Between Data and Interpretation – When Language Models Become the New Barometer of Corporate Reputation

Between Data and Interpretation – When Language Models Become the New Barometer of Corporate Reputation

Reputation is increasingly shaped by what AI systems “know” about companies. AI-based search systems such as ChatGPT, Perplexity, or Google AI don’t simply collect information; they synthesize opinions, weigh content, and thus actively influence public discourse. For communications professionals, this means that simply measuring visibility in AI models is not enough. What matters most is how a company is portrayed in AI responses—in what tone, with what level of credibility, and with what associations.

This article was written by: Dr. Lydia Prexl

More and more people are no longer relying primarily on traditional search engines for information, but are instead turning to AI language models (LLMs) for answers. However, these work fundamentally differently from Google and similar services: They do not link to individual websites, but rather synthesize information from numerous sources, reorganize it, and provide an interpreted answer. As a result, an AI response is not just information, but always an interpretation as well—and it is precisely this interpretation that is increasingly shaping perceptions of companies, brands, and topics.

Communications professionals are therefore faced with five key questions:

1. Visibility: How visible is our brand in AI-generated responses compared to relevant competitors (“Share of LLM Voice”)?
2. Accuracy: How accurately does AI portray our company (e.g., functions, services, key facts)?
3. Competitive Position: How does my company compare to competitors in terms of AI-generated responses, and what alternatives are mentioned?
4. Tone: In what tone do AI systems portray my company (positive, neutral, critical)?
5. Narrative & Role: Within what overarching narrative does my company appear in AI responses, and how consistent is this portrayal?

What Is LLM-Visibilitytools can do —and what they don’t

To systematically measure these factors, numerous specialized LLM monitoring tools are now available. In my analysis, I examined Peek AI, Otterly AI, Profound, Nightwatch, and Semrush. Despite their different modes of operation and areas of focus, these solutions essentially promise the same thing: they aim to reveal how relevant a company is in AI-powered search and response systems and to identify ways to improve that visibility.

Over the past few days, I have thoroughly analyzed and tested various LLM tracking tools. My conclusion is nuanced—especially with regard to the five key questions faced by communications professionals.

When it comes to visibility (1), these tools offer clear added value: They provide transparency regarding whether, how often, and in what contexts companies appear in AI-generated responses—across all models, from ChatGPT to Gemini or Perplexity. This enables a reliable estimate of “Share of LLM Voice.”

The competitive position (3) can also be analyzed effectively. The tools show which competitors are commonly mentioned in AI responses, what alternatives AI systems suggest, and how your own brand stacks up in comparison. These insights are extremely valuable for market and competitive analyses in the AI landscape.

In addition, some tools provide insights into sentiment (4) by classifying mentions as positive, neutral, or critical. These assessments are helpful for an initial evaluation, but they are often relatively rough and should not be confused with a comprehensive reputation analysis.

These tools are of little use when it comes to determining accuracy (2). While the tools can show how a brand appears in AI systems, they cannot indicate whether it is represented correctly. A thorough assessment of factual accuracy and communicative appropriateness still requires human expertise.

Ultimately, these tools also reach their limits when it comes to narrative and role (5). They provide metrics on visibility and tone, but offer little basis for reliable conclusions about the overarching narrative in which a company appears or the role that AI systems reproduce in the long term.

The bottom line is that LLM tracking tools are an important foundation for assessing visibility and competitive positioning within the AI ecosystem. However, when it comes to deeper issues of reputation, narrative, and the strategic role of the brand, they do not replace a strategic communications analysis—rather, they make such an analysis necessary in the first place.

What is reputation, and how can it be measured?

Which brings us to another question that has been on my mind a lot over the past few days: Can’t the language models themselves be used directly as an early-warning system for reputational risks?

Reputation describes the overall perception of a company or brand—that is, how it is viewed, evaluated, and categorized by others. The key point here is the phrase “by others”: Reputation is not an objectively measurable characteristic, but rather an external attribution. It is not something one possesses, but rather the result of collective perception.

Measuring reputation has therefore been very time-consuming and costly up to now. Established reputation measurement models, such as the RepTrak model or the Reputation Quotient, aregenerally basedon extensive, standardized surveys. These models each result in a reputation score between 0 and 100, which quantifies how positively a company is perceived by the public.

LLMs as a Potential Early Warning System for Reputation

I am well aware that language models cannot replace traditional surveys or sophisticated market research. At the same time, media-based approaches are already used today to indirectly measure reputation by analyzing the attention and tone of media coverage. Media reports thus serve as a proxy: an increase in positive tone is interpreted as an improvement, while a predominance of negative reports is interpreted as a deterioration in reputation.

Against this backdrop, I believe language models can be understood as another level of synthesis. They are trained on enormous amounts of text and are particularly good at identifying and summarizing dominant narratives. And they have real-time access to opinions and ratings from the media, social networks, forums, and review platforms. When queried systematically, they can thus reflect how public discourse and mainstream perceptions regarding a company are evolving.

Reputation Monitoring Using Language Models: A Pragmatic Approach

A key prerequisite for reliable comparisons is the use of identical prompts. What does Brand X or Company Y stand for? Which five providers are the best (or most affordable, or most innovative…) for Z? Is W a reputable company? What are three strengths of Company A? What are three weaknesses of Company B?

These are just examples—it depends on what you actually want to measure. Communications professionals need to define a few, but relevant, reputation dimensions and experiment a bit with prompts (at least that was the case for me). The rule here is: better to have fewer, but make sure they’re consistent.

I submit these prompts once a month or once a quarter to several language models. This approach becomes particularly insightful when comparing competitors. All relevant companies are queried with identical prompts, allowing me to compare relative positioning, dominant narratives, and recurring patterns of argumentation.

It’s important to note that language models don’t measure “reality,” but rather a condensed representation of public discourse. However, they do provide a good reflection of what customers, journalists, and partners see in their search queries. The result is not a substitute for traditional reputation studies. But it helps me analyze discrepancies and highlight trends—especially in a communications landscape where AI-generated responses are increasingly shaping public perception.

About Dr. Lydia Prexl

Lydia Prexl is a communications strategist with over fifteen years of experience. Since 2021, Prexl, who holds a Ph.D. in English, has been responsible for internal and external communications at the European payment service provider Unzer. Prior to that, she established the communications function at the fintech insurer Getsafe. She is also the author and editor of several books and guides on communication and writing, including *How Do Startups Communicate?*

OTHER BLOG ARTICLES