Key takeaways (detailed version):
- 64% of a brand’s AI citations come from third-party sources, not its owned domain (Stacker, March 2026).
- Earned media distribution produces a median 239% lift in LLM visibility (study on 87 stories and 8 AI platforms).
- Up to 90% of citations that drive brand visibility in LLMs come from earned media (Edelman, 2025).
- The winning budget split combines 40% on-site and 60% earned media once the fundamentals are covered.
Why 64% of your AI citations don’t come from your site
LLMs don’t answer with your site first. They answer with the full set of sources that talk about you, weighted by authority and freshness. When ChatGPT, Perplexity, Gemini or Claude cite your brand, they rely mainly on third-party media, external blog posts, forum discussions or Wikipedia pages, not on your own domain.
The Stacker study published in March 2026, conducted in partnership with Scrunch on 87 stories distributed for 30 brands and measured across 8 AI platforms, precisely quantifies this structural bias. When a brand is cited by LLMs, 64% of those citations point to third-party publisher sources, not to the brand’s own site. In 1 story out of 5, third-party sources are even the sole source used by the AI to talk about the brand.
This figure aligns with Edelman’s 2025 analysis, which documents that up to 90% of citations driving brand visibility in LLMs come from earned media, meaning third-party content that is neither owned nor paid. The operational conclusion is direct: your on-site GEO strategy mechanically plateaus if it isn’t extended by an earned media strategy.
The common mistake SEO teams make in 2026: investing 100% of the GEO budget on their own site (intro optimization, refresh, semantic structure). These efforts remain necessary, but they only capture at most a third of the citation volume. The remaining two-thirds play out off your domain.
What the Stacker 2026 study changes in our understanding of GEO
Stacker conducted the first large-scale study with statistical controls on the impact of earned media distribution on LLMs. Methodology: 87 stories distributed through Stacker’s publisher network, 30 brands represented, 8 AI platforms queried with about 30 prompts per story. Measurement at a fixed post-distribution point, across two dimensions: citations of the brand’s owned domain, and citations of the network’s publisher sources.
The results statistically validate (p < 0.006) what agencies suspected:
- 97% of distributed stories get at least one AI citation, compared to 82% for owned content alone. The gap is narrow but statistically significant.
- Median lift of 239% in AI visibility through earned distribution (vs 325% in a pre-study on 8 stories, demonstrating statistical robustness).
- 5.3 times more likely that the distributed version becomes the sole source of a story’s AI visibility, compared to the owned site.
- Cross-platform coverage goes from 5.4% to 17.9% with earned distribution. Your presence becomes 3 times more uniform across ChatGPT, Perplexity, Gemini and Claude.
The strongest signal beyond these raw numbers concerns challengers. Brands with the smallest initial AI footprint benefit most from earned media. LLMs weight relevance, freshness and specificity more than accumulated domain authority. A fresh, specific story distributed through credible publishers can surpass a much larger historical competitor.
This opening for challengers rarely exists in traditional SEO. It will close as large players invest in distribution, but it remains exploitable in 2026.
The 4 earned media levers that generate AI citations
Not all earned media is equal. LLMs weight sources differently based on authority, freshness and topical relevance. I identify 4 levers that produce the most measurable AI citations, ranked by decreasing impact.
Distribution through publisher networks
Distribution through publisher networks (Stacker, Outbrain, regional networks) means syndicating an original content to third-party media with link and attribution. This is the fastest lever to produce documented AI citations. The model works because it multiplies your content’s presence across domains recognized by LLMs as authoritative.
The Stacker study proves that this lever alone produces a 239% median lift. The entry ticket for an SMB in 2026 ranges from 2,000 to 5,000 euros per quarter depending on volume and publisher tier.
Targeted PR on industry media
Traditional PR keeps strong GEO value, as long as you target industry media rather than mainstream press. An article in Siècle Digital, Frenchweb or a vertical business media carries more weight in LLM citations than an article in a general daily, because models associate the source with a recognized domain of expertise.
Operational rule: each press outreach must contain an original figure, a sharp position, or a documented use case. Generic press releases produce zero AI citations. LLMs extract what is unique and substantial, not what is copy-pasted from corporate boilerplate.
Interviews and expert speaking opportunities
Podcast interviews, expert panels and conference talks produce indirect but durable AI citations. The mechanism: the interview is transcribed, turned into an article, indexed, then used by LLMs to source responses attributing a quote or position to a named expert.
For an SMB or consultant, targeting 5 to 10 speaking opportunities per year on high-authority formats (industry podcasts, professional conferences) produces measurable returns within 6 to 12 months. The expert name associated with the brand creates a trust signal that LLMs preferentially surface.
Original research and studies to distribute
Publishing original research with transparent methodology is the most powerful of the 4 levers, but also the most expensive in production. A figured, dated study with sample and protocol becomes a source LLMs cite directly and durably, far beyond the lifecycle of a classic article.
The Stacker study itself is the example: a few months after publication, it is already cited by dozens of articles and regularly surfaces in ChatGPT and Perplexity responses on GEO queries. An SMB can produce a mini-study on 50 to 200 cases with a budget of 3,000 to 8,000 euros, then distribute it through the 3 other levers to maximize exposure.
