Semantic consensus: How Digital PR builds your authority beyond your website

Generative visibility is not earned on your website. It is won in the citation graph, and most brands are still optimizing the wrong territory.

The “short list” has moved

For decades, marketing professionals have competed for a single goal: placing their brand in the consumer’s mental short list. That “short list” of preferred, pre-established options that used to be the primary factor in any purchase decision.

Advertising bought space in the most popular channels for its target audience, hoping that repeated exposure to the brand in connection with a differentiating attribute (Volvo → safety, Coca-Cola → the spark of life, Ariel → white clothes) would become recall and influence when making a purchase decision.

Let’s face it: with AI, that list of options has moved. When a potential B2B customer asks ChatGPT “what is the best CRM for a 200-person sales team?”, or when a user asks Perplexity “which smart lighting system works best with Apple Home?”, artificial intelligence returns a list of three to five recommended brands. It is no longer the user, with their biases and sociodemographic conditioning, who created that list. We did it—collectively—through every page, every comment, every piece of content shared online.

In other words, we no longer compete—or not only—to capture attention in the consumer’s mind. Now we also compete to be included in each LLM’s recommendations.

GEO: when visibility is cultivated across the entire ecosystem

Traditional SEO operated within a scope limited to your own domain. Technically flawless pages, on-page semantics, internal linking within the information architecture, authority provided by backlinks… SEOs orchestrated multiple tasks whose destination was invariably the URLs under our control.

According to numerous studies, visibility on AI platforms, however, no longer works like this:

• A Muck Rack study from Q4 2025 revealed that 89% of all AI citations come from earned media, not from content on a brand-owned property (its website, its social media profiles…).
• SparkToro’s 2026 analysis confirms that approximately 85% of brand mentions in AI answers originate on third-party pages.
• Ahrefs recently tested structured data markup (schema) and found that, contrary to the intuitive perception that structured markup should be more effective for machine reading, its implementation did not change AI citations at all.

In short: if your entire strategy to be discoverable through generative answers is still focused on optimizing your website, you are only leveraging about 15% of the signal surface. This should be a warning sign for companies that continue to focus their online visibility strategy exclusively on optimizing their website. But it is also a major strategic opportunity for anyone willing to innovate and adapt.

Semantic consensus: what LLMs really value

Large language models (LLMs) do not rely on a single source, especially if it is the company’s own website or a brand-sponsored publication. They remain an important part of the company’s digital footprint, but when it comes to attributing certain qualities and attributes to a brand, AI platforms seek to confirm agreement across multiple sources. When Forbes, TechCrunch, a respected industry publication, and an analyst report describe a brand with consistent attributes—such as whether its products are durable, have an appealing design, or offer good customer service—the model interprets that convergence as a consensus of authority.

The Stacker/Scrunch research is very clarifying in this regard: a single appearance in a Tier-1 outlet produces a citation rate of approximately 7.7% in relevant prompts. But if we distribute that same story across four or five authoritative publications, the citation rate skyrockets to 34%! Same brand, same news, but four times more visibility. This demonstrates that AI systems treat alignment across multiple sources as confirmation of the truth of that information, not as redundancy.

The BrightEdge AI Catalyst analysis adds further evidence. When analyzing five AI platforms, the overlap of sources between any two of them ranges from 36% to 59%, but the overlap of cited brands is between 35% and 55%. In other words, LLMs do not agree on whom to cite, but they do agree on whom to recommend. That alignment is the clearest proof of semantic consensus.

Entities and sources

The first exercise is clear: start by defining the entities and attributes you want LLMs to associate with your brand: which category, which use cases, which adjacent technologies, which differentiators. Let us remember how advertising has always done it: Volvo → safety.

Next, audit your industry’s citation graph—that is, which sources the different models cite most for likely prompts your customers might use. Amsive’s research found, for example, that ChatGPT most frequently cites Wikipedia, Perplexity cites Reddit and YouTube, and Microsoft Copilot cites Forbes and Gartner. The mix changes by vertical: in beauty, social networks and influencers dominate; in B2B SaaS, analyst reports and the trade press carry more weight; in connected hardware, manufacturer ecosystems, niche media with specialized reviews, and community forums do most of the confirmation work.

Once you know which sources the model uses as references for your category, the strategy is clear: you must be present in those sources, described consistently and aligned with the entities and attributes you have chosen.

At Human Level, we applied exactly this approach with Signify / Philips Hue, our finalist case for Best Use of Search in Retail/Ecommerce at this year’s European Search Awards. In addition to optimizing the website at a technical, semantic, and domain authority level, we identified the most cited sites for prompts related to the smart lighting category in AI engines, discovered which attributes had the greatest influence on the purchase decision (ecosystem compatibility, design authority, energy efficiency), and aligned communications across owned and external media to reinforce those associations in the sources the models were referencing.

Digital PR gains new momentum

Corporate communications have traditionally served institutional objectives: reputation, executive visibility, crisis response, investor narrative… Driving customer acquisition was not usually part of its goals; that was marketing’s job, measured in GRPs, clicks, conversions, or GMV.

This division of tasks no longer works.

Today, LLMs turn earned media into direct training data for tomorrow’s purchase recommendations. Stacker’s research found that distributing earned media increases AI citations by an average of 239%. This is no longer a brand metric; it becomes a sales driver directly attributable to the work of communications teams.

The conclusion is obvious: at least part of any serious PR and Digital PR program must be allocated as soon as possible to organic visibility (findability). Not at the expense of institutional priorities, of course, but in parallel with them. And here, many organizations’ structures may need to overcome strong inertia to adopt the necessary change.

Note: we are not talking about “black hat GEO”

If you are thinking this sounds like LLM poisoning, this is a good time to clarify it.

“Poisoning” an LLM (LLM poisoning) means deceiving a model with manipulated or false information to alter its outputs. At Human Level, we have always advocated White Hat SEO. And we are not going to change now.

What we are describing here is the opposite: ensuring that the sources LLMs already trust (and therefore cite) describe your brand in a coherent way aligned with the positioning you have actually chosen. It is the old discipline of brand consistency applied to a new reference environment.

This recent SparkToro study by Rand Fishkin confirmed in January 2026 that individual AI answers are random, not deterministic: identical prompts return different lists and cite different sources across runs. However, the visibility share of different brands (cited by different sources each time) across many runs is stable, measurable, and shaped by the citation graph. In other words, we cannot control the LLM, but we can influence the media from which it builds its consensus.

Five steps to move forward with this model

At Human Level, we believe these are the actions we must put in place today to secure our brand visibility in AI tomorrow:

1. Map your target entity-attribute associations. What should an LLM say about you, and alongside which entities?
2. Audit the most cited sources in your category. Identify the 10–15 sources AI cites most in your commercial prompts. That should be the destination of your communications.
3. Focus on Tier-1 and vertical press. Two appearances in high-authority outlets outperform twenty in low-authority outlets. Consistent category citations typically come with 3–5 well-distributed Tier-1 hits.
4. Maintain a freshness cadence. Content updated in the last 30 days receives significantly more AI citations than content that is not modified.
5. Measure visibility share, not rankings. Run each prompt multiple times across different platforms. Track presence, sentiment, and information accuracy—not the exact position.

The future customer’s “short list” is being written today

A brand’s authority requires a considerable investment of time and money, but once built it becomes a lasting asset. The brands that start aligning their PR with their visibility strategy now will be the brands AI recommends in the future.

The “short list” is no longer only in the consumer’s mind—it has moved to LLMs. Are you already working to secure a place for your brand in them?