Does earned media really drive 82-89% of AI citations?

That statistic comes from two methodologically valid studies: Muck Rack's What Is AI Reading? report which analyzed over 1 million citations across a general prompt set, and a University of Toronto academic study testing four specific verticals (automotive, consumer electronics, software, and local services). Both studies are accurate within the categories they measured. For those categories — consumer electronics, automotive, software, and consumer products — earned media does dominate. For B2B industrial verticals like manufacturing, HVAC, industrial equipment, precision agriculture, and industrial IoT, a separate 547-citation study by Tampa Web Technologies found the opposite: owned content accounts for 48.3% of citations versus 12.4% for earned editorial. Both findings are correct. The determining factor is which vertical you operate in.

Why does earned media dominate AI citations in consumer categories but not industrial?

Consumer categories have thick independent review infrastructure. When someone asks AI about the best laptop, electric car, or SaaS platform, the web contains thousands of expert reviews, analyst articles, and comparison pieces from outlets like CNET, Wirecutter, The Verge, G2, Capterra, and TrustRadius. AI engines have abundant earned media to cite because the content exists at volume. B2B industrial categories have thin or nonexistent earned media infrastructure. There is no Wirecutter for commercial HVAC, no CNET for CNC spindle repair. Trade publications exist but represent only 6.5% of B2B industrial AI citations because most trade editorial is not built for AI extraction. When AI engines need to answer industrial questions, they have no choice but to pull from manufacturer-owned documentation.

What verticals did the Muck Rack and University of Toronto studies actually cover?

Muck Rack's methodology statement says the prompt set covered many industries and topics with some naming specific companies. The prompts reflect the campaigns run by Muck Rack's PR software customers, which skew heavily toward consumer, entertainment, finance, healthcare, and general business categories. The University of Toronto study is more explicit: it tested four specific verticals — automotive, consumer electronics, software products, and local services. Neither study specifically segmented B2B industrial categories like manufacturing, HVAC, industrial equipment, precision agriculture, mining, or industrial IoT. Those verticals were simply not in the sample.

What AEO strategy works for B2B industrial brands?

For B2B industrial brands, the primary AEO lever is owned content built for AI extraction — deep technical documentation, product pages with comprehensive schema markup, factually dense specifications, and high Page Structure Score architecture. Earned media is a supplementary channel, not primary. Trade publication placement helps only when the articles themselves are structured for AI extraction (factual density, schema markup, clean URL structure). The brands winning AEO in industrial categories are the ones investing in technical documentation on their own domains, not chasing press coverage that does not exist or does not get cited at meaningful volume.

Where can I read the original studies?

The Muck Rack What Is AI Reading? report is published at muckrack.com/blog/2025/08/13/what-is-ai-reading/ with a December 2025 update adding Perplexity data. The University of Toronto academic paper is available at arxiv.org/pdf/2509.08919. Commentary and analysis of the University of Toronto findings by Michael Brito, Global Head of Data + Intelligence at Zeno Group, is available at britopian.com/research/earned-media-dominates-ai-search. Both studies are worth reading in full — the methodology sections clarify which categories their conclusions apply to.

Does this mean Muck Rack and University of Toronto got it wrong?

No. Both studies are methodologically sound within the verticals they tested. The Muck Rack research reflects PR industry prompt patterns. The University of Toronto research explicitly names its four test verticals. Both conclusions are correct for those categories. The issue is how the findings have been generalized on LinkedIn and across the AEO discourse into universal laws of AI citation. They are not universal. They are specific findings about specific verticals. Any AEO strategy that treats consumer-category research as applicable to industrial categories is optimizing for the wrong surface. The correct framing is that AEO strategy must be vertical-specific.

AEO Research

The “Earned Media Drives AI Citations” Finding Is Real — But Not for B2B Industrial

Two widely-cited studies conclude that 82-89% of AI citations come from earned media. A 547-citation study we ran across 16 B2B industrial verticals found the opposite. Both conclusions are correct. Here is why the verticals you operate in decide which is true for you.

The stat nobody is questioning

If you read LinkedIn posts about AEO (Answer Engine Optimization) or GEO (Generative Engine Optimization), you have seen some version of this claim repeated as settled fact: 82 to 89% of AI citations come from earned media. Get more press coverage. Focus on third-party publishers. Vendor blogs are dead.

The claim has two real sources worth reading directly.

First, Muck Rack’s What Is AI Reading? report (July 2025, updated December 2025). Their Generative Pulse team analyzed over 1 million citations from ChatGPT, Gemini, Claude, and Perplexity. They found 94% of citations came from non-paid sources and 82% specifically from earned media.

Second, a University of Toronto academic study published in September 2025, available on arxiv.org/pdf/2509.08919. The researchers tested ChatGPT, Perplexity, and Gemini against Google across four verticals. In software specifically, US AI engines returned 72.7% earned, 26.7% brand, and negligible social content. In consumer electronics, 92.1% earned in the US.

Both studies are methodologically sound. Both conclusions hold up under review. But there is a critical detail in their methodology sections that the LinkedIn discourse keeps skipping.

What each study actually measured

Methodology matters more than headlines. Read the methodology sections of both studies carefully and a pattern emerges immediately.

Muck Rack “What Is AI Reading?”

Muck Rack’s own methodology statement says the prompt set “covered many industries and topics, with some naming specific companies and others not.” The study is built to reflect PR industry prompt mix — which is heavily consumer, entertainment, finance, healthcare, and general business. They do not specifically segment B2B industrial verticals in the public summary.

Who the study serves: PR and communications teams. Muck Rack is a PR software company. Their prompts reflect the campaigns their customers run. That is not a flaw; it is their mission. But it shapes what the data represents.

