AI PM WEEKLY AI PM Weekly

Glossary

The AI terms you keep hearing.

Every word, phrase, model, mode, limit, and acronym you bump into at work. Each row tells you what tools the term shows up in, what it means, and the one thing that matters about it for your work. Click any column header to sort. Use the search box or category filter to narrow the list.

Definition Why this matters
Tool All Tool Think of the tool as the app you log into. ChatGPT, Claude, Copilot, Gemini, Perplexity, NotebookLM. Each one is a wrapper built by a single company (OpenAI, Anthropic, Microsoft, Google, Perplexity, Google) around one or more models. Vendors compete on the wrapper, the interface, the integrations, and the price, more than on the brain underneath. Copilot inside Word feels different from ChatGPT in a browser even when similar models are running underneath. Pick the tool that fits where you already work, not the one with the loudest ad campaign.
Model All Model Think of the model as the actual brain doing the thinking. The tool (ChatGPT, Claude, Copilot) is the wrapper around it. The model (GPT-5.5, Claude Sonnet 4.6, Gemini 3 Pro) is what is inside answering you. Two people both using "ChatGPT" can get very different answers because one is on a free lightweight model and the other is on a paid flagship. When someone shares a great result, ask which model. When you upgrade your plan, you are really paying for access to a smarter model, not a different app.
Chat Mode All Chat Mode Think of a chat mode as a switch inside the same tool that changes how the AI thinks. Fast mode answers in two seconds. Thinking mode (Claude’s Extended Thinking, ChatGPT’s Thinking, Gemini’s Deep Think) takes a minute and reasons step by step before answering. For "summarize this email," fast mode is right. For "find the flaw in this contract clause," fast mode will miss it and thinking mode will catch it. Picking the wrong mode is the number one reason people say "AI gave me a dumb answer." It was not dumb. You used the speed-reader when you needed the analyst.
Limit All Limit Think of a limit as a ceiling the vendor puts on what you can do per message, per hour, or per day. How many words you can paste in. How many messages you can send before being cut off. How many files you can upload at once. Limits are why your prompt that worked Tuesday "stops working" Thursday. You hit a daily cap, or the vendor quietly lowered it. Free plans have tight limits designed to push you to paid. Knowing the limit for your tool tells you when to break a big task into smaller chunks instead of staring at an error message.
Feature All Feature Think of a feature as a button, panel, or capability inside one of the tools. Artifacts is a Claude feature. Audio Overview is a NotebookLM feature. Action chips are a Claude feature. Each one is a specific thing you can click, not a chat mode or a model. Features are how vendors differentiate. Same model under the hood, different feature wrapped around it, and suddenly one tool is great at podcasts and another is great at long writing. Knowing the feature name lets you ask for it directly instead of describing it and hoping the AI guesses.
PM Term All PM Term Think of PM terms as the project-management vocabulary that shows up in AI work: stakeholder, transcript, RAID log, executive summary. Not AI-specific, but used so often in our examples that they deserve a definition. AI prompts get sharper when the PM word is right. “Summarize this for the stakeholder” tells Claude something different than “summarize this for the team.” Knowing the PM term you mean is half of writing a clear prompt.
AI Concept All AI Concept Think of these as the under-the-hood ideas that explain why AI behaves the way it does. Tokens, context windows, hallucinations, embeddings, temperature, grounding. You do not need them to use AI. But knowing four or five explains 90% of the weird stuff. When Copilot "forgets" the start of your meeting transcript, that is the context window concept. When two people ask the same question and get different answers, that is the temperature concept. Learning a handful of these turns AI from a magic box into a machine with predictable rules you can work with.
Acronym All Acronym Think of acronyms as the three- and four-letter shorthand AI people throw around. LLM, RAG, API, GPT, MCP, SLM. Most of them name a thing you already understand once you unpack the letters. AI conversations move fast because everyone shortcuts to acronyms. Knowing five of them (LLM, RAG, API, GPT, prompt) is enough to follow any AI article, podcast, or vendor pitch without nodding politely while feeling lost.
ChatGPT ChatGPT Tool Think of ChatGPT as OpenAI’s flagship app: the one that started the consumer AI race in 2022. Free, Plus, Pro, and Business tiers. Most people who say "ChatGPT" mean the free or Plus version they use in a browser. When a coworker says "use ChatGPT," ask which version. The free tier runs a smaller model, has smaller limits, and switches off features the Plus tier keeps on. The same prompt can produce a brilliant answer on one tier and a flat one on another.