How to split your budget between on-site and earned media
The budget split depends on your site’s GEO maturity. As long as the on-site fundamentals are not covered (optimized intros, monthly refresh, semantic structure, measured AI Share of Voice), investing in earned media produces limited ROI. LLMs need a correctly structured owned source to validate brand authority.
Once the on-site base is consolidated, the optimal split shifts strongly in favor of earned media. Here is the matrix I apply with clients based on maturity stage.
| GEO maturity stage | On-site budget | Earned media budget | Priority |
|---|---|---|---|
| No GEO structure | 100% | 0% | Intro optimization, refresh, schema |
| Fundamentals covered | 60% | 40% | First industry PR actions |
| AI visibility measured | 40% | 60% | Publisher network distribution |
| Thought leadership target | 30% | 70% | Original study + media tour |
Pitfall to avoid: switching to earned media too early. A brand spending 10,000 euros on PR while their site has unoptimized intros and no refresh in 18 months wastes two-thirds of the budget. The AI citations generated by earned media point to a site LLMs cannot validate, which limits conversion.
How to measure the real impact of your earned media on LLMs
Measuring earned media GEO requires a baseline/post-distribution method. Without a baseline, you have no way to prove ROI. The method I apply:
- Baseline before action: capture of AI Share of Voice across 30 to 50 strategic prompts through a tool like Cockpyt AI, with Coverage Breadth measurement (cross-platform presence across ChatGPT, Perplexity, Gemini, Claude).
- Distribution or PR action: execution of the earned media action (distributed story, press article, podcast, published study).
- Measurement at 30 and 60 days: rerun of the same prompt panel, comparison with baseline, identification of prompts where citation appears for the first time.
- Coverage Breadth delta: measurement of cross-platform coverage progression. A successful action moves your presence from 1 or 2 LLMs to 3 or 4 LLMs on the same query.
The central KPI to track is Coverage Breadth, a concept formalized by Stacker in 2026. It expresses the share of your strategic prompts where you appear on at least 3 of the 4 major LLMs. This KPI is more informative than raw citation frequency because it measures uniformity of presence, not volume.
Without a dedicated tool, measurement remains possible but plateaus at around a dozen prompts and loses statistical reliability. The natural variability of LLM responses (same prompt, same day, different answers) requires multiple executions to obtain a robust signal, which a tool natively automates.
FAQ on earned media and AI citations
What is earned media in a GEO context?
Earned media encompasses all content published about your brand on platforms you do not own and do not directly pay for. This includes industry press articles, reuse by third-party publishers, podcast interviews, mentions in external studies, and discussions on forums or Wikipedia.
Does earned media replace on-page SEO?
No. The two are complementary. Earned media generates the majority of AI citations (64% according to Stacker), but LLMs use your site to validate brand authority. Without a correctly structured site, earned media citations point to a weak source, which degrades conversion.
How much does a GEO earned media strategy cost in France?
Budget depends on chosen levers. For an SMB, count 2,000 to 5,000 euros per quarter for a publisher distribution strategy, 3,000 to 8,000 euros for an original study to distribute, and 500 to 1,500 euros per targeted press outreach. A complete annual program starts at 15,000 euros and can exceed 80,000 euros for a major account.
Which publishers really target LLMs in France?
LLMs mainly recognize industry media with established authority: Siècle Digital, Frenchweb, Journal du Net for digital, Les Échos and Usine Digitale for business, and vertical sector media (legal, health, finance depending on your industry). Mainstream general media produce fewer thematic AI citations.
Do paid publications (advertorials) count as earned media?
Technically no. An advertorial is paid media, not earned. LLMs weight less the content clearly identified as advertising. To maximize GEO effect, prioritize non-sponsored content whose publication results from genuine editorial interest, even if you pitched the topic.
How long between an earned media action and its effect on AI citations?
Between 30 and 90 days for most actions. LLMs update their sources with variable delay depending on the models: Perplexity and engines with active web search (GPT-5 with web) integrate faster, models with strict cutoff require waiting for a training refresh or remaining visible on the web for connected engines to capture them.
Does a local SMB benefit from earned media for GEO?
Yes, as long as you target locally. For a local SMB, regional press, geo-localized industry media, and mentions in studies about the city or region produce a strong effect on local queries in ChatGPT and Perplexity. Entry budget is lower than for a national player, with often better cost-per-citation ratio.
Sources
- Stacker + Scrunch, Coverage Breadth Study: The Latest GEO Research on Expanding Brand Visibility Across LLMs, March 2026, https://stacker.com/blog/latest-research-on-expanding-brand-visibility-across-llms
- Edelman, How Brands Can Stay Visible in an AI-Driven Search World, Nick Taylor VP Product Marketing AI, May 2025, https://www.edelman.com/insights/how-brands-stay-visible-ai-search
- Muck Rack, What AI Is Reading: Source Attribution in LLM Citations, August 2025 (relayed by Stacker), https://muckrack.com/blog/2025/08/13/what-is-ai-reading/
- Stanford HAI, AI Index Report 2025, https://hai.stanford.edu/ai-index/2025-ai-index-report
