On 4 March 2026, ChatGPT switched its default model to GPT-5.3 Instant. Within 24 hours, the average number of unique domains cited per response dropped from 19.1 to 15.2 — a 20.5% contraction measured across 27,000 responses and 400 prompts tracked by Resoneo and Meteoria. The Bigfoot Effect, as they named it, did not change how deeply ChatGPT reads a given domain. The URLs-per-domain ratio held stable at 1.26 throughout. It changed which domains get invited into each response at all.
GPT-5.4 made this structural rather than incidental. Analysis by Lily Ray and Chris Long found that the newer model now runs 10+ fan-out queries per response and uses explicit site: operators targeting Clutch and G2 by domain name. The model is not preferring these platforms in a general, fuzzy sense. It is literally searching them by domain during its retrieval process. If you are not on Clutch or G2, you are not in that retrieval step. Not ranked lower — simply not consulted.
Aaron Haynes tested this from a different angle: 0 out of 300 press release citations across 300 platform-query combinations. Zero. The content that companies spend the most money distributing — press releases on wire services, corporate announcements, product launches — earns no AI citations whatsoever. The content that earns citations is editorial: articles in publications journalists wrote, reviews from clients who chose to leave them, entries in directories editors compiled, Wikipedia articles where notability was established.
What Loren Baker Got Right
Loren Baker, Founder of Search Engine Journal and one of the people who has watched this industry longer than almost anyone, described the mechanism with clarity in April 2026: “Unlinked brand mentions are now doing work that used to require a backlink. When large language models train on data, they absorb patterns of association. If your brand shows up consistently alongside a topic, you become part of that topic’s semantic neighbourhood. You become the answer before anyone asks the question.”
That is not a new observation for people who have worked in entity SEO. It is the same principle that makes Wikidata entries valuable, that makes consistent NAP data across directories matter, that makes review platform presence compound over time. What has changed is the commercial urgency. The training data that determines ChatGPT’s parametric citations is being updated on a faster cycle. The concentration is tightening — 67% of AI citations in any category captured by the top 30 domains (Kevin Indig, Growth Memo, 21,482 citation rows). The window between “no one is doing this” and “the top 30 domains have locked in their positions” is closing.
The 2003 Parallel Is Not a Cliché
Baker made the comparison explicitly: “The brands building this kind of presence right now are doing the same thing early SEOs did in 2003 — establishing authority before the channel got crowded.” That comparison is worth taking literally, not rhetorically.
In 2003, most businesses had no systematic approach to building inbound links or domain authority. The businesses that started building editorial relationships, getting mentioned in the early blog ecosystem, earning citations from authoritative directories — they compounded an advantage that later entrants could not replicate at the same cost. By 2008, the gap between those with established link equity and those trying to build it quickly was wide and expensive to close.
The AI editorial record is in the same pre-consolidation stage in 2026. The businesses getting editorial mentions, building Clutch profiles with verified client reviews, establishing Wikipedia notability, creating the cross-source entity corroboration that AI systems use for provider recommendations — they are building the equivalent of that 2003 link equity. In three years, the concentration will have hardened. The editorial record will have a moat around it that later entrants will find expensive to breach.
What “Getting Into the Editorial Record” Actually Requires
It does not mean press releases. It does not mean sponsored content. It means appearing in sources that AI systems were trained to recognise as editorially independent — because the fundamental mechanism is AI systems trying to avoid recommending businesses based only on self-promotional content.
The University of Toronto measured this across 13 industries in September 2025 and found that 92.1% of AI citations came from third-party sources, not the brand’s own content. Muck Rack’s Generative Pulse analysis of over one million AI response links confirmed the direction: 82% of all AI citations come from earned media. The AI systems that generate recommendations have learned what editorial independence looks like, and they weight it accordingly.
The actions that build editorial record at scale: genuine client reviews on platforms AI systems index (Clutch, G2, TripAdvisor, Trustpilot — depending on sector). Contributed articles in industry publications that journalists and editors choose to run. Expert quotes in roundup pieces where your specific expertise adds something others cannot. Wikipedia articles for entities notable enough to qualify. Wikidata entries for founders, products, and organisations. Conference presentations and academic citations where your work is referenced. None of these can be manufactured at speed. All of them compound.
The Payoff: 14.2% vs 2.8%
Seer Interactive measured the downstream commercial effect across twelve million website visits: traffic arriving via AI citation converts at 14.2% versus 2.8% for standard organic search. That is a 5x quality multiplier on the same volume of visitors — because a visitor who arrives via AI citation has received a recommendation from a platform they chose to consult for sourced, evidence-based answers. They are not browsing. They are evaluating. The decision to contact you was largely made before they clicked.
That quality difference is why editorial presence in AI training data is not a brand awareness exercise with vague ROI. It is a lead quality investment with a measurable payoff. The businesses building it now will not just have more AI citations in three years. They will have better-converting traffic, a higher share of the AI recommendation surface in their category, and a compounding advantage that compounds faster as AI search captures a larger proportion of commercial queries.
The editorial record is open right now. Most of your competitors have not started building it. That gap will not stay open. — Sean Mullins, SEO Strategy Ltd, April 2026.
For the framework that determines whether individual pages are citation-eligible, see the CITATE standard. For the full citation dominance strategy including the off-site editorial layer, see AI Citation Dominance. For the platform-specific mechanics of how ChatGPT’s editorial source pool works, see How to Rank in ChatGPT.