Claude Claude Tool Think of Claude as the careful writer in the room. Made by Anthropic, lives at claude.ai. Free and Pro tiers. Known for handling long documents, writing in a natural voice, and pushing back when something seems off. When you paste a 60-minute meeting transcript and want a clean executive summary, Claude usually wins. In our Issue 01 testing, it produced the most readable status report on the first try. If your work involves long documents and careful prose, this is the one to learn first.
Copilot (Microsoft) Copilot Tool Think of Copilot as Microsoft’s AI built into the apps you already use. There are actually two products: the free Copilot at copilot.microsoft.com, and the paid M365 Copilot that lives inside Teams, Outlook, Word, and Excel. They are not the same thing. Free Copilot caps every message at 10,240 characters, which a real meeting transcript blows past in one paste. The paid M365 Copilot can see your calendar, your inbox, and your files, which is powerful and worth knowing about even if you do not use it.
Gemini Gemini Tool Think of Gemini as Google’s AI that already lives inside Google Docs, Gmail, Drive, and Meet. Free and AI Pro tiers, at gemini.google.com. It is the only chat tool that already has your Google Meet transcripts, your Gmail threads, and your Drive files at hand without any setup. If your team runs on Google Workspace, this is the lowest-friction AI you can hand someone today.
Perplexity Perplexity Tool Think of Perplexity as the AI that always shows its work. Every answer comes with numbered citations to live web pages you can click and verify. When you need a real link to back up a claim in a board memo or a vendor pitch, Perplexity gives you the link instead of making one up. It is the safest tool to use when "where did you get that number?" is a question someone will ask.
NotebookLM NotebookLM Tool Think of NotebookLM as a private notebook built from documents you upload. It only answers questions from those files. Nothing else. Made by Google. Free at notebooklm.google.com. It refuses to make things up. If the answer is not in your uploaded files, it tells you so instead of inventing one. That is gold for legal research, policy review, or any work where a hallucinated answer would be a fireable offense.
Notion AI Notion AI Tool Think of Notion AI as the assistant baked into Notion that summarizes pages, drafts new content, and answers questions about your workspace, without you ever leaving Notion. If your team already lives in Notion, this is the lowest-friction way to use AI on the docs you already have. No copy-pasting between tabs. The downside: it only sees what is in your Notion, not the rest of your company’s knowledge.
Instant ChatGPT Chat Mode Think of Instant as ChatGPT’s sprinter. Answers in two to five seconds. The model picks the fastest path and runs. For "rewrite this email" or "what does this acronym mean," Instant is the right call. You will burn months of life if you wait for Thinking mode on questions Instant can finish before you blink.
Thinking ChatGPT Chat Mode Think of Thinking mode as ChatGPT slowing down to actually reason. It writes out its steps before answering. Takes 30 seconds to 2 minutes per response. This is the mode you reach for when the answer needs to be right the first time. Finding the flaw in a contract clause. Catching the bad assumption in a budget. Reviewing a hard memo before it goes to the VP. The wait pays for itself.
Pro Mode ChatGPT Chat Mode Think of Pro Mode as Thinking mode’s deluxe sibling. Paid ChatGPT tiers only. Takes minutes per response and produces the most careful, deepest answer the product can give. Worth the wait once or twice a week on truly high-stakes work: a board summary, a competitive teardown, a strategic recommendation that will be remembered. Overkill for daily work.
Deep Research ChatGPT, Gemini, Perplexity, Claude Chat Mode Think of Deep Research as turning the AI into a junior analyst. You give it a question, it spends 5 to 30 minutes browsing the web, reading sources, and writing a full report with citations. A market scan that used to take you an afternoon now happens while you eat lunch. Not the right tool when you need an answer in the next two minutes, but a game changer when you actually have time to wait for quality.
Web Search Mode ChatGPT, Gemini, Claude, Copilot Chat Mode Think of Web Search as the toggle that lets the AI look things up live instead of relying on what it learned during training. The model goes out to the internet, reads pages, and uses them in its answer. Turn this on when the answer depends on something that happened this week: a rate change, a new release, a news story, a today’s price. Off, the model can only tell you what was true on its knowledge cutoff date.
Canvas / Artifacts ChatGPT (Canvas), Gemini (Canvas), Claude (Artifacts) Chat Mode Think of Canvas (in ChatGPT and Gemini) and Artifacts (in Claude) as a side panel where the AI builds a real document, table, or piece of code you can edit live, while the chat stays in the other panel. No more scrolling back through chat history hunting for the latest draft. The artifact stays put, gets revised in place, and you can copy it out cleanