Contents
AI search is no longer a single visibility problem
YouTube is now AI infrastructure, not just a video channel
Wikipedia and NIH are becoming shared trust layers
Earned media remains the backbone, but press releases are losing weight
Citation rate compression is a key signal marketers should watch
Different AI models now trust different ecosystems
Data journalism shows how to make content AI-legible
What brands should do now
Conclusion: The next phase of AI visibility is source strategy
FAQ
AI search is no longer a single visibility problem
Generative AI is changing how consumers discover brands, compare products and make decisions. But the strategic question for PR and marketing leaders is no longer simply whether a brand appears in AI-generated answers, but which sources do AI platforms cite when a brand is mentioned.
Meltwater’s May 2026 analysis of more than 8 million citations across eight major LLMs shows that AI visibility is becoming more fragmented, more measurable and more source-dependent. The latest data confirms that large language models are not converging around one universal citation hierarchy, instead, they are developing distinct trust patterns shaped by source type, format, authority and model design.
That means a brand can perform well in one AI ecosystem and remain almost invisible in another. For communications teams, this makes generative engine optimization a reputation strategy, not just a search strategy.
| Rank | Source | Total Citations | Visibility Score | Citation Rate | Cit. Rate Score | MoM Visibility |
|---|---|---|---|---|---|---|
| 1 | YouTube | 188,863 | 18% | 0% → | 14% | +38.5% ↑ (+56.4% citations) |
| 2 | 111,860 | 11% | 16.7% ↓ | 10% | 0% → (+13.0% citations) | |
| 3 | NIH | 91,227 | 8% | 0% → | 6% | +33.3% ↑ (+41.3% citations) |
| 4 | Wikipedia | 83,191 | 9% | 14.3% ↑ | 8% | +50% ↑ (+55.2% citations) |
| 5 | Statista | 74,875 | 8% | 11.1% ↓ | 8% | 0% → (+11.8% citations) |
| 6 | Yahoo | 64,447 | 7% | 22.2% ↓ | 7% | -12.5% ↓ (-6.6% citations) |
| 7 | Forbes | 55,975 | 6% | 0% → | 7% | 0% → (+30.4% citations) |
| 8 | 53,481 | 6% | 12.5% ↓ | 7% | -14.3% ↓ (-9.7% citations) | |
| 9 | 41,908 | 5% | 16.7% ↓ | 5% | 0% → (+4.4% citations) | |
| 10 | Nerdwallet | 40,752 | 4% | 20.0% ↓ | 4% | 0% → (+32.1% citations) |
Meltwater's GenAI Lens helps businesses like your measure and boost their visibility in AI search - find out what it can do for you!
YouTube is now AI infrastructure, not just a video channel
YouTube remains the clearest winner in the May data. It generated 188,863 citations, making it the most-cited source overall, and grew 56.4% month-over-month.
YouTube now appears to be foundational AI infrastructure, particularly for Google AI Mode, Google AI Overviews, Gemini and Perplexity. Google AI Mode alone generated 73,813 YouTube citations, while AI Overviews generated 38,002, Gemini generated 37,924 and Perplexity generated 25,949.
For marketers this shows that AI platforms are not just responding to video popularity, they are reading the structured information around the video, especially transcripts, descriptions, titles and metadata.
That makes YouTube optimization a core GEO lever. Product explainers, category education, executive interviews, research summaries and comparison content can all become AI-citable assets when they are structured clearly enough for models to parse and reuse.
Wikipedia and NIH are becoming shared trust layers
If YouTube is the dominant reach source, Wikipedia and NIH are the clearest authority signals.
Wikipedia reached 83,191 citations in May, growing 55.2% month-over-month - its visibility score also increased by 50%, while its citation rate improved by 14.3%. That combination makes Wikipedia one of the strongest multi-metric gainers in the dataset.
NIH/NCBI entered the top three sources overall with 91,227 citations, up 41.3% month-over-month, surpassing both Wikipedia and Statista in raw volume.
Together, these movements suggest that AI systems are increasingly rewarding sources that offer neutrality, authority, structure and evidence. Wikipedia is becoming a broad trust layer across multiple models, while NIH’s rise points to a stronger weighting toward scientific, medical and government-adjacent authority.
This goes beyond healthcare, signaling a wider shift in AI citation behavior: models appear to be placing more weight on sources that are factual, structured and externally validated.
Earned media remains the backbone, but press releases are losing weight
| Source Type | May Count | May % | April % | MoM Change | Multi-Month Trend | Trend |
|---|---|---|---|---|---|---|
| Other | 4,292,419 | 53.7% | 50.3% | +3.4 pts | Dip Apr, rebounds May | → Inconclusive |
| Earned/News | 3,006,799 | 37.6% | 39.5% | -1.9 pts | ↑ Mar→Apr, slight dip | → Stable band |
| Social | 463,279 | 5.8% | 7.2% | -1.4 pts | Volatile — spike Apr, dip May | → Mixed |
| Reviews | 127,956 | 1.6% | 1.7% | -0.1 pts | Stable | |
| Press Release | 22,679 | 0.2% | 0.4% | -0.2 pts | ↓ Declining | ↓ Under pressure |
At the category level, earned media remains one of the strongest levers for AI visibility.
Earned/News media accounted for 37.6% of citation domain count in May. That is down slightly from 39.5% in April, but the multi-month picture suggests a stable band of roughly 37% to 40%. In other words, earned coverage remains one of the most reliable ways for brands to enter the sources AI systems trust.
The more concerning signal is the decline in press releases. Press Release citation share fell from 0.4% in April to 0.2% in May, the sharpest category-level decline in the dataset.
That does not mean press releases no longer matter. They still serve important distribution, disclosure and media relations functions. But as an AI visibility lever, the data suggests they are much weaker than earned editorial coverage, structured data, authoritative third-party sources and AI-readable video content.
The lesson here is that publishing a claim is not the same as earning authority for that claim.
Citation rate compression is a key signal marketers should watch
One of the most important findings in May is not about which sources grew, but how some sources grew.
Several major platforms maintained strong citation volumes while seeing citation rate declines. Reddit grew to 111,860 citations, up 13.0% month-over-month, but its citation rate declined 16.7%. NerdWallet grew 32.1% in citation volume, but its citation rate fell 20.0%. LinkedIn reached 53,481 citations, but its citation rate declined 12.5%. Yahoo’s citation rate fell 22.2%, while its total citation volume declined 6.6%.
This suggests that some sources are being cited more broadly but less intensively per topic. In practical terms, volume alone may not tell the full story. A source can appear frequently across the ecosystem while becoming less concentrated, less efficient or less dominant within specific categories.
What does this mean for marketers? Source rank matters, but so does citation concentration. The next generation of GEO measurement needs to track not only whether a brand is cited, but where, by which model, through which source and with what level of topic-level consistency.
Different AI models now trust different ecosystems
May’s data reinforces one of the most important patterns from earlier research: AI models behave like distinct information ecosystems.
Claude remains the data-centric analyst. Statista leads Claude citations at 21,620, the highest single model-source pairing in the dataset. NIH follows at 18,640, confirming Claude’s continued preference for structured, data-rich and authority-led sources.
ChatGPT remains more institutionally anchored. Its citation footprint is concentrated around Wikipedia and NIH, with 9,010 Wikipedia citations and 8,053 NIH citations. It shows no meaningful citation volume for YouTube, Reddit, LinkedIn or Facebook in the top-source table, reinforcing its authority-first profile.
Copilot’s B2B profile has shifted. Forbes now leads Copilot citations at 12,176, ahead of LinkedIn at 9,546 and Statista at 9,718. That suggests professional and business media coverage may be especially important for brands seeking visibility in Copilot.
Google AI Mode remains the reach engine. It generated the highest raw citation volumes across YouTube, Reddit, Yahoo, LinkedIn and Facebook. For brands seeking broad visibility in Google’s AI ecosystem, video, social discussion and mainstream web presence remain essential.
Gemini is becoming the most balanced model in the dataset. It draws heavily from YouTube, Reddit, Wikipedia, NIH and Forbes, combining social, video, institutional and editorial signals.
Perplexity continues to behave like a research engine. YouTube remains its top source, followed by Wikipedia and NerdWallet, while Reddit remains absent from its top citation profile.
Grok is also evolving. Earlier data suggested a more social-first profile, but May shows a broader mix including Reddit, LinkedIn, Wikipedia, Forbes and NIH. The fully social characterization no longer holds.
The implication is simple but strategically significant: brands need to stop thinking about AI visibility as one channel. Visibility in ChatGPT, Claude, Gemini, Perplexity, Copilot, Grok and Google’s AI experiences requires different source strategies.
Data journalism shows how to make content AI-legible
The May report also reveals an important pattern at the journalist and creator level.
Three of the top 10 most-cited journalists and content creators are from Visual Capitalist, a publication known for structured, chart-led and data-rich explainers. This is not just a media relations insight, but also a content-format insight.
AI systems appear to reward content that is easy to parse, summarize and cite. Structured data, clear headings, visual explainers, rankings, tables, statistics and source-backed narratives all make content more machine-legible.
This means the format of a story may matter almost as much as the outlet that publishes it. Dense narrative content may be less useful to AI systems than clearly structured, data-backed content with extractable facts.
The Visual Capitalist effect is a useful lesson: AI visibility is impacted heavily by information design.
What brands should do now
1. Treat YouTube as foundational GEO infrastructure
YouTube’s citation dominance is now too large to treat as incidental. Brands should prioritize clear titles, detailed descriptions, accurate transcripts, chaptering, schema-friendly metadata and educational formats.
The key is not simply producing more video. It is producing video that AI systems can understand.
2. Invest in Wikipedia as a trust layer
Wikipedia’s growth across multiple models makes it a foundational source for AI visibility. Brand, product and category information should be factual, neutral, current and well-structured.
This is not about promotional editing. It is about ensuring that public, factual information is accurate and complete.
3. Build authority through third-party validation
Earned media remains one of the strongest pathways into AI answers. Press releases may still support communications workflows, but the AI visibility premium appears to sit with trusted editorial coverage, authoritative databases, review platforms and structured third-party sources.
4. Make data citable
The rise of NIH, Statista and Visual Capitalist reinforces the value of structured, sourced and dated data. Brands should turn proprietary insights into charts, benchmarks, rankings, explainers and research-led narratives that are easy for AI systems to reference.
5. Measure AI visibility by model
A blended AI visibility score can hide critical differences. Claude, ChatGPT, Copilot, Google AI Mode, Gemini, Perplexity and Grok each show different citation preferences.
Brands should measure visibility by model, by source type and by citation pathway.
Conclusion: The next phase of AI visibility is source strategy
The May 2026 data suggests that AI search is entering a more mature and more selective phase. YouTube continues to dominate, Wikipedia is becoming a shared trust layer, NIH has emerged as a major authority source, and data-rich content is gaining measurable influence.
At the same time, the differences between models are becoming more pronounced. Some AI systems lean toward institutional authority. Others lean toward video, social discussion, business media or structured data.
For brands, this means generative search visibility is no longer simply about publishing more content. It is about appearing in the sources AI systems already trust.
The organizations that understand that shift earliest will be best positioned to build authority, influence discovery and maintain visibility as AI search continues to evolve.
FAQ
1. What is AI visibility?
AI visibility refers to how often and where a brand, source or piece of content appears in AI-generated answers. It includes not only whether a brand is mentioned, but which sources AI systems cite when producing those answers.
2. Why does YouTube matter so much for AI visibility?
YouTube matters because AI systems can use transcripts, titles, descriptions and metadata as structured information sources. In May 2026, YouTube was the most-cited source overall, with 188,863 citations.
3. Is earned media still important for generative search?
Yes. Earned/News accounted for 37.6% of citation domain count in May and has remained one of the most stable citation categories across multiple months. This suggests that editorial coverage remains a major pathway into AI-generated answers.
4. How should brands measure AI visibility?
Brands should measure AI visibility by model, source type and citation pathway. A single aggregate score can hide major differences between platforms such as ChatGPT, Claude, Gemini, Copilot, Perplexity, Grok and Google AI Mode.
