Custom GPTs FAQ & Glossary: Every Term Explained
Custom GPTs come with their own vocabulary, and clients (and even some agencies) often nod along through a discovery call without actually knowing what “Actions” or “knowledge retrieval” mean in practice. This glossary breaks down every term that actually matters when planning, building, or buying a Custom GPT, in plain language, with enough context to use each term correctly in a real project conversation.
Read it top to bottom for a full primer, or use it as a reference when a specific term comes up during a build. We wrote this the way we wish it existed when we started fielding client questions about “AI assistants” — most of the confusion isn’t about whether the technology works, it’s about which specific term applies to which specific tool, since “Custom GPT,” “AI agent,” and “chatbot” get used interchangeably in casual conversation even though they mean genuinely different things to build and price.
The Core Building Blocks
These are the pieces that make up every Custom GPT, regardless of what it’s used for.
- Custom GPT: a customized version of ChatGPT, configured with specific instructions, knowledge, and optional tool connections, built through OpenAI’s no-code GPT Builder.
- Instructions (system prompt): the private text that defines how the GPT behaves — its voice, its rules, its boundaries, and what it should do when it doesn’t know an answer. Users never see this text directly.
- Knowledge (knowledge files): documents uploaded to the GPT that it can reference when answering questions, such as PDFs, spreadsheets, or text files containing FAQs, policies, or pricing.
- Conversation starters: the suggested prompts shown to a user when they first open the GPT, meant to guide them toward the questions the assistant actually handles well.
- Capabilities: the built-in tools a GPT can be given access to, such as web browsing, image generation (DALL·E), or code execution.
Retrieval and Knowledge Terms
Understanding how a GPT actually “reads” its uploaded documents matters for setting realistic expectations with clients.
- Retrieval: the process by which a GPT searches its uploaded knowledge files for relevant passages before answering a question, rather than having the entire document memorized.
- Chunking: the way long documents get broken into smaller sections behind the scenes so retrieval can find the most relevant piece rather than searching an entire file at once. This is why poorly organized source documents (with information scattered rather than grouped clearly) often retrieve worse than clean, well-structured ones.
- Hallucination: when a language model generates an answer that sounds plausible but isn’t actually supported by its instructions or knowledge files — the central risk that thorough testing is meant to catch before launch.
- Grounding: the practice of writing instructions that force the GPT to answer only from its provided knowledge rather than its general training, reducing hallucination risk on business-specific facts like pricing or policies.
Actions and Integration Terms
This is where Custom GPTs move from “smart FAQ” to “connected to real systems,” and where the terminology gets more technical.
- Actions: a feature that lets a Custom GPT call an external API — for example, checking live inventory, pulling an order status, or submitting a form to a CRM — during a conversation.
- OpenAPI schema: the technical specification document that tells a GPT’s Actions feature how to talk to an external API: what endpoints exist, what data they expect, and what they return. Building this correctly is usually the job of a developer, not a marketer.
- Authentication (API key / OAuth): the credentials an Action needs to securely connect to an external system. Handling these credentials properly — never hardcoding them where a client’s whole team can see them — is a basic security requirement, not an optional step.
- Webhook: a way for an external system to send data back out when something happens, sometimes used alongside Actions in more advanced integrations (for example, notifying a Slack channel when a GPT captures a new lead).
Publishing and Access Terms
Once a GPT is built, these terms describe who can find and use it.
- Private GPT: visible only to the account that created it.
- Link-sharing: a GPT accessible to anyone with the direct link, without appearing in any public directory — the most common setting for client work.
- Workspace / Team GPT: a GPT shared only within an organization’s ChatGPT Team or Enterprise workspace, useful for internal-only assistants.
- GPT Store: OpenAI’s public directory of published Custom GPTs, searchable and browsable by any ChatGPT user — the right destination only for GPTs intended for broad public use, not most client-specific tools.
Related but Different: Assistants API and Agents
Clients and even some agencies frequently confuse these adjacent terms with a Custom GPT, so it’s worth separating them clearly.
- Assistants API: a developer-facing tool for building AI assistants that run inside a company’s own app or website, rather than inside ChatGPT’s interface. This is what’s actually needed if a client wants an embedded website chatbot, not the consumer GPT Builder.
- AI agent: a broader term for an AI system that can take multi-step actions toward a goal, sometimes with minimal human input at each step — a Custom GPT with Actions is a simple form of this, but the term also covers more autonomous developer-built systems well beyond the GPT Builder’s scope.
- RAG (retrieval-augmented generation): the general technical approach of combining a language model with a retrieval step over external documents — the same underlying idea that powers a Custom GPT’s knowledge files, and the term you’ll hear used more broadly in custom AI development.
- Fine-tuning: a separate, more technical process of retraining a model’s underlying weights on custom data, distinct from a Custom GPT’s instructions-and-knowledge approach and rarely necessary for typical agency or local-business use cases.
Business and Project Terms
Beyond OpenAI’s own vocabulary, a few practical terms come up constantly in client-facing project conversations.
- Scope (in a GPT project): the explicit definition of what a GPT is and isn’t responsible for answering — the single most important planning decision in any build.
- Escalation rule: the instruction that tells a GPT when to stop and hand off to a human rather than attempt an answer, critical for any assistant handling real customer interactions.
- Knowledge refresh: the ongoing process of updating a GPT’s uploaded documents as business information changes, usually the basis for a recurring maintenance retainer.
- Voice (or brand voice): the tone and phrasing style an assistant is instructed to use, ideally drawn from real examples of how a business actually communicates rather than a generic description.
Frequently Asked Questions
What's the difference between a Custom GPT and the Assistants API?
A Custom GPT is built through ChatGPT's no-code interface and lives inside ChatGPT; the Assistants API is a developer tool for embedding a similar kind of assistant directly into a company's own website or app.
Do I need to understand OpenAPI schemas to build a basic Custom GPT?
No. A knowledge-and-instructions-based GPT requires no technical schema work at all; OpenAPI schemas only come into play when connecting Actions to external systems.
What does "hallucination" actually mean in practice?
It's when the assistant confidently states something that isn't true or isn't supported by its actual source material — the main risk that careful instructions, grounding, and adversarial testing are designed to catch.
Is the GPT Store the same as publishing a website?
No. The GPT Store is OpenAI's own public directory inside ChatGPT; most client-built GPTs are shared privately by link rather than listed there.
What is "grounding" and why does it matter for client work?
Grounding means writing instructions that force the GPT to rely on its provided knowledge files rather than guessing from general training — it's the main technique for keeping business-specific answers like pricing and policy accurate.
Is fine-tuning something most agencies need to worry about?
Rarely. Fine-tuning retrains a model's underlying weights and requires significant technical resources; the instructions-and-knowledge approach used in a standard Custom GPT covers the vast majority of agency and local-business use cases.
Terry has 30+ years in software and SEO. He’s the founder of Salterra Digital Services and SEO Spring Training, host of the Roundtable SEO Mastermind, and lead instructor at SEO University — teaching the exact tactics his team uses on client work.
Continue learning
Ready to master this?
This guide is one lesson from the Building Custom GPTs & AI Assistants for Client course. Get every lesson, framework and checklist — plus the full 38-course catalog — inside SEO University.
Practitioner-focused training across the full digital marketing stack — from technical SEO to conversion optimization and the AI search era. By Salterra Digital Services, since 2011.