Definitions for the terms used across this site's training courses, its audit methodology and the two regulations every foundational programme is built against.
No term matches that. Clear the search to see them all.
The difference between how accurately AI and experienced people make the same call. The site's task and decision article reports AI at around 65% in ambiguous, judgement-heavy situations against roughly 85% for experienced humans, narrowing to 75 to 80% against 85 to 90% on structured predictions such as candidate shortlisting.
A vendor relabelling an existing chatbot or automation tool as an AI agent without adding any new capability. The phrase was coined by industry analysts for products rebranded as agentic on the strength of the label alone.
The approach rather than the individual system: designing work so that AI carries out a sequence of steps on its own instead of answering one prompt at a time. Adoption that is early and specific to one workflow consistently outperforms broad, general rollout.
How far AI is actually used in daily work, as distinct from how many tools the business has bought. Organisations with formal AI training programmes achieve 2.3 times faster adoption, according to the training article.
A system built to carry out a sequence of steps on its own, calling tools and making decisions along the way, rather than producing a single response to a single prompt. Whether a workflow needs to be an agent, or something simpler, is a question the identify and test steps of an audit answer.
The context you give an AI tool before it starts, set out the way a new hire's brief would be. Incomplete context, not poor wording, was linked to 72% of the iteration cycles in the study the site cites.
A page named as a source inside an AI-generated answer, on Google or in an assistant such as ChatGPT, Gemini, Copilot or Perplexity. It is tracked separately from rank position, because the two no longer move together.
The AI-generated answer Google places above the ordinary search results. It answers the query directly, which is why more searches now end without a click going anywhere, and it names the pages it drew on as sources.
Turning recorded or live speech into text with an AI speech-to-text model. In Microsoft Teams it is switched on during a meeting, and the transcript is what Copilot's meeting recap is built from.
Links pointing to a page from other websites, long a signal in conventional search. The site's citation article reports that brand mentions correlated roughly three times more strongly with AI visibility than backlinks did, at 0.664 against 0.218.
Searches that include a business's name. A generic query is the kind an AI Overview is most likely to answer completely on its own, so branded search is one of the places a click still survives.
A tool you hold a conversation with, answering one prompt at a time. Any chatbot rests on three pillars (instructions, a knowledge base and tools), and what makes something an agent rather than a chatbot is that it carries out a sequence of steps on its own.
OpenAI's generative AI assistant, used as a destination of its own rather than reached from inside the applications a business already runs. Four courses in the catalogue are built on it, two of those on ChatGPT Enterprise.
Anthropic's AI assistant, covered in the foundational courses as one of the leading chatbots alongside ChatGPT, Gemini and Copilot. It runs as a service on the provider's infrastructure rather than on hardware you control, which is the distinction that decides where obligations attach.
The plan for what a business publishes, for whom, and how it will know whether the content worked. The site's articles say AI-generated answers move where the effort goes: towards depth on specific questions, structure a model can extract and evidence stated plainly.
How much text an AI model can take into account at once, measured in tokens: the conversation so far, any documents supplied and the answer being written. A longer window lets one conversation run longer and lets the model work across longer documents.
The baseline AI chat that comes with eligible Microsoft 365 licences. It answers from the web and from content a user supplies, while automatic access to emails, files, meetings and calendars through Microsoft Graph comes only with the paid Microsoft 365 Copilot add-on.
A version of ChatGPT set up for one recurring job, with its own instructions and, where needed, its own documents. It lets repeatable work be handed to a ready-made assistant instead of rewriting the same prompt each time.
Standing rules you set once so an AI tool applies them to every request, rather than restating the same context each time. They are the stable half of personalisation; memory is the remembered half.
The slow slide in an AI tool's output quality, access and rules once nobody is looking after it. Unowned AI rarely fails loudly. It drifts.
The authority signals search engines describe for judging a source. The AI Overviews article spells them out as demonstrated experience, subject-matter expertise, authoritativeness and trustworthiness, and they matter more, not less, when an AI model has to choose which few sources to attribute an answer to.
Adjusting an AI model by training it further on a set of examples, so that it writes, formats or classifies in a consistent way afterwards. It solves a behaviour problem rather than a knowledge problem, and what it learns is absorbed into the model itself.
Google's AI assistant, covered in the foundational courses as one of the leading chatbots alongside ChatGPT, Claude and Copilot. The advanced programme goes further and works through its agents.
AI that produces new text, images or other content in response to a prompt, rather than only classifying or retrieving what already exists. It is the foundational subject of every course in the training catalogue.
