Why does AI rely so heavily on third-party sources when describing brands?

Large language models are trained on and retrieve from a large corpus of public web content. The University of Toronto study (Chen et al., September 2025) measured this directly: AI search engines exhibit a systematic, overwhelming bias toward earned media over brand-owned and social content. The bias holds across all tested regions, verticals, and AI systems (ChatGPT, Perplexity, Gemini, Claude). For ChatGPT specifically, the study found 93.5% of citations for well-known brand queries went to earned third-party sources. Branded websites are present in the training corpus but make up a much smaller share of what AI systems trust enough to cite.

Which third-party sources matter most for AI brand representation?

The high-leverage categories are Wikipedia (heavily cited by ChatGPT), Reddit (heavily cited by Perplexity), G2 and other review aggregators (cited across systems for product evaluation queries), Hacker News (technical and SaaS categories), and news outlets (recency and authority signals). Per-source AI weight varies by AI system and by query type, which is why Kodiac scores each source independently.

How is Source Intelligence different from AEO and GEO?

AEO and GEO focus primarily on owned content - making your website and structured data more extractable by AI. Source Intelligence focuses on the third-party sources AI reads to form its view of a brand. The two are complementary. AEO and GEO matter for the content you own; Source Intelligence matters for the 74-92% you don't (University of Toronto, 2025).

Source Intelligence / A Kodiac discipline

Most of what AI says about your brand isn't written by you.

Q: What is Source Intelligence?

Source Intelligence is the discipline of monitoring and operationalising the third-party sources that AI systems read to form their representation of a brand. Reddit, Wikipedia, G2, Hacker News, news, and other public sources account for the overwhelming majority of how AI describes brands. Independent research from the University of Toronto (September 2025) measured AI citing third-party publications 92.1% of the time in consumer electronics queries and 81.9% in automotive. Source Intelligence covers per-source AI weight scoring, sentiment monitoring, and remediation tracking for each source.

Source Intelligence is the discipline of monitoring and influencing the third-party content AI uses to describe your brand. Reddit threads. Wikipedia articles. G2 reviews. Hacker News discussions. News coverage. Most AI visibility tools never look here. Kodiac made it a core layer of Kodiac Audit: part of understanding how AI describes your organisation before you improve the information it relies on.

Sign up for a month See Kodiac Audit

University of Toronto: Chen, Wang, Chen & Koudas, Sept 2025 / "AI Search exhibit a systematic and overwhelming bias towards Earned media over Brand-owned and Social content."

AuditSource intelligence across the systems shaping your brand.

The gap most tools miss

What is Source Intelligence?

Source Intelligence measures and improves the third-party content layer that shapes how AI systems describe your brand. It is the missing layer in single-source AI visibility tools, and the largest single contributor to brand representation in generative AI.

Per-source AI weight scoring

Every source is weighted by how heavily each AI system relies on it when describing your brand, from the URLs Perplexity and Gemini cite to the sources inferred behind ChatGPT and Claude answers. The weights tell you which interventions actually move the needle.

Every source AI cites

Reddit. Wikipedia. Wikidata. G2. Hacker News. ProductHunt. Industry news. Trade press. Partner directories. Forum discussions. Review sites. Every source AI systems cite when describing your brand is collected on each scan and tracked with sentiment alerts.

Measured score uplift per fix

Every fix is measured against a baseline scan and the first scan after the change, with the uplift attributed to the inaccuracy, page or source it addressed and a confidence band on the result. Real numbers, not generic recommendations.

Remediation tracking

Every source you act on moves through not started, in progress, remediated or failed, with mention volume before and after and a 30 to 60 day verification window. You see which interventions worked, not just which were attempted.

92.1%

The number most tools hide

of how AI describes your brand comes from sources you don't own.

Independent research from the University of Toronto (Chen, Wang, Chen, Koudas - September 2025) measured how ChatGPT, Perplexity, Gemini, and Claude source their answers about brands. In consumer electronics, AI cited third-party publications 92.1% of the time. In automotive, 81.9%. In software, 74.2%. The pattern is consistent across regions, languages, and verticals: AI systems exhibit a systematic, overwhelming bias toward earned media over anything brands publish themselves.

