with high engagement (Reddit, specialised forums)

From Bing web search (queries with web search enabled):

  • LinkedIn: 14.3% of ChatGPT Search responses cite LinkedIn — the fastest-growing citation source in 2025-2026
  • Review platforms: G2, Capterra, Clutch — cited heavily for software and service categories
  • News articles: breaking and recent coverage

What this means for you: for ChatGPT, invest in Wikipedia eligibility if possible, secure a position in at least one recognised sector ranking, and maintain an active LinkedIn company presence.

Perplexity — earned media first

Perplexity's citation bias is the most documented and the most distinctive.

A Fullintel-UConn study (February 2026) on 1,702 Perplexity citations:

  • 47% from journalistic sources — newspapers, magazines, sector media
  • 89%+ earned media overall — content published about a brand, not by a brand
  • Reddit and forums: significant presence on queries involving peer recommendations
  • LinkedIn: lower weight than ChatGPT, but growing

Perplexity applies the most aggressive freshness filter of any engine. Content older than 30 days loses visibility rapidly. The implication: you need ongoing press/media presence, not a single article.

What this means for you: Perplexity rewards earned media above everything else. One article in a well-indexed sector publication is worth more than fifty articles on your own blog. Build relationships with sector journalists and publications before optimising anything else.

Gemini — Google ecosystem depth

Gemini draws from Google's index with E-E-A-T as the primary filter.

A Yext analysis (October 2025) on 6.8 million Gemini citations:

  • 52.15% from brand sites — unique among AI engines, Gemini favours your own well-structured pages
  • Google Business Profile: directly integrated, not just referenced
  • YouTube: 34% growth in citations since Gemini 3 (January 2026) — now 18.2% of all citations outside top 100
  • Press articles: cited when they appear in Google's authoritative index for a topic

What this means for you: for Gemini, your own site and Google Business Profile matter more than for other engines — but they must be well-structured with clear E-E-A-T signals. YouTube content is an underexploited lever.

Copilot — Microsoft ecosystem

Copilot's citation logic is the most Microsoft-specific:

  • Bing index is the exclusive source — nothing outside Bing exists for Copilot
  • LinkedIn: highest weighting of any engine (LinkedIn is a Microsoft property)
  • Microsoft AppSource and partner directories: strong signal for software/service companies
  • Bing News: particularly for recent queries
  • G2 and Capterra: cited heavily for software and professional services

What this means for you: Copilot requires a different investment than other engines — specifically Bing indexing and LinkedIn presence. These are often missing for French-speaking B2B companies.

Claude — verifiability above all

Claude's source hierarchy is driven by its Constitutional AI approach to verification:

  • Brave Search top results — 86.7% overlap (Profound, 2025) — Brave is the retrieval engine
  • Primary sources with explicit citations: academic papers, official statistics, institutional reports
  • Specialised sector publications with identifiable authors
  • Professional associations and certification bodies
  • G2, Capterra for software — with author-verified reviews

What this means for you: for Claude, content verifiability is more important than content volume. One article with explicit methodology, dated sources, and an identifiable expert author outperforms ten general articles.

The universal signals — what works across all engines

Despite the differences above, five source types consistently increase citation probability across all AI engines:

1. G2/Capterra/Clutch profiles with recent reviews Every engine cites these review platforms for service and software queries. A complete profile with 10+ reviews from the last 6 months is one of the highest-leverage actions available. Brands present on 4+ review platforms are 2.8 times more likely to appear in ChatGPT responses.

2. LinkedIn company page with regular content LinkedIn has become the single highest-growth citation source across AI engines. Between November 2025 and February 2026, LinkedIn went from 11th to 5th in ChatGPT citations (Profound, 1.4 million citations analysed). For professional queries, LinkedIn is now the most cited domain across the six main AI platforms.

3. Sector ranking inclusions Being listed in a recognised sector ranking — Les Échos rankings, Gartner Magic Quadrant, Forbes lists, regional business rankings — creates training-data-level authority that persists long after the ranking is published. These are the most durable signals available.

4. Wikipedia presence (if eligible) Wikipedia is cited across all engines and remains the strongest single entity anchor. Eligibility requires demonstrable notability — but even indirect Wikipedia references (being mentioned in a Wikipedia article about your sector) create entity signals.

5. Consistent structured data on your own site Organization, Product/Service, FAQPage, Article with author and dateModified — these schemas are read directly by LLMs and anchor your entity across all platforms.

The French-speaking market gap

For Belgian and French B2B companies, there's a structural challenge that global data doesn't capture: the dominant citation sources in LLMs are overwhelmingly English-speaking.

G2, Capterra, Clutch, Forbes, TechCrunch — these are the tier-2 sources LLMs trained primarily on English text know best. French-speaking equivalents (e-marketing.fr, Le Journal du Net, Décision Achats) exist but have lower LLM authority.

The practical implication: a French SME that secures a listing on Clutch (even with an English profile) will often appear more frequently in LLM responses than an equivalent company only listed on French-language directories.

This creates a concrete opportunity: French-speaking B2B companies willing to build a minimal English-language presence on global review platforms gain disproportionate LLM visibility relative to their sector competitors who stay exclusively on French-language sources.

A prioritised action framework

Given the source hierarchy above, here's a prioritised framework for a B2B company starting from a low-visibility position:

Month 1 — Entity anchoring (highest leverage, fastest impact)

  • Complete G2/Capterra/Clutch profile with structured description
  • Optimise LinkedIn company page: 750+ character description, all specialities filled, 2x/week publishing
  • Implement Organization + FAQPage Schema.org
  • Submit sitemap to Bing Webmaster Tools and Brave Webmaster

Month 2-3 — Content authority building

  • Publish 2-3 substantive articles with identified authors, dated sources, methodology
  • Add dateModified to all existing articles
  • Create or update Google Business Profile to 100% completion
  • Identify sector journalists and publications for earned media outreach

Month 3-6 — Earned media acquisition

  • Secure 1-2 expert contributions in sector publications
  • Pursue a position in a sector ranking relevant to your category
  • Build 3+ reviews on each review platform with recent dates
  • Consider Wikipedia eligibility assessment

Sources: Profound analysis of 680 million AI citations (2025-2026), Fullintel-UConn study on Perplexity citation patterns (IPRRC, February 2026), Yext analysis of 6.8 million Gemini citations (October 2025), Profound data on LinkedIn as citation source (1.4 million citations, February 2026).