Fifty Websites Now Control What AI Tells You About Everything

There's a quiet power struggle happening inside every AI chatbot you use, and most people have no idea it's going on. When you ask ChatGPT who makes the best electric vehicle, or ask Perplexity about a breaking news story, or query Google's Gemini about a medication, the answer you receive isn't coming from some neutral, omniscient oracle. It's being shaped by a surprisingly small group of websites that these systems have learned to trust above all others.

A new index from Morningstar is putting hard numbers on something the tech industry has been quietly aware of for a while. According to their AI Citation Index, roughly 50 websites now account for a disproportionate share of what major AI assistants cite when generating answers for users. Fifty. Out of the billions of pages on the open internet.

The Citation Bottleneck Nobody Talked About

To understand why this matters, you need to think about how AI language models actually work in practice, especially the ones with real-time retrieval capabilities. Systems like Perplexity AI, ChatGPT with browsing enabled, and Google's AI Overviews don't just generate text from their training data. They actively pull from live sources, rank them by perceived credibility, and weave those citations into their responses. The sources they trust most heavily end up shaping the answer.

What Morningstar's index surfaces is the concentration risk buried inside that process. A handful of legacy publishers, major news organizations, government databases, and established reference sites have emerged as the de facto authorities for AI systems. Wikipedia predictably ranks near the top. So do Reuters, the Associated Press, and major financial data providers. Academic databases and government health agencies appear frequently. These are not surprising choices on their own.

The problem is what gets excluded. Thousands of legitimate, high-quality sources, including specialized trade publications, independent researchers, regional news outlets, and niche expert communities, rarely make it into AI-generated answers at all. The AI doesn't know they exist in any meaningful operational sense, even if it technically trained on their content.

Search Visibility Is About to Get Much Worse for Most Publishers

Traditional SEO was already a brutal game. Google's algorithm changes could wipe out years of traffic overnight, and smaller publishers spent enormous resources chasing ranking signals they barely understood. But at least the old system had hundreds of ranking positions. Page one of Google had ten results. Page two had ten more. A motivated user could dig through dozens of sources.

AI chat interfaces don't work that way. They give you one answer. Maybe two or three sources cited at the bottom if you're lucky. The entire concept of "ranking on page two" becomes irrelevant when the interface doesn't have pages. You're either in the answer or you don't exist for that user in that moment.

This is the structural shift that should be keeping media executives up at night, and increasingly is. If 50 websites capture the majority of AI citations, everyone outside that group faces a visibility cliff that makes Google's algorithm updates look gentle by comparison. A regional newspaper that built decades of trust in its community, a specialized EV industry publication covering battery chemistry in depth, a medical research nonprofit producing genuinely useful public health content, none of them can easily compete with the institutional authority signals that Reuters or WebMD carry in an AI's training data.

Commerce and Brand Influence Are Already Being Reshaped

The implications extend well beyond media companies feeling squeezed. Consumer behavior is changing fast. A growing percentage of product research now starts with an AI query rather than a search engine. Ask an AI what the best home EV charger is, and the brands that appear in that answer are getting a recommendation from a source that millions of users have come to trust implicitly, sometimes more than they trust traditional search results.

For brands, this creates a new kind of marketing problem. Traditional digital advertising could reach consumers through paid placement. Traditional SEO could at least be worked on, optimized, and improved over time with enough resources. AI citation is different. You can't buy your way into a language model's trusted source list. You can't run an A/B test on your schema markup and watch your AI citation rate improve next Tuesday.

What you can do, to some extent, is get covered by the sources that AI systems trust. If Reuters writes about your product, if a major academic institution publishes data about your industry that references your company, if Wikipedia has a well-maintained article about your brand, you have a better chance of appearing in AI-generated answers. This effectively hands enormous gatekeeping power to the same legacy institutions that have always held it, and perhaps concentrates it further.

Smaller EV startups, for instance, face a genuine challenge here. A company building innovative charging infrastructure in the Pacific Northwest might produce genuinely useful products that local installers swear by. But if that company hasn't been covered by the 50 dominant sources, it may as well not exist from an AI assistant's perspective, regardless of how good its products actually are.

Regulators Are Starting to Notice, Slowly

Antitrust regulators in the US and Europe have spent the last several years building frameworks around search engine dominance, app store gatekeeping, and social media algorithmic control. The AI citation question is going to force them to think about an entirely new category of information chokepoint.

The EU's Digital Markets Act and the ongoing FTC interest in AI systems both create potential hooks for scrutiny. If a small group of websites effectively determines what information gets surfaced to hundreds of millions of users, and if those websites were selected through opaque training processes that nobody fully understands or audits, that's a significant concentration of informational power. Whether existing regulatory frameworks can actually address it is a different question entirely.

There's also a transparency problem that regulators haven't fully grappled with. When Google's algorithm favored certain sites, independent researchers could study SERPs, run experiments, and build reasonable models of how the ranking worked. AI citation processes are considerably more opaque. The weights assigned to different sources during training aren't published. The real-time retrieval logic in systems like Perplexity involves proprietary ranking signals. Even the companies building these systems may not have complete visibility into why their models trust certain sources more than others.

What the AI Companies Themselves Are Doing About It

OpenAI, Google, Anthropic, and Perplexity have all been in active negotiations with major publishers about licensing deals and citation policies. The New York Times lawsuit against OpenAI brought some of these tensions into public view, but licensing negotiations are happening across the industry. These deals matter partly for legal reasons and partly because they may formally encode which sources get privileged treatment in future model training.

Perplexity has perhaps been most explicit about its citation model, showing users sourced links alongside answers and allowing some degree of source inspection. Google's AI Overviews have been more controversial, particularly after early incidents where the system confidently cited unreliable sources or generated factually wrong answers with legitimate-looking citations attached.

The deeper issue is that every AI company is making editorial decisions whether they acknowledge it or not. Choosing which sources to include in training data, which sites to allow real-time retrieval from, which domains to trust for factual claims, these are editorial choices with enormous downstream consequences. Calling them "algorithmic" doesn't make them neutral.

Where This Leaves Everyone Else

Consumers are in an awkward position. AI assistants are genuinely useful, often dramatically more efficient than traditional search for many types of questions. But the concentration of trusted sources means users are getting a filtered view of the information universe, one curated by systems they can't inspect, toward sources they didn't consciously choose, with commercial and institutional biases baked in at the training level.

For people in the EV space specifically, this has real consequences. Questions about charging infrastructure, battery performance, range claims, government incentives, and new model comparisons increasingly get answered by AI systems. If those systems consistently surface information from a narrow set of sources, the nuance and depth that comes from specialized coverage gets lost. A general-purpose news organization covering EVs once a week will get cited ahead of a publication that covers nothing else.

Morningstar's index is valuable precisely because it forces a concrete conversation about something the AI industry has been treating as an implementation detail. It isn't. Who gets cited by AI systems is a question of informational power, commercial influence, and public epistemology. Fifty websites shaping what hundreds of millions of people learn when they ask questions deserves a lot more scrutiny than it's currently getting, from regulators, from civil society, and from the users who've started trusting these systems without fully understanding how they work.

The old gatekeepers of information didn't go away when the internet arrived. They adapted. The AI era may simply be the next round of that same story, with higher stakes and less visibility into who's actually pulling the levers.