ChatGPT is still commonly described as a chatbot. Gemini and Claude are usually called AI assistants. Perplexity is positioned as an AI-powered search engine, while Microsoft Copilot is primarily presented as a productivity assistant.
Each of these descriptions is valid. The problem begins when one of them is treated as a complete definition of the product.
Modern AI systems combine capabilities that once belonged to separate categories of software. They hold conversations, retrieve current information, analyse documents, write text and code, conduct research and, in some cases, use tools to carry out a sequence of actions.
Describing these systems with a single label is becoming increasingly difficult.
One Product, Multiple Functions
Product positioning usually emphasises either the system’s primary role or the way users interact with it. Its actual capabilities often extend far beyond that initial category.
| Platform | Usually described as | Also performs |
|---|---|---|
| ChatGPT | Chatbot | Search, writing, coding, research |
| Gemini | AI assistant | Search, research, productivity tasks |
| Claude | AI assistant | Writing, analysis, coding, research |
| Perplexity | AI search engine | Synthesis, research, conversation |
| Microsoft Copilot | Productivity assistant | Writing, coding, search |
| Grok | AI assistant | Search, analysis, coding |
| DeepSeek | AI assistant | Reasoning, coding, research |
Any table of this kind inevitably simplifies the products it describes. Its purpose is to show how extensively their functional boundaries now overlap.
ChatGPT, for example, can respond conversationally, search the web or carry out a more extensive research process. OpenAI describes deep research as an agentic capability that searches, analyses and synthesises information from multiple online sources.
The interface remains familiar, but the function being performed can change from one interaction to the next.
Different Dimensions, Not One Label
Much of the confusion comes from the fact that chatbot, assistant and search system answer different questions.
Chatbot and assistant describe different aspects of the product: the first points to the conversational interface, while the second suggests the role the system is expected to play for the user. Search system belongs to another level altogether, describing the retrieval, selection and presentation of information.
A conversational interface may remain unchanged while the process behind it varies considerably. A straightforward request may be handled using the model’s existing knowledge, while a question about recent events may require access to web sources. In deep research, retrieval becomes part of a broader process involving the evaluation, analysis and synthesis of information.
As discussed in our article What Is AI Search?, AI Search is better understood through the function being performed rather than through the appearance of the interface alone. The same principle may apply to modern AI systems more broadly.
Agentic capability introduces another dimension: the degree of autonomy with which a system can select actions, use tools and work towards a defined objective. An AI agent therefore does not sit neatly beside chatbot or search system as an equivalent product category.
Agents can be built on different models and used for research, search, coding or operational tasks. Agentic capabilities can also be integrated directly into an existing product, as deep research has been integrated into ChatGPT.
A modern AI system can therefore be examined through several dimensions:
- its interface;
- its role;
- the function it is performing;
- its degree of autonomy.
A similar multidimensional approach appears in more formal classification efforts. The OECD Framework for the Classification of AI Systems, for example, evaluates AI systems across several dimensions, including their context, inputs, models, tasks and outputs.
These dimensions provide a practical way to describe how an AI system behaves in a particular context.
What This Changes for AI Search
For AI Search, this distinction has direct methodological consequences.
The statement that a business is “visible in ChatGPT” sounds clear, but it reveals very little about how that visibility occurred.
Visibility can arise in several ways. A company may first be found through integrated web search, while one of its pages is later used as a cited source. The brand may then appear in the answer, be presented as a suitable option, or ultimately be selected from a group of alternatives.
These are five distinct forms of AI Search Visibility:
- discovery — whether the system can find the company;
- citation — whether it uses the company’s content as a source;
- mention — which queries lead to the brand appearing in an answer;
- recommendation — under what conditions the company is presented as a suitable option;
- selection — whether it is ultimately chosen from the available alternatives.
One appearance in ChatGPT does not demonstrate consistent visibility across all five. The same applies to Gemini, Claude, Perplexity and other AI systems.
An AI Search Audit should therefore evaluate these forms separately. Without that distinction, an audit can easily become a collection of arbitrary prompts followed by a broad conclusion such as “the brand appears in ChatGPT”. That result is difficult to interpret, compare or translate into an optimisation strategy.
Why This Matters Beyond Product Labels
We explored a similar issue in article Why the European Commission Wants Google to Open Android and Google Search.
That article examined how the European Commission’s measures concerning Google treated competing AI products across more than one functional context. A system presented to users as an assistant could also compete with search services for access to data, device functionality and critical points of user interaction.
For AI Search, the practical consequence is clear. An audit built around a single visibility score will conceal the differences between being found, cited, mentioned, recommended and selected. A more useful framework measures those outcomes separately and shows where visibility is gained, where it is lost and which part of the process requires optimisation.
That shift turns AI Search auditing from a collection of prompt tests into a structured analysis of how a system moves from discovering information to choosing what appears in the final answer.
Resources
- OpenAI, Introducing Deep Research
An overview of deep research as an agentic capability for multi-step online research, analysis and synthesis. - OECD, OECD Framework for the Classification of AI Systems
A multidimensional framework for describing AI systems according to their context and characteristics. - Anthropic, Building Effective AI Agents
A practical discussion of agentic systems, workflows, tool use and different degrees of autonomy. - European Commission, Commission Provides Guidance to Google for AI Interoperability on Android and Sharing of Google Search Data