Writing and structuring content so that AI answer engines choose it as a source when they assemble a response. It builds on traditional SEO fundamentals rather than replacing them.
Tying an AI system's answers to a specific set of source material, usually an organisation's own documents, rather than letting it answer from its training alone. Grounding answers in your own documents is a stated outcome of the ChatGPT programmes, and which content a tool is grounded in is a distinction the Copilot Chat course works through directly.
The limits built into an AI workflow that set what it may do on its own and when a person steps in. The site's audit method names oversight, logging, escalation and the point at which a person takes over, designed into the build rather than added once something has gone wrong.
A confident, plausible-sounding answer from an AI system that is factually wrong. Recognising one, and checking for one, is covered in every foundational course this site runs.
The technology behind tools such as ChatGPT and Copilot: a model trained on a very large amount of text to predict and generate language. It is the basis both for the courses we teach and for the workflows an audit identifies.
What an AI tool keeps from one conversation to the next, so it does not have to be told the same context again. It is the remembered half of personalisation, against custom instructions, which are the stable rules.
Microsoft's generative AI assistant, reached from inside the applications a business already runs rather than as a separate destination, and grounded in an organisation's own Microsoft content through the Microsoft Graph. Several of the core AI courses in the catalogue are built on it.
Mistral's AI assistant, and one of the three named tools this catalogue teaches with a course of its own alongside ChatGPT and Microsoft 365 Copilot. It is also one of the leading chatbots the foundational courses compare, and the advanced programme works through its agents.
An open standard for linking AI assistants and agents to other systems, so they can read from and act in tools such as document storage, email or a business application. A connection built on it is often called an MCP server.
A model that works across more than one kind of material (text, image, video and voice) rather than being built for one. Choosing between a broad multimodal model and a specialised one for a given job is the judgement that matters.
AI that runs on infrastructure the organisation itself owns and maintains, rather than on a cloud platform run by a provider. Its cost is front-loaded into hardware, licensing and the staff to run it, and it keeps processing under the organisation's direct control.
An AI tool that answers a question with a synthesised response and cites the sources it drew on, rather than returning a list of results. It is one of the chatbots the courses cover, and one of the systems whose citations decide whether a business is named in an AI answer at all.
The instruction a person gives an AI system to produce a result. Writing effective prompts is one of the practical skills every core AI course teaches.
Structuring a prompt deliberately, rather than typing the first phrasing that comes to mind, so the result is closer to what is actually needed. Taught as a practical skill in the foundational courses, not as theory.
A kept set of prompts that have already been shown to work, reused instead of rewritten each time. It is the move from writing prompts to reusing reliable tools, and it is how a repeated task costs less the second time.
What happens when the model behind an AI answer breaks one search into several related sub-questions and picks a source for each, instead of summarising a single top-ranking page. A page can lose the main query and still be cited through one of the sub-questions.
An AI model that works through a problem step by step before it answers, rather than replying straight away with the first likely answer. It does not know more than a fast model; it spends longer thinking, which makes it slower but better at complex tasks.
Giving a model a set of documents to search at the moment it answers, so the reply is drawn from that material rather than from its training alone. It is the mechanism behind a knowledge base, and it is taught under that name on the AI Agents in Action day.
The practice of improving a page so that search engines can find it, rank it and show it to people searching. It is still the foundation, but AI-generated answers have changed what a good result looks like.
How much specific, quantitative data a page carries compared with general assertion. In the peer-reviewed study the site's citation article describes, adding statistics produced the largest single gain of nine techniques tested, a 41% increase in AI visibility.
Content laid out so a reader or an AI model can take an answer from it as written, with a direct answer near the top of a section and clearly labelled lists, steps, tables or question blocks. The site's articles treat it as what decides whether a trusted page actually gets used in an AI answer.
Building software by describing what you want to an AI tool in plain language and accepting the code it writes, often without reading it. It suits prototypes and small internal tools with no sensitive data, and needs review by someone qualified for anything that handles logins, personal data or payments.
Systematically unfair results from an AI system, such as screening out qualified candidates. In hiring it is a named concern, and recruitment and candidate-evaluation AI is classified as high-risk under the EU AI Act.
The rules an organisation sets for how its staff may use AI: which tools are approved, where third-party tools stop, who stays responsible for a result and how it is verified. Covered governance-first in the foundational programmes rather than bolted on at the end.
A failure of an AI tool in use, such as an assistant sending wrong information to a client or surfacing something it should not. Who is told, and in what order, should be agreed before it happens rather than worked out under pressure.
