Vanity vs. Impact: Here’s What Actually Matters for GEO
By Tanner Skotnicki
Ask any PR or marketing professional for their industry predictions for this year and you can guarantee they’ll emphasize the importance of generative engine optimization (GEO). Much like how SEO became the talk of the town in the early 2000s, GEO has quickly found itself a central component of searchability for brands – and for a good reason. According to McKinsey and Company, 50% of consumers use AI-powered search with 44% of users citing it as their “preferred source of insight.”
With buyers being fed results the moment they hit enter for (soon to be) 75% of queries, external communication professionals across industries were quick to learn how to get their brand’s name into that new “AI Overview” landscape. And while the consensus agrees on how (third-party validation, optimized owned content, external media placements) these strategies work, not many have delved into why.
Foreword
Before going into the reasoning behind why specific content is so heavily favored by AI, I have to confess — I’ve been lying to you. The “AI” that’s been described up until now is not actually artificial intelligence. What we are discussing in terms of enhancing searchability are large language models (LLMs).
This distinction is necessary because LLMs do not actually “think” for themselves. Rather, they are trained on a vast amount of data and regurgitate information based on internal “scores.”
These scores include:
- Original source ranking (i.e., page 1 of Google is weighted more heavily than page 10)
- E-E-A-T — experience, expertise, authoritativeness and trustworthiness
- Query intent, which assesses the user’s end search goal
Because models are reliant on these scores for their outputs, they can be retrained to produce different outputs. And that’s precisely what comms pros have figured out.
Why this and not that?
As self-serving as it would be to say these LLMs weigh earned media placements heavily just because, there is a rationale.
Content Scan-ability: Structured Text vs. Static Files
LLMs are trained primarily on raw, machine-readable text. What that means in practice is that content embedded directly in a webpage’s HTML (think a paragraph wrapped in a <p> tag or a headline in an <h2>) is far more accessible to a model during training than the same content locked inside a PDF, image or slide deck.
When a PDF is uploaded to a website, the text inside it is effectively invisible to most crawlers unless the site has taken extra steps to extract and republish that text as HTML. A model training on web data will index the page, find no parse-able body copy, and move on.
Meanwhile, a brand that publishes that same content as a structured webpage — with headers, short paragraphs, and semantic HTML — has significantly higher odds of having that content absorbed during training and surfaced in a generative response.
This is why press release wire services matter more than they may appear. A well-structured press release published to a high-authority distribution platform creates a machine-readable, indexed, linkable text artifact. It’s something a model can cleanly ingest, associate with your brand, and draw from when a relevant query is posed.
Semantic Context and Entity Association
The second factor is less intuitive but equally impactful: LLMs don’t just read words — they map relationships between entities. An “entity” in this context is anything a model can identify and categorize: a company, a person, a product, a location, a concept.
When your brand name appears consistently alongside relevant entities (i.e. your industry, your competitors, your key use cases, credible third parties) models begin to build what’s called a “high-confidence entity association.” Think of it like a web of credibility. The more often your brand appears in the same sentence as authoritative, contextually relevant subjects, the stronger that association becomes in the model’s internal weighting.
This is why earned media placements and brand mentions from credible publications carry so much weight in GEO. A quote from your CEO in a Forbes article about AI in supply chain doesn’t just give you a backlink for SEO purposes. It teaches the model that your brand is a relevant, trusted voice in that category. Repeat that signal across enough sources, and you’ve effectively lobbied the model into including you in relevant generative results.
Press releases, analyst inclusions, award recognition, and category listings on third-party sites all serve the same purpose: they create distributed, consistent, semantically rich associations between your brand and the topics you want to own — across domains that the model treats as authoritative.
Source Authority and Citation Chains
This third factor circles back to the original scoring system: not all content is created equal in the eyes of an LLM. Models are trained on data that has, in effect, been pre-ranked by the internet’s existing authority structures. Content from high-domain-authority sources like major publications, government sites, academic institutions or recognized industry bodies are weighted more heavily than content from low-traffic blogs or brand-owned microsites.
What this creates is a citation chain effect. If a Tier 1 publication covers your brand, that coverage then becomes a source that other sites may reference, quote or link to. Each subsequent mention reinforces the authority signal and broadens the entity web described above. A model training on this ecosystem will encounter your brand name in progressively more authoritative contexts, increasing the confidence score associated with including your brand in a relevant response.
This is why award programs, industry analyst rankings, and trade publication features punch above their weight in a GEO environment. They inject your brand into the citation ecosystem that models treat as ground truth.
It also explains why vanity metrics like social shares or branded hashtag campaigns have minimal GEO impact. They generate noise, not signal. A post shared 10,000 times on a platform that most models don’t train on does almost nothing to improve your model confidence score.
The Bottom Line
GEO is not a new checklist to layer on top of your existing marketing strategy. It is a fundamentally different way of thinking about what it means to be discoverable — one where the audience is a machine making probabilistic judgments about your brand’s authority, relevance, and trustworthiness.
The tactics that move the needle — press releases, earned media, third-party validation, structured owned content — are not vanity plays dressed up in new language. They are the specific inputs that LLMs are architected to favor. Understanding the mechanics of why is what separates a GEO strategy that compounds over time from one that simply checks boxes.
