Publishing more content on your own site no longer guarantees visibility in AI-driven search. That idea fell apart after two GEO experiments tracked 775 citation events across ten platforms. The results were clear: third-party mentions now dominate the way large language models surface brands.
The second experiment flipped old assumptions. Owned listicles-once a staple of AI optimization-performed far worse than earned placements on external sites. The most-cited sources weren’t brand-owned, but high-frequency listicles and review sites already favored by LLMs. One agency’s self-published listicle took 18 days to get its first citation. In contrast, third-party placements were picked up overnight and accounted for 85.8% of all mentions.
According to a 2026 industry report, earned media now accounts for approximately 84% of all AI citations, while paid or advertorial content contributes just 0.3%.
What changed? The experiments started with a simple question: does controlling your own narrative guarantee AI visibility? The first test, run over several months for an established brand, targeted 15 commercial-intent keywords across ChatGPT, Claude, Gemini, and Perplexity. The approach mixed listicle placements, PR, guest posts, and LinkedIn activity. By the end, the brand appeared for 10 to 12 of the tracked keywords, peaking at 37.01% keyword presence. But the real driver was a comprehensive listicle on Indeed SEO, which alone generated 190 citations-more than all other sources combined.
Source power and decay
The authority and relevance of the source made the difference. Government and education domains, news publications, and industry-specific sites consistently outperformed generalist outlets. Getting placed on a source already cited by LLMs quickly boosted visibility. But the effect faded fast: half of all sources stopped being cited within 30 days, and one MEXC placement dropped from 29 to 11 mentions in a week. PR could amplify a strong placement but couldn’t save a weak one.
Peer associations mattered, too. When a listicle included the brand alongside recognized experts like Lily Ray and Aleyda Solis, citations jumped. Removing those names led to an immediate drop, suggesting LLMs weigh the company you keep as much as the mention itself. The same pattern showed up in PR: being named with industry leaders outperformed solo features, even on stronger outlets.
The August 2026 Meltwater report highlights that the share of earned/news sources in the AI citation mix dropped to 29.4% from 33.1% in July, marking the third consecutive month of decline. Despite this, earned/news sources still generated about 2.1 million domain-level citations, underscoring their continued significance in AI search visibility.
Meltwater, AI Search Visibility Report
Keyword precision and content structure
Exact-match keyword targeting still worked. Dedicated assets for phrases like "Best LLM SEO Consultant" delivered results, while broader coverage didn’t fill the gap for similar queries. Content structure also mattered: answers in the first 100 words, visible FAQ sections, and question-based headings all improved citation rates. Freshness-recent data, current references, and visible publication dates-correlated with better performance.
Query capitalization and type affected which sources LLMs pulled. Capitalized and lowercase queries returned different citations in repeated tests. Software and tool queries favored high-authority review sites, while service queries surfaced more listicles. Matching placement type to query intent increased the odds of being cited.
Cold start and earned placement dominance
The second experiment, a 30-day cold start for a SaaS link building agency, added Google AI Mode and Grok to the mix. Starting from zero, the brand appeared 298 times in a month, with Gemini and Google AI Mode making up two-thirds of the total. Referral traffic told its own story: 18.5% of new users arrived via referrals, and 3.25% through GA4’s AI Assistant channel. Even a single Perplexity referral converted to a paying customer, despite low overall volume.
Building the outreach list from observed citations-not just domain rating-worked better. Indie Hackers, Bruce Jones SEO, and TechBullion saw citation jumps of 232%, 165%, and 147% after placements went live. But not every placement delivered: some shortlisted sites earned no mentions, and citation volume didn’t always mean more clicks. Indie Hackers led in citations but saw flat referral traffic, while TechBullion’s lower citation count drove a 64-session spike.
Comparative content and business impact
Comparative content outperformed advocacy. Updating the owned listicle to include leading competitors triggered a 12-fold increase in mentions, from four to 49, and daily visibility peaked at 95. This mirrored the peer-set effect from the first experiment, confirming that LLMs reward category resources over self-promotion. Claude focused on high-intent terms, while Gemini spread appearances more broadly-platform distribution mattered as much as raw totals.
For creators and marketers, the lesson is direct: run your target keywords through AI platforms before investing in placements. Build outreach lists from observed citations, not spreadsheets. Budget for ongoing maintenance, since half of sources decay within a month. Pay close attention to who appears alongside you-entity associations can make or break visibility. And don’t assume citation volume equals business impact; a single low-volume referral can outperform the most-cited source.
Traditional SEO’s focus on ranking pages is now outpaced by LLM-driven search, where associations, source authority, and recency decide who gets surfaced. The experiments break the myth that owned content is the main lever for AI visibility. Instead, the data points to a new playbook: prioritize earned placements on sources LLMs already trust, reinforce those mentions with PR and guest posts, and keep your assets fresh and comparative. For a related look at how unconventional data sources can reveal hidden opportunities, see this earlier breakdown.
Gemini, Google’s flagship AI platform, launched in 2024 and now powers AI search features across Google’s core products. By 2026, Gemini is integrated into Google AI Mode, which has rolled out to over 1.2 billion users worldwide. The platform’s citation and answer mechanisms are updated weekly, and Google reports that over 40% of commercial-intent queries now trigger Gemini-powered responses in supported regions.
Recent industry analyses show that SEO rankings alone no longer guarantee AI visibility. Instead, trust in the source and the frequency of its mentions on already-cited LLM pages have become more influential, as detailed in the 2026 ABNewswire-distributed industry report. This shift means brands need to focus on building authority and earning mentions from reputable third-party domains.
The difference between "being cited" and "being selected" is now central in the GEO and AEO landscape. As explained in a 2026 analysis by Paul A. C., optimizing for generative answers now requires attention to both external mentions and machine-readable content structure, not just traditional SEO ranking.