Reddit alone accounts for 6.6% of Perplexity citations. Wikipedia accounts for 5% of ChatGPT citations. Most B2B brands have either a Wikipedia stub or no entry at all. Most B2B brands have never audited what's said about them on Reddit.

Audit-only tools tell you the symptom. Source Intelligence tells you the root cause.

Use case

A worked example.

A digital lead at a mid-market B2B SaaS company runs a Kodiac three-layer audit and sees an AI visibility score of 34/100. Audit-only tools would tell her to publish more content. Source Intelligence tells her exactly where the problem actually is.

Layer 01 view: Output

Per-system breakdown shows Perplexity worst at 19/100. ChatGPT 38. Gemini 41. The score has not improved despite 6 months of website optimisation.

Layer 02 view: Input

Website Score 47/100. 10-dimension breakdown shows acceptable structured data and AI-readability. The owned site is not the bottleneck.

Layer 03 view: Source Intelligence

A single Reddit pricing thread is being cited by ChatGPT in 4 of 12 buyer queries. Wikipedia stub is missing. Three out-of-date G2 reviews dominate the snippet. The fix isn't more content - it's source intervention.

The result

The team addresses the Reddit thread (+11 points), creates a proper Wikipedia entry (+9 points), and refreshes G2 (+7 points). Within 6 weeks the AI visibility score moves from 34 to 67. None of those interventions would have been visible to an audit-only tool.

Frequently asked

Source Intelligence: common questions.

How does Source Intelligence differ from social listening?

Social listening tracks brand mentions across social channels for sentiment and engagement. Source Intelligence tracks specifically the sources AI systems cite when generating answers, weighted by how heavily each system relies on each source. They are complementary disciplines with different goals.

Which sources matter most for B2B brands?

For most B2B brands the highest-leverage sources are Wikipedia (heavily cited by ChatGPT and Gemini), G2 and similar review platforms (cited for product comparisons), Reddit (cited by Perplexity and increasingly by ChatGPT for opinion-rich queries), and trade press. The exact ranking depends on category and AI system mix.

Can I improve a Reddit thread that's hurting my AI visibility?

Yes - through legitimate community engagement, not manipulation. The right approach is product team participation, AMA hosting, transparent correction of factual errors, and earning natural mentions through genuine value. Kodiac tracks each intervention and verifies whether the citations change. Manipulating Reddit will damage you. Engaging properly will improve every layer of your AI visibility.

Does my brand need a Wikipedia entry?

For any B2B brand large enough to be evaluated against competitors in AI-mediated buyer journeys, yes. Wikipedia notability requires independent reliable sources, so the prerequisite is press coverage and trade citations. Notability assessment, sourcing and proper editorial process come first; Kodiac then tracks whether AI starts citing the entry.

How long does Source Intelligence take to improve AI visibility?

Per-source interventions typically appear in AI responses within 4–8 weeks once the source is updated and re-indexed. Wikipedia and G2 update fastest. Reddit and Hacker News are influenced over months through community participation. Sustained improvement of overall AI score takes 3–6 months.

Where it leads

Be Seen. Be Understood. Be Asked.

Most organisations begin with the audit. Content gets the information behind the answers approved and kept current. The agent is where it leads.

Be Seen / Kodiac Audit

See how AI describes your organisation and identify what needs attention.

Explore Audit

Be Understood / Kodiac Content

Connect and maintain trusted information for AI to use.

Explore Content

Be Asked / Kodiac Agent

Give customers and their AI someone to ask.

Explore Agent

Visibility is the entry point. The destination is a Brand Agent of your own: one your customers and their AI can ask directly, answering only from information you have approved, remembering with permission, on the AI model you choose.

Explore Kodiac Agent

See it for your brand

See the sources shaping your brand.

Understanding your sources is where the journey starts. Scan your brand across six AI systems. Per-source AI weight. Measured score uplift per fix.

Sign up for a month Book a demo