The working understanding of AI that staff need to use it safely in their own jobs. Built around the specific workflows staff already do rather than as generic AI literacy content, because that relevance is where the largest effectiveness gain comes from.
One named person per AI tool who can answer four questions: what it is used for, who can use it, what data it can reach and when it was last checked. They do not have to do the technical work. They have to be the person the question goes to.
A single list of the AI tools actually in use across a business, who uses each one, what for and on what kind of data. It includes tools staff pay for themselves and AI features switched on inside software the business already has.
The record an AI system or workflow keeps of each run: the input, what the model returned, what was approved and by whom, and what changed downstream. It is what lets any single run be reconstructed after the fact.
A rule that forbids staff from using AI tools for work, either outright or tool by tool. A blanket ban removes the approved route but not the task, so use tends to move onto personal accounts the business cannot see.
Using AI to rank or filter job applicants so that a person can decide who goes forward. The site places it at the Recommend rung of the delegation ladder, with a person accountable for the final call.
Sorting information by how sensitive it is, so that the rule for AI use is drawn by the kind of data and not by the name of the tool. The site's customer data article uses three tiers: green for any tool, amber only for an approved business tool, and red kept out of general AI tools.
How personal data may be collected, used and stored wherever an AI system touches it. Taught as a set of practical choices rather than as law, and checked alongside the EU AI Act at an audit's Identify step.
Where an AI tool stores and processes a business's data, such as inside the EU or in another region. Providers state it per product and per plan, so EU residency cannot be assumed across every tool carrying the same name.
The EU AI Act role of an organisation that uses a high-risk AI system, as opposed to building one. A deployer has duties of its own, and a vendor handling compliance does not remove them.
The duty the EU AI Act places on a high-risk AI system to be documented and to have a risk management process. It attaches to the system and its risk, whichever supplier or platform runs it.
The European Union's regulation governing how AI systems may be built and used, with obligations that scale to the risk a system carries. Assessed inside every audit's Identify step, and built into the foundational training programmes rather than added as a separate module.
The EU's data protection regulation, covering how personal data may be collected, used and stored. Relevant wherever an AI system touches personal data, and checked alongside the EU AI Act at the same Identify step.
A system in the EU AI Act's higher risk classification. The Act's obligations attach to the system and its risk rather than to where it is hosted, so a system in this class still needs a risk management process and documentation whichever way it is run.
The date from which the EU AI Act's obligations for high-risk AI systems apply. The site's review of the Act reports that the Digital Omnibus moved it from 2 August 2026 to 2 December 2027 for stand-alone systems under Annex III, and from 2 August 2027 to 2 August 2028 for systems embedded in regulated products under Annex I.
A set-up in which a named person checks an AI result before it reaches a client, a decision or a public output. The check has to name who reviews, at what stage and against what standard.
AI used in hiring, such as targeting job adverts, filtering CVs or scoring candidates. The EU AI Act classifies these uses as high-risk, whichever supplier or platform runs them.
A date fixed in advance on which an AI tool, or the rule that governs it, is checked again. The site's ownership article puts it on each tool's one-page record so that the check does not depend on anyone remembering.
Sorting each AI use case by what happens if it goes wrong, who is affected, and whether it touches personal data, a regulated decision or client-facing output nobody checks. It is what makes a policy proportionate rather than a blanket rule.
AI tools and personal accounts that staff use for work without the business having approved or reviewed them. It is usually a bigger exposure than the tools an organisation bought on purpose, because a written policy rarely covers it and nobody can see it.
A diagnostic, not a named product: one five-step methodology (map your processes, document how they run, identify the opportunities, test a prototype, build and scale) applied to an organisation's own processes, run the same way for an SME and for an Enterprise.
A single, specific process rebuilt around AI, with a clear input, a clear output and a defined point where a person reviews the result. It is one of the three things this business sells, and the one that follows an audit.
Building a workflow directly on a vendor's API, the published interface one system uses to call another, rather than assembling it in a no-code builder. It costs more to build first, because nothing exists until it is written, and then runs closer to the vendor's raw API cost.
Handing a repeatable piece of work to a system instead of a person. An audit separates the processes worth automating from the ones that should be left alone, and the AI and automation systems themselves are developed after that audit.
A training programme built from an organisation's own processes and documents, rather than a published outline, scoped in a conversation.
The argument for investing in an AI project, set out so a finance function can judge it. A strong one names the weakest gap, attaches a plan for closing it and ties the spend to a defined business outcome.
