AI Models, Infrastructure and the AI Search Ecosystem
Terms that explain how large language models work and the AI search ecosystem. Model infrastructure, bot access, content licensing and the use of AI in search algorithms are covered under subheadings.
All Terms · A–Z
You can search in English or Turkish.AI Search Surfaces and Visibility Measurement
- AI Search Engine
- Turkish: AI Arama Motoru
- Search interfaces that produce a synthesised answer instead of a list of links (ChatGPT Search, Perplexity, Gemini, Copilot).
- Google AI Mode
- Google's mode that moves the whole search into a conversational, generative answer flow.
- SGE (Search Generative Experience)
- The name of AI Overviews during its experimental period. Older sources refer to it by this name.
- Conversational Search
- Turkish: Konuşma Tabanlı Arama (Conversational Search)
- A query progressing as a dialogue that carries context rather than as a one-off search.
- Query Fan-out
- Turkish: Sorgu Dağıtımı (Query Fan-out)
- Generative search systems splitting a single user query into many sub-queries in the background and searching in parallel. The technical reason content must also answer those sub-queries in GEO.
- Follow-up Query
- Turkish: Takip Sorgusu (Follow-up Query)
- A deeper query the user asks after the first answer while keeping the context.
- Citation
- Turkish: Alıntılanma (Citation)
- A generative answer citing a site as a source. The GEO counterpart of an organic ranking.
- Citation Rate
- Turkish: Alıntılanma Oranı (Citation Rate)
- The share of answers in a tracked prompt set in which the brand is cited as a source.
- Prompt Tracking
- Turkish: Prompt Takibi (Prompt Tracking)
- Asking models a defined set of queries at regular intervals and recording the answers. The core method of GEO reporting.
Model and Infrastructure Concepts
- Token
- The smallest unit a model uses when processing text. Costs and limits are calculated in tokens.
- Context Window
- Turkish: Bağlam Penceresi (Context Window)
- The total number of tokens a model can process at once.
- Parameter
- Turkish: Parametre
- The number of weights a model learns during training. A rough indicator of capacity; it does not determine quality on its own.
- Inference
- Turkish: Çıkarım (Inference)
- A trained model producing a response to a given input. The stage where operating costs arise.
- Knowledge Cutoff
- Turkish: Bilgi Kesim Tarihi (Knowledge Cutoff)
- The last date covered by a model's training data. Information after this date only arrives through external access.
- Temperature
- Turkish: Sıcaklık (Temperature)
- A parameter that sets how random the output is. A low value produces consistency, a high value produces variety.
- System Prompt
- Turkish: Sistem Talimatı (System Prompt)
- A top-level instruction layer defining a model's role, limits and rules of behaviour.
- Zero-shot / Few-shot Prompting
- Instructing a model without examples and with a few examples. Examples markedly improve format consistency.
- Chain of Thought
- Turkish: Düşünce Zinciri (Chain of Thought)
- A model explicitly producing intermediate steps on its way to a conclusion. It improves accuracy on tasks that require multi-step reasoning.
- RLHF (Reinforcement Learning from Human Feedback)
- Turkish: RLHF
- The training stage used to align a model's responses with human preferences.
- Distillation
- Turkish: Damıtma (Distillation)
- Transferring the behaviour of a large model to a smaller, cheaper model.
- Open-Weight Model
- Turkish: Açık Ağırlıklı Model (Open-weight)
- A model whose weights can be downloaded and run on your own infrastructure. Preferred where data sovereignty is required.
- Multimodal Model
- Turkish: Çok Modlu Model (Multimodal)
- A model that can process text, images, audio and video together.
- Model Collapse
- Turkish: Model Çöküşü (Model Collapse)
- Models losing diversity and accuracy as a result of being trained mainly on AI-generated data.
- Guardrail
- Turkish: Guardrail (Koruma Bandı)
- A control layer that keeps a model's output within defined policy and brand boundaries.
- Function Calling / Tool Use
- Turkish: İşlev Çağırma (Function Calling / Tool Use)
- A model's ability to call an external tool or API while producing a response.
- MCP (Model Context Protocol)
- An open protocol that connects models to external data sources and tools in a standard way.
- Vector Database
- Turkish: Vektör Veritabanı
- A database where embedding vectors are stored and retrieved through similarity queries. A component of RAG architecture.
- Cosine Similarity
- Turkish: Kosinüs Benzerliği
- A method of measuring the semantic closeness between two vectors.
- Chunking
- Turkish: Parçalama (Chunking)
- Splitting long content into meaningful units for retrieval. Poor chunking is the most common reason the right content is not found.
Bot Access, Copyright and Content Provenance
- GPTBot / ClaudeBot / PerplexityBot
- The content collection and access bots of generative engines. They can be managed individually through robots.txt.
- Google-Extended
- Google's directive that lets permission for using content in generative models be managed separately from search crawling.
- AI Crawler Access Management
- Turkish: AI Botu Erişim Yönetimi
- Setting a policy for which generative engines may crawl your content. Blocking them also shuts off AI visibility.
- Content Licensing
- Turkish: İçerik Lisanslama (Content Licensing)
- Publishers tying the use of their content in model training or retrieval to paid agreements.
AI in Search Algorithms and Entity Management
- NLP (Natural Language Processing)
- Turkish: NLP (Doğal Dil İşleme)
- A family of methods that let machines make sense of text.
- RankBrain
- Google's first machine-learning ranking component, used to interpret queries it had not seen before.
- BERT (Bidirectional Encoder Representations from Transformers)
- Turkish: BERT
- A language model that evaluates the context of words in a sentence bidirectionally. Effective for queries where prepositions and conjunctions change the meaning.
- MUM (Multitask Unified Model)
- Turkish: MUM
- A model aimed at multilingual and multimodal understanding. Developed for complex, multi-step queries.
- Wikidata / Entity Registration
- Turkish: Wikidata / Varlık Kaydı
- Defining a brand in machine-readable entity databases. The basis of recognition in the knowledge graph and in model answers.
