How LLMs describe your company (and how to change it)
Every day, millions of people ask artificial intelligence systems about companies, brands, people, and organisations. They receive answers that seem objective, complete, and authoritative. In reality, they are the result of a selection and synthesis process that almost no company monitors, and even fewer are able to influence.
The question is not whether AI systems are talking about your company. It's what they are saying — and how to change it if necessary.
What happens when an LLM receives a question about your company?
The process is less mysterious than it seems, but more complex than many imagine.
When a user asks ChatGPT, Gemini, Claude, or Perplexity about a company, the system doesn't "search" for information in real time like a traditional search engine would. It draws from a knowledge ecosystem built during its training — supplemented, in the latest systems, by real-time source retrieval.
In both cases, the selection mechanism favours:
- Consistency across different sources: information that appears on multiple authoritative sources carries more weight than an isolated statement
- Structure and readability: content with clear headings, verifiable data, and precise language is more easily integrated into the summary
- Source authority: media, institutions, public databases, and Wikipedia carry more weight than the company's own proprietary content
- Distribution over time: continuous and updated signals take precedence over one-off publications
The result is a representation that reflects the company's information ecosystem — not its official communication.
Why can what an LLM says about your company be wrong?
There are two distinct mechanisms that produce incorrect representations, regardless of any corporate error.
The first is semantic misalignment: the company has online content, but it is not structured in a way that is readable by AI systems. The result is an incomplete representation — not false, but partial. Entire chapters of the corporate identity are simply not read.
The second is hallucination: the system generates false information with full confidence. It might attribute positions to the company that it never expressed, cite non-existent data, or associate the brand with events that never happened. This information appears with the same authoritative tone as the correct information — and is invisible to traditional monitoring systems.
Both mechanisms have the same reputational effect: whoever queries the system receives an image of the company that the company has not authorised and often is not aware of.
How do you measure what LLMs say about a company?
Measuring AI Reputation requires a systematic and multi-model approach.
It is not enough to query a single AI system: each LLM has different training characteristics, different weights on sources, and different synthesis logic. A stable representation emerges only from a comparative analysis across multiple systems — ChatGPT, Gemini, Claude, Perplexity, Mistral — queried with prompts calibrated to the dimensions relevant to that organisation.
The dimensions to measure are:
- Sentiment: the overall tone of the representation — positive, neutral, negative, or ambivalent
- Visibility: how often the company emerges in queries relevant to its sector
- Thematic coverage: which dimensions of the corporate identity are mentioned (innovation, sustainability, leadership, products) and which are missing
- Narrative coherence: do the different sources converge into a coherent narrative or do they contradict each other?
- Reputational risk: does negative, outdated, or potentially harmful information emerge?
The result of this analysis is a map of the distance between the real identity and the algorithmic representation — the starting point for any intervention based on the Aurelian Lab AI Reputation Framework.
What are the sources that LLMs use to build a company's reputation?
One of the most surprising elements for those involved in corporate communication is the hierarchy of sources used by AI systems.
Contrary to intuition, AI systems do not prioritise a company's proprietary content. They favour sources they consider more authoritative and independent. In practice, this means that:
- Media and news outlets carry much more weight than the corporate website
- Wikipedia — when it exists and is up-to-date — is a primary reference source for almost all AI systems
- Institutional reports and databases (Consob, Borsa Italiana, public registries) carry high weight
- Editorial and social content — including Reddit, LinkedIn, and industry forums — is integrated more than many companies imagine
- Press releases and proprietary ESG reports have limited weight unless they are picked up by third-party sources
This hierarchy has direct implications for strategy: building AI Reputation does not mean producing more content. It means building an ecosystem of external confirmations — earned media, authoritative citations, independent validations.
How to change what an LLM says about your company?
Changing a company's algorithmic representation is possible, but it requires time and method. There is no shortcut.
The process is structured into three levels, consistent with the Aurelian Lab AI Reputation Framework:
1. Make what exists readable The first intervention concerns the structure of existing content. Texts without clear headings, decontextualised data, and statements without supporting evidence are invisible to AI systems. Restructuring content into an algorithmically readable format is the starting point.
2. Build an ecosystem of external confirmations The second level concerns the distribution of evidence. Earned media, citations in authoritative sources, a presence on Wikipedia, verifiable data published by third parties — anything that reinforces and distributes the corporate narrative outside of proprietary channels.
3. Monitor and correct continuously The third level is structural monitoring. AI Reputation is not managed with a one-time intervention. It requires continuous oversight of the representations generated by the systems, with the ability to intervene quickly when distortions or hallucinations emerge.
How long does it take to change a company's representation in AI systems?
There is no single answer, as it depends on the distance between the current and target representations, the quality of the existing information ecosystem, and the organisation's ability to produce consistent signals over time.
In general, AI systems with real-time retrieval (like Perplexity) respond more quickly to new content. Systems based primarily on training (like the offline versions of other LLMs) require more time.
What is certain is that those who start first build a cumulative advantage that is difficult to overcome. Citation authority — like domain authority in traditional SEO — accumulates over time. The organisations that invest in AI Reputation today will be the ones that AI systems cite tomorrow.
Start with the basics → What is AI Reputation: definition, risks, and opportunities Discover why press releases are no longer enough → Why press releases are no longer enough See how sustainability becomes a strong signal → Sustainability and reputation in AI systems Back to the complete model → Governing Corporate Reputation in the AI era: the Aurelian Lab pillar Want to measure your AI Reputation? → Aurelian ESG Monitor · Contact me