What an engagement is meant to leave behind: less repetitive work, with EU AI Act and GDPR obligations already built into how the work runs rather than checked after it. It is the end of a sequence that starts with training and an audit, and it is reached one proven workflow at a time rather than in a single rollout.
A training session that runs to a published outline, rather than being built from scratch. Six of the courses in the catalogue are core courses.
Whether the data an AI project depends on is governed, reachable and good enough to use. It is one of the most common weaknesses in an AI project, and one of the most directly fixable.
Handing a call that carries judgement to a system, such as which candidate gets a callback or which claim is flagged as fraud. Getting one wrong has a cost a checked task does not, so automating it transfers accountability and needs to be deliberate.
A four-level test that sorts each step of a workflow into Assist, Recommend, Execute or Never delegate. The level depends on volume, consequence, reversibility, how clear the rules are and whether the decision needs legitimacy a model cannot supply.
The one way to start an audit or a bespoke training programme. Nothing is quoted or scheduled before it; it establishes which processes are in scope and who needs to be in the room.
The larger of the two segments this business describes a client by, alongside SME.
What a workflow does when a case falls outside what it was built for: a malformed input, a low-confidence result, a connected system that is down. The right answer names an owner, a queue or a notification, and says whether the action is reversed or held.
The person inside an organisation who owns an AI project's day-to-day progress. With a named executive sponsor and a defined success metric, it is one of three traits shared by projects that move into production.
One round of asking an AI tool again, or editing what it returned, before the result is usable. In the study the site's briefing article cites, the average fell from 3.8 cycles to 2.0 once people assembled structured context before prompting.
How much of what staff learned in training they still remember and use afterwards. The site's training article treats it as the figure worth tracking, since completing a course and being able to use what it covered are not the same thing.
Building a workflow in a visual builder from a library of pre-built connectors, rather than writing code against each system. Platforms of this kind charge by usage, so the cost grows with every run for as long as the workflow exists.
A general-purpose AI product built for many businesses at once and bought as a subscription, as opposed to a custom workflow built against one organisation's own process. It suits a process shared across most businesses in a sector.
The third step of an audit: documented processes are assessed for what AI can do reliably against them and, as importantly, what it cannot. Each candidate is weighed on how often it repeats, what the manual version costs in effort, whether the data it needs exists in a usable state and what regulatory exposure attaches to it. What comes out is a prioritised roadmap, each opportunity carrying the reason it sits where it sits.
The point at which what a prototype proved is built for daily use, with oversight, logging, escalation and the moment a person takes over designed in. It is step five of the audit, Implement and Scale, and it happens one proven workflow at a time.
The time or cost saved when AI is used on a piece of work, counted across the whole cycle from start to accepted result. It only exists if checking, fixing, re-prompting and handoff together take less time than the task did before.
A working build of the highest-priority AI or automation opportunity, tested with the people who will use it on suitable data or realistic samples to judge accuracy, usability and benefit. It is step four of the audit's five steps, and its result decides whether the next step happens.
What an AI tool or workflow returns against what it costs, usually counted in time saved, cost reduced or revenue gained, and the time it takes to pay back. It can only be shown against a baseline taken before the tool arrived.
Whether a decision or action can be undone at low cost if it turns out wrong. A routing choice or a password reset can be; a dismissal, a declined application or a public commitment made in a client's name cannot.
One of the two segments this business describes a client by, alongside Enterprise, used instead of naming the client on the case studies this site publishes and everywhere else a client's size matters.
Handing a repeatable step with a right answer that can be checked to a system, such as formatting a report, routing a ticket or resetting a password. Once the rules are set, a step like this is safe to hand over completely.
A one-hour meeting about a week before AI training, run by the manager rather than the trainer. It names the tasks in scope, asks each person to bring one real example and says what is not allowed near the tools.
The full cost of an AI tool or workflow over its working life, not only the subscription or the build invoice. For a workflow it covers model usage, platform and hosting, human review, maintenance and governance.
Training existing staff to use AI tools well in the work they already do. The site's training article finds that programmes built around real workflows do better than generic content and better than staff teaching themselves.
A single, narrow AI workflow with one purpose, a defined input and output and a point where a person reviews the result. Every step inside it has been sorted for how much authority it gets, rather than approved as one block.
The stage between naming an opportunity and building anything: one process redrawn end to end, with its input, its output and the point where a person reviews the result all settled before a prototype exists. Scope is the decision that determines the result, which is why it is fixed at that stage rather than discovered during the build.