The question shows up in nearly every brand strategy meeting now: how do we get ChatGPT to mention us? The honest answer is more useful than the marketing-pitch version. Here is what the available evidence supports about how citations get made and what communications teams can actually influence.
How ChatGPT decides what to cite
ChatGPT search, which OpenAI rolled out broadly in early 2025, works through a combination of training data, real-time web retrieval, and partnerships with select publishers. When a user asks a question that triggers web search behavior, the model retrieves results, synthesizes them, and presents an answer with citations. When it relies on training data alone, citations are absent but brand mentions still appear, drawn from whatever the model absorbed during training.
Two variables determine whether a brand surfaces. The first is whether the brand exists at sufficient density and authority across sources the model trusts. The second is whether the specific page that should rank is structured in a way the retrieval system can parse and prefer.
Communications teams can affect both.
Authority signals that actually matter
The hierarchy of source authority for retrieval-augmented systems looks broadly similar to the hierarchy journalists already use. Established news outlets, government and academic sources, and well-maintained reference content rank above content farms, low-authority blogs, and self-published material. This is good news for traditional PR work — earned placements in Tier 1 outlets carry forward into the AI layer.
A few specifics worth flagging.
Wikipedia. Brand entries that meet notability thresholds and are well-sourced anchor entity recognition across nearly every major model. Wikipedia is often quoted directly or paraphrased closely. Brands without an entry, or with a poorly-sourced one, surface less reliably.
Trade press in your category. ChatGPT often pulls vertical authority from category-specific outlets — Skift for travel, WWD for fashion, Adweek for marketing, Becker's for healthcare admin. A placement in the right vertical outlet often outweighs a generalist business outlet for category-specific queries.
Wire releases. Press releases distributed via PR Newswire, Business Wire, and similar services do appear in retrieval results, though they carry less authority than original journalism. They are useful for surfacing specific claims (a product launch, a hire, a financial result) rather than building general authority.
Page-level structure that helps retrieval
The technical requirements are not exotic. Schema.org Article and Organization markup, clean H1/H2 hierarchy, fast load times, and machine-readable publication dates all assist retrieval. Schema.org's Article specification has been the standard for over a decade and is well-supported by Google's AI surfaces and most third-party retrieval systems.
Two structural cues are underused. The first is direct question-and-answer structure on FAQ and resource pages. Models reward content that closely matches the conversational query format. The second is clear date metadata. Models trying to give recent answers prefer recent content, and an unstamped page is treated as an unknown vintage.
What does not work
Stuffing brand mentions into press release boilerplate beyond the standard count. Buying low-authority links. Duplicating content across many domains. Submitting to AI-specific "submission services" that promise indexing into LLMs — the major models do not have submission portals. Anyone selling one is selling vapor.
Coordinated inauthentic activity is also a poor strategy. Models trained on web text are good at detecting spam patterns at scale, and a brand whose visibility depends on artificially inflated content is taking on reputation risk that compounds over time.
A practical sequence
For a comms team starting from scratch, the order of operations is:
First, audit current presence. Run twenty to thirty representative queries across ChatGPT, Claude, Perplexity, and Google AI Overviews. Document what comes back.
Second, fix the entity layer. Confirm Wikipedia accuracy, ensure Crunchbase and Wikidata are clean, check that LinkedIn and major directories have current information.
Third, identify the source gap. If the model is citing a competitor's blog post when answering a category question, the work is to earn coverage in higher-authority sources that address the same question.
Fourth, structure owned content properly. Tag Article schema, add FAQ structure to resource pages, ensure publication dates are visible.
Fifth, measure on a recurring cadence. Quarterly audits track trend lines that single-point measurement cannot.
The discipline is not new. It is media relations, executive thought leadership, and digital hygiene reorganized around a different distribution endpoint. The teams that recognize the continuity will move faster than the teams looking for an AI-native silver bullet.
