AI glossary
The words that come up when a small team starts using AI, explained without the jargon. Where a term matters to a specific piece of work, it links to the solution that uses it.
30 minutes, no cost. Bring the terms you’ve been nodding along to.
Also called: Answer engine optimization, AI search optimization
The work of getting your business mentioned and cited in AI-generated answers from tools like ChatGPT, Claude, Perplexity and Google’s AI Overviews. It overlaps with SEO, but leans more on third-party mentions, structured data and clearly stated facts about who you are and what you do.
Where it shows up: Found by AICase Study Engine
A process where an AI agent handles several steps in a row, with checks between them. For example: research a company, draft an email, wait for a person to approve it, then schedule it.
Where it shows up: Signal OutboundIntake Autopilot
Also called: Agent
An AI system that works toward a goal by taking steps on its own, such as looking things up, using tools and deciding what to do next, rather than answering a single prompt. The agents that work well for small businesses are narrow: one job, clear limits, and a person approving anything important.
Where it shows up: Signal OutboundIntake Autopilot
Also called: Copilot
An AI assistant built into the tools people already use that drafts or suggests work while a person stays in control. The person edits and approves; the copilot removes the blank page.
Where it shows up: Delivery Copilot
Also called: Meeting assistant
A tool that records or joins meetings, transcribes them and writes a summary. Granola, Fathom and Gemini in Google Meet are common examples.
Where it shows up: Call-to-ProposalDay-One Kickoff
Google’s AI-generated summaries that appear at the top of some search results. They cite their sources, which makes them one of the places AEO tries to earn a mention.
Where it shows up: Found by AI
An AI tool that responds with a direct answer instead of a list of links, such as ChatGPT, Perplexity or Google’s AI Overviews. More buyers now start their research in one.
Where it shows up: Found by AI
Also called: Application programming interface
An application programming interface is the way one piece of software talks to another. When an AI tool “connects” to your CRM, inbox or calendar, it’s usually through an API.
Software doing a repeatable task without a person, usually started by a trigger. AI extends automation to tasks that involve reading and writing language, like sorting email or drafting a reply.
Where it shows up: First ResponderDay-One Kickoff
Also called: Voice guide, Style guide
A written description of how a person or brand sounds: word choice, rhythm, tone and what to avoid. Giving AI a voice guide plus real examples is the difference between generic drafts and drafts that sound like you.
Where it shows up: Voice Engine
A final, polite sales email saying you’ll stop following up unless the prospect wants to continue. It often gets a reply because it’s honest and easy to answer.
Where it shows up: Deal Rescue
Software that holds a text conversation with a person. Modern chatbots run on large language models, and the trustworthy ones for business are grounded in approved content and hand off to a person when unsure.
Where it shows up: Always-On AnswersSetup Concierge
Also called: Churn rate, Customer churn
Customers who stop buying from you over a period of time. Churn rate is the share of customers lost in that period.
Where it shows up: Never Out of TouchRenewal Ready
Anthropic’s family of AI models and the assistant built on them. It’s the model I use most, because it’s strong at writing, following detailed instructions and connecting to the tools teams already use.
Where it shows up: Voice EngineDelivery Copilot
Also called: Agent skills
Reusable packages of instructions, examples and files that teach Claude how to do a specific task your way, such as writing in your brand voice or following your report format. Once a skill exists, anyone on the team gets the same result.
Where it shows up: Voice EngineDelivery Copilot
The amount of text an AI model can consider at once, measured in tokens. A larger context window lets a model read a long document, contract or call transcript in one pass.
Also called: Customer relationship management
Customer relationship management software tracks your contacts, deals and conversations. HubSpot, Pipedrive and Salesforce are common examples, and many AI workflows read from and write to the CRM.
Where it shows up: Fit FilterDeal RescueNever Out of Touch
Also called: Lead enrichment
Automatically adding details to a contact or company record, such as industry, size, location or job title, from outside data providers. It saves researching each lead by hand.
Where it shows up: Fit FilterSignal Outbound
Also called: Training on your data
Whether an AI provider uses the information you send it to train future models. Business and API plans from the major providers generally don’t by default, while consumer plans may, so check your plan’s terms before connecting client data.
Also called: Email deliverability
Whether your emails land in the inbox instead of the spam folder. It depends on your domain setup (SPF, DKIM and DMARC records), how much you send and how recipients respond.
Where it shows up: Signal Outbound
Numerical representations of text that capture its meaning, so software can find passages that mean similar things even when the words differ. Embeddings power semantic search and retrieval-augmented generation.
Where it shows up: Company Brain
Also called: Examples in prompts
Including a few examples of the output you want in the prompt, so the model follows the pattern. It’s one of the most reliable ways to get consistent results.
Where it shows up: Voice EngineDelivery Copilot
Further training an existing AI model on your own examples to change how it behaves. Small businesses rarely need it; clear instructions, good examples and grounding usually get there faster and for less.
Also called: GenAI
AI that creates new content, such as text, images, audio or code, rather than only sorting or predicting. ChatGPT, Claude and Gemini are generative AI tools.
Also called: Generative engine optimization
Another name for AEO, used for the work of being cited by generative AI search tools. The tactics are the same: clear facts, structured data and credible third-party mentions.
Where it shows up: Found by AI
Limiting an AI to answer from a specific, approved set of information, such as your FAQ, pricing rules or help docs, instead of whatever it learned in training. Grounded assistants can say “I don’t know,” which is what makes them safe to put in front of customers.
Where it shows up: Always-On AnswersSetup ConciergeSupport Desk Copilot
Rules and checks that keep an AI system within safe limits. Examples include topics it won’t discuss, actions that need a person’s approval, and a fallback to “I don’t know.”
Where it shows up: Always-On AnswersSupport Desk Copilot
When an AI model states something false with confidence. Grounding, clear instructions and human review reduce it.
Where it shows up: Always-On AnswersSetup Concierge
Also called: HITL
A design where a person reviews or approves AI output before it has any effect, such as sending an email or publishing a post. It’s the default in almost everything I build.
Where it shows up: First ResponderSignal OutboundDelight Trigger
Also called: ICP
A written description of the companies you serve best: their industry, size, situation and problems. The useful version comes from studying your best past customers.
Where it shows up: Signal OutboundFit Filter
A collection of documents and answers that a team or an AI assistant draws on. An AI knowledge base lets people ask questions in plain language and get answers with links to the sources.
Where it shows up: Company BrainSetup ConciergeSupport Desk Copilot
Also called: LLM
An AI model trained on large amounts of text to understand and generate language. Claude, GPT and Gemini are large language models.
Giving each lead a score based on how closely it matches your ideal customer and how ready it seems to buy, so you know who to talk to first.
Where it shows up: Fit Filter
Also called: Model Context Protocol
An open standard, introduced by Anthropic, for connecting AI assistants to tools and data such as your inbox, calendar, CRM or files. One assistant can work with many systems through a common kind of connector.
Also called: AI model
The trained AI system that does the work, such as Claude Sonnet or GPT. Different models trade off speed, cost and capability, so the right one depends on the task.
Describes an AI model that works with more than one kind of input or output, such as text, images, audio and PDFs.
Also called: NLP
The field of computing that deals with understanding and generating human language. Large language models are the current state of the art.
The instructions and content you give an AI model. A good prompt says what you want, gives context and examples, and describes what the output should look like.
Writing and refining prompts so a model produces reliable, useful output. In practice it’s mostly clear instructions, good examples and specific constraints.
Also called: RAG
A way of making an AI answer from your own documents: it looks up the most relevant passages first, then writes an answer based on them. It’s how a website assistant or company knowledge base stays on your facts.
Where it shows up: Always-On AnswersCompany BrainSetup Concierge
Search based on meaning rather than exact keywords. A question like “how do we handle refunds?” finds the refund policy even if it never uses the word “handle.”
Where it shows up: Company Brain
Detecting the tone of a piece of text: positive, negative or neutral. It helps spot happy customers to ask for referrals, and unhappy ones who need attention.
Where it shows up: Delight TriggerCustomer Voice ReadNever Out of Touch
Also called: Lead response time
The time between an inquiry arriving and your first reply. Shorter is better, because buyers often contact several providers at once.
Where it shows up: First Responder
Also called: Schema markup
Code on a web page that describes its content in a standard format (schema.org), so search engines and AI tools can read facts like your services, location and FAQs reliably.
Where it shows up: Found by AI
Standing instructions given to an AI model before a conversation starts, setting its role, rules and tone. The people using the assistant usually don’t see it.
A setting that controls how varied a model’s output is. Lower temperatures give more predictable answers; higher ones give more creative, less consistent ones.
Finding the recurring topics across many comments, such as reviews, survey answers or tickets, and counting how often each comes up. Counting per customer instead of per comment stops one loud voice from skewing the result.
Where it shows up: Customer Voice Read
Resolving a support question without a person, usually through a help article or an AI assistant. The deflection rate is the share of questions resolved this way.
Where it shows up: Support Desk CopilotSetup Concierge
The unit AI models use to measure text, roughly three-quarters of a word in English. Usage pricing and context windows are both counted in tokens.
Turning recorded speech into text. It’s the first step for call summaries, proposal drafting and voice AI.
Where it shows up: Call-to-Proposal
The event that starts an automation, such as a form submission, a deal closing, an email arriving or a date approaching.
Where it shows up: First ResponderDay-One KickoffRenewal Ready
A database that stores embeddings so software can quickly find the passages most relevant to a question. Most retrieval-augmented generation systems use one.
Where it shows up: Company Brain
Also called: Voice agent
AI that listens and speaks in real time. Small businesses use it for virtual receptionists, sales practice partners and presentation coaching.
Connecting apps so a sequence of tasks runs on its own, using tools like Zapier, Make or n8n. AI steps, such as drafting or classifying, can sit inside these workflows.
Where it shows up: First ResponderDay-One KickoffIntake Autopilot
30 minutes, no cost, no pressure.