ChatGPT leans toward primary sources, Ahrefs data shows

Reddit still tops ChatGPT's citation tables in the United States. But the sites gaining ground beneath it tell the more useful story. Per Search Engine Journal, Ahrefs data shows government health agencies, research institutions and regulatory bodies climbing ChatGPT's U.S. citation rankings, while general-purpose explainer content, the WebMDs and listicle-style guides of the world, is losing share.

This is not a reshuffling at the margins. It is a statement about what ChatGPT now considers trustworthy enough to cite when a user asks something that touches health, policy or regulation. The model is increasingly reaching past the intermediary that explains a topic and toward the body that issued the underlying data or rule. A question about drug interactions pulls from the FDA rather than a health-advice aggregator. A question about labour statistics pulls from the Bureau of Labor Statistics rather than a news explainer summarising them.

The intermediary's shrinking job

For two decades, the commercial logic of search rewarded whoever explained a primary source most clearly, in the most scannable format, with the best on-page SEO. That is the business model behind every "What is X" content hub built since 2010. Ahrefs' data suggests that logic is weakening inside ChatGPT's retrieval layer. The model does not need an explainer to make a regulatory filing legible; it can parse the filing and generate the explanation itself, on demand, tailored to the question asked. The explainer's traditional value, translation, is being absorbed into the model. What remains valuable is the thing being translated: the primary document, the dataset, the agency statement.

This matters most for sectors that already sit closer to the primary-source end of that spectrum than they realise. A multilateral like UNDRR publishes disaster-risk data that others summarise into think-pieces and news articles. A regulator-adjacent body like ISO issues standards that consultancies spend whole careers explaining to clients. CGAP, sitting inside the World Bank system, produces financial-inclusion research that gets cited, requoted and diluted across hundreds of secondary sources. Each of these organisations has spent years treating its own publications as inputs to someone else's content strategy, rather than as the destination. Ahrefs' findings suggest that posture now works against them in LLM answers.

Why "citable" now means something narrower

The shift also changes what counts as citable content for a financial-services or industrial brand. A bank's own regulatory disclosures, compliance statements and published research carry more retrieval weight under this pattern than a blog post interpreting a competitor's earnings call. A standards body's original text outranks a trade publication's summary of it. An industrial group's safety data sheet or technical specification, long treated as back-office documentation, becomes a more valuable citation asset than the marketing copy built around it.

The practical consequence is uncomfortable for communications teams built around the explainer model. Years of investment went into content designed to simplify, contextualise and humanise primary material for a general audience. That content has not become worthless, but its citation value in ChatGPT answers is drifting downward relative to the primary document itself. Brands that hold original data, regulatory standing or research output now have reason to publish that material directly, structured and accessible, rather than routing it exclusively through a layer of secondary explanation aimed at search-engine readers.

There is a second-order effect worth watching. If ChatGPT increasingly cites the regulator over the explainer, the explainer's traffic and authority both decline, which weakens the secondary ecosystem that has historically amplified primary-source organisations' reach. A UN agency or central bank that relied on journalists and analysts to carry its findings into public discourse may find that LLMs now do that carrying directly, cutting out the intermediary and, with it, some of the framing control that intermediary provided. Primary sources gain citation share. They do not automatically gain narrative control over how that citation gets used.

Ahrefs' data covers the U.S. market and one model. But the direction is the one to track: retrieval systems rewarding origin over explanation. Organisations that sit on primary data, standards or regulatory authority should treat that asset as a citation product in its own right, not raw material for someone else's content.