Our new AI/Chatbot Research Tool provides you with in-depth insights into the visibility of your brands, entities and sources within the most important AI chatbots. We currently support analyses for ChatGPT, Gemini and Deepseek.
Similar to our established tools for Google, Amazon and social media, all our customers have access to the same comprehensive database. This enables you to perform precise analyses based on a broad range of information.
A huge advantage of this tool is the lightning-fast availability of results. Because all data has already been determined in advance you get your answers in a fraction of a second without long setup and waiting times. In the future, we will also integrate a history and tracking function that allows you to track changes over time.
Prompts
To provide you with meaningful analyses, we use a comprehensive dataset of 10 million prompts for each of the supported languages: German, English, French, Spanish and Italian.
We carefully select these prompts so that they represent a cross-section of actual chatbot usage in the respective countries and languages. These prompts come from various sources. These include, for example, the prompt suggestions that ChatGPT offers when you enter text, but also frequently asked user questions from Google search results and other relevant data sources.
To ensure that our data remains up to date, we regularly update our prompt database. This means that we add new prompts and remove those that are used less frequently or no longer used at all. In this way, we ensure that the data reflects current usage habits.
It is important to note that the data situation and the availability of reliable foundations for LLM/AI analyses are currently even more challenging than, for example, at Google or Amazon. However, this is a development that we are familiar with from the early days of Google analysis. There, too, data quality has gradually improved. We are confident that this will also be the case with LLM data.
Chatbot responses
In the “Search” section, we query the responses of three major, relevant LLMs for 10 million prompts per language. We currently support OpenAI ChatGPT, Google Gemini, and DeepSeek. If new, important market players emerge, we plan to include them as well.
In the responses from the LLMs, we recognise entities such as brands, people, places and more, as well as the associated source references. We prepare this data for your analysis. Similar to Google and Amazon, this enables reverse searches: you can find out which prompts mention your brand or those of your competitors.
We always query the current, relevant models. When a provider introduces a new model, we endeavour to make it available as soon as possible. We continue to keep older models available so that you can make comparisons over time.
Querying the models more frequently does not add any value, as the information in these foundation models is static. This means that multiple queries only change the wording slightly, but the underlying information remains the same. Combining LLMs and web search, as with Perplexity or ChatGPT Websearch, leads to more dynamic results. You can track these individually in your projects.
We currently support these foundation models for chatbots:
- OpenAI ChatGPT
- gpt-4o-mini
- gpt-4.1-mini
- Google Gemini
- gemini-2.0-flash-lite
- gemini-2.0-flash
- gemini-2.5-flash
- Deepseek
- v3
Position and ranking
Within AI responses, we determine the ranking of brands and other entities separately. In other words, we measure the position at which a particular brand is mentioned in the response. The same applies to all other entity types such as people, places, etc., but in this case they are grouped together.
AI Visibility Index
The calculation of the Visibility Index for brands, entities and sources basically works in the same way as you are already familiar with from Google. There are three steps:
- Collecting responses: First, we collect the responses of the LLMs for 10 million representative prompts.
- Weighting responses: In the second step, we weight these responses. It is important to consider the position of the information in the response and how often the prompt is used.
- Summarise values: Finally, all weighted values for all entities, brands and sources are summarised.
There is a separate Visibility Index for brands, entities and sources for each language and chatbot. A total of 1 million visibility points are always distributed.
The total AI Visibility Index for a brand, entity and source is then calculated by adding together the individual visibility values of the three chatbots. The position shows you where your brand ranks in the order of the most visible brands. The brand with the highest Visibility Index is therefore ranked #1, and the rest are ranked in descending order.
Frequently asked questions and answers
Why aren’t the results changing?
The chatbots use so-called foundation models from OpenAI, Gemini or Deepseek. These models work with a fixed database that is only updated when a new model is released. If you ask the same prompt multiple times, the wording of the answer may change slightly, but the facts remain the same because the underlying model remains the same. We collect new data whenever a new foundation model becomes available, for example when switching from GPT-4o to GPT-4.1.
How are positions measured?
In the chatbots’ text responses, we identify entities such as brands, places, people, sources and other types. For each of these entities, we record the position of the respective mention in the response text. The counting is done separately: for brands, only other brand mentions are counted; for all other entities, the positions of all other entities are evaluated together.
Why are brand logos sometimes missing or incorrect?
The chatbot responses contain brand names as continuous text. We analyse this text and recognise brands and other entities in the respective context. To improve recognisability, we enrich well-known brands with logos. We obtain these from various third-party interfaces. In most cases, this works reliably, but it can happen that no logo or an incorrect logo is displayed. We strive to minimise these cases and can also update logos manually. Please contact our support team for assistance.
Why is my brand not recognised as an entity?
In general, the research area of AI/chatbot analysis works by automatically querying millions of prompts from different chatbots in different languages and analysing the results for the presence of entities such as brands. If a particular brand is not recognised there, there can be various reasons for this: the simplest reason is that the brand was not mentioned in any of the chatbot responses. Even though chatbots generally have quite comprehensive knowledge, they do not know every brand and do not mention every brand. It can also happen that chatbots use a different spelling than the brand itself. It can also happen that we do not recognise a brand as an entity.
