Concepts and Glossary
Concepts and Glossary
A self-contained glossary so any reader — engineer, lawyer, marketer, or AI agent — can interpret the rest of the documentation without context.
LLM and bot terminology
LLM (Large Language Model) — Generative AI systems such as ChatGPT, Gemini, Claude, Perplexity, and Copilot.
LLM bot / LLM crawler — An automated agent operated by an LLM provider that visits public web pages to collect content for training, indexing, or live retrieval. Examples: GPTBot (OpenAI), ClaudeBot (Anthropic), PerplexityBot (Perplexity), Google-Extended (Google).
User-Agent — An HTTP header that identifies the requesting client. LLM bots declare themselves here.
Verified LLM bot — A bot whose User-Agent matches Maya's curated allowlist. Maya only ingests traffic from verified bots.
RAG (Retrieval-Augmented Generation) — Pattern in which an LLM retrieves fresh content at query time and uses it to construct an answer. Citations come from this layer.
Grounding — The mechanism by which an LLM uses external content (your pages) to substantiate its answer.
Maya-specific terms
Prompt set — A curated list of queries Maya runs against LLMs on a defined schedule. Example: "What's the best CRM for a 20-person sales team?"
Mention — An instance of a brand appearing in an LLM's response.
Mention rate — Percentage of prompts in a set where the brand is mentioned.
Mention position — Where the brand appears in the LLM's response (first, mid-list, last).
Citation — A specific URL the LLM links to as a source.
Citation rate — Percentage of prompts in a set where one of the brand's URLs appears as a citation.
Markdown rendering — Serving a markdown-formatted version of a page when, and only when, a verified LLM bot requests it.
Source-side filter — A filter that runs inside the brand's perimeter, retaining only allowlisted bot traffic and stripping any sensitive fields before transmission to Maya.
Allowlist (data) — The explicit list of fields Maya will accept. See Data Minimization.
Denylist (data) — Categories Maya explicitly will not accept. See What Maya Does Not Collect.
Architectural terms
BFF (Backend for Frontend) — An architectural layer that sits between the frontend and core backend services, often used as the canonical place to enforce policies and produce logs.
Endpoint pull — Integration pattern in which Maya calls a brand-hosted endpoint over a mutually authenticated TLS tunnel, rather than receiving a batch upload.
Batch upload — Integration pattern in which the brand POSTs a filtered NDJSON batch to Maya's ingestion endpoint on a schedule.
Feature flag — A runtime switch used to enable or disable a capability without deployment. Maya recommends gating markdown rendering behind a feature flag for safe rollback.
Reversibility — The property that any Maya-recommended change can be backed out without lasting effect on production. Every recommendation in this documentation is reversible.
Compliance terms
Data Controller — The party that determines the purposes and means of processing personal data. For brand engagements, this is the brand.
Data Processor — The party that processes personal data on behalf of the controller. Maya operates as a processor for any data covered by the DPA.
KVKK — Türkiye's data protection law (Kişisel Verilerin Korunması Kanunu, Law No. 6698).
GDPR — The EU's General Data Protection Regulation.
DPA (Data Processing Agreement) — Contract that governs the controller / processor relationship.
DPIA (Data Protection Impact Assessment) — A formal assessment of risk to data subjects from a planned processing activity. Maya assists with DPIAs when requested.
PII (Personally Identifiable Information) — Information that identifies, or could reasonably identify, a natural person.
Sub-processor — Any third party engaged by the processor (Maya) to perform processing activities. Sub-processors are listed in Annex B of the DPA.
Zero-retention endpoint — A model provider endpoint configured contractually so that prompts and completions are not retained for training or service improvement.
Process terms
Pilot — The initial deployment phase, during which Maya delivers measurable insights for a single brand or scope before expansion.
Onboarding — The process of provisioning a Maya account, loading the prompt set, connecting Search Console / GA, and standing up the log integration.
Insights report — Periodic deliverable summarizing bot crawl behavior, mention rates, and citation patterns, with prioritized recommendations.
Backlog grooming — The product process by which engineering work (e.g., markdown rendering implementation) is reviewed, sized, and prioritized.
Change management — The brand's internal process for approving and tracking infrastructure changes. Maya integrations always follow the brand's change management.