University of Toronto ktau.ai study

The paper is even more explicit. It tests four verticals specifically: automotive, consumer electronics, software products, and local services. Each vertical gets its own percentages. The “earned media dominates” conclusion is true within those four.

What is not in the sample: manufacturing, industrial equipment, HVAC, roofing, industrial IoT, precision agriculture, mining automation, industrial energy, healthcare devices, data center infrastructure. None of these verticals were tested.

This matters because the LinkedIn discourse treats these findings as universal laws of AI citation. They are not. They are conclusions about specific categories: consumer products, software, and general business topics. The verticals that drive your pipeline if you sell to factories, municipalities, fleet operators, hospitals, or industrial contractors were not tested.

What we measured — the B2B industrial gap

Tampa Web Technologies ran a 16-vertical, 550-citation study across ChatGPT, Perplexity, and Gemini, specifically targeting the categories the existing research did not cover: B2B industrial, manufacturing, and trade services.

The verticals: industrial robotics, industrial equipment manufacturing, HVAC building automation, roofing contractors, industrial energy, industrial IoT, healthcare digital, data center cooling, precision agriculture, mining automation, utility vehicles, work boots and industrial footwear, automation and robotics. Plus a consumer-proximity control (athletic footwear).

48.3% Owned content share of citations

12.4% Earned editorial share of citations

16 B2B industrial verticals analyzed

550 Total AI citations examined

In B2B industrial categories, brand-owned content outperformed earned editorial by nearly 4 to 1. That is the reverse of what both Muck Rack and U of T found in their sample categories.

This is not a contradiction. It is the predictable result of studying different verticals with different publication ecosystems.

Why the difference makes perfect sense

The split between the consumer finding and the B2B industrial finding is not a puzzle. It reflects how buyers in different categories actually research products, and how the publication ecosystems around those categories are structured.

Consumer categories have thick independent-review infrastructure

When someone asks AI about the best laptop, the best electric car, or the best accounting software, the web has a massive backlog of expert reviews, analyst content, and comparison articles. CNET, Wirecutter, The Verge, TechRadar, G2, TrustRadius, Capterra — thousands of articles covering every product in depth. AI engines have abundant earned media to cite because the content exists at volume.

B2B industrial categories do not

When someone asks AI about the best HVAC building automation system, the best shot blast machine, the best industrial spindle, or the best precision agriculture software — the earned media layer is thin or nonexistent. There is no Wirecutter for commercial HVAC. There is no CNET for CNC spindle repair. Trade publications exist, but as we documented separately, trade publications are only 6.5% of B2B industrial AI citations because most trade editorial is not built for AI extraction.

With no earned media to cite, AI engines pull from owned content

If you ask Perplexity “what is the most reliable HVAC building automation platform,” and the earned media landscape is thin, Perplexity has to pull from somewhere. That somewhere is the manufacturers themselves — Carrier.com, Trane.com, Johnson Controls, Schneider Electric. Because that is where the documentation actually lives.

The consumer research conclusion says: earned media dominates. That is true in consumer. The B2B industrial research conclusion says: owned content dominates. That is true in B2B industrial. Both are correct. Both apply exclusively to the verticals they studied.

What this means for your AEO strategy

Before you invest a single dollar in earned media outreach for AEO purposes, figure out which bucket your category falls into.

Consumer electronics, automotive, software, apps, SaaS

Earned media placement — trade pubs, review sites, analyst coverage

Owned content for entity verification and technical documentation

Local services (auto repair, dental, roofing, legal, etc.)

Structured data + machine-readable service details + authoritative review presence

Traditional local SEO as a floor, not a ceiling

B2B industrial, manufacturing, HVAC, trades

Owned content — deep technical documentation, product pages with schema, high-PSS architecture

Selective trade pub placement where factual density is strong

Healthcare (clinical software)

Mixed — owned content plus earned coverage in clinical and health-tech publications

Academic and peer-reviewed content where available

The lesson is not that Muck Rack or U of T got it wrong. The lesson is that AEO strategy must be vertical-specific. Applying consumer-category playbooks to industrial categories is the fastest way to spend marketing dollars on levers that do not move your AI citation share.

Methodology and sources

Tampa Web Technologies 550-citation study

Queries were run across ChatGPT, Perplexity, and Gemini between late 2025 and early 2026. Each citation was classified by ownership type (owned, independent, earned editorial, third-party, paid) and scored against a 5-factor Page Structure Score (PSS) rubric measuring answer extraction, content formatting, technical readability, schema markup, and topical alignment. Sixteen B2B industrial verticals were included. Raw data is documented and available on request.

Sources cited in this article

Muck Rack — What Is AI Reading?
Research by Matt Dzugan and Linda Zebian. Available at muckrack.com/blog/2025/08/13/what-is-ai-reading/. Updated December 2025 with Perplexity data added and refined findings. Published via Generative Pulse.

University of Toronto study (ktau.ai)
Academic paper published September 2025. Available on arxiv.org at arxiv.org/pdf/2509.08919. Tests ChatGPT, Perplexity, and Gemini against Google across automotive, consumer electronics, software, and local services verticals in US and Canadian markets.

Commentary and analysis by Michael Brito (Global Head of Data + Intelligence, Zeno Group): britopian.com/research/earned-media-dominates-ai-search.

B2B industrial AEO audits

Tampa Web Technologies specializes in AEO for industrial, manufacturing, and B2B trade brands — the categories the consumer-focused research does not cover. We audit your real AI citation share, identify the gaps in your owned content architecture, and build the technical foundation for AI citation capture in your specific vertical.

Request a B2B AEO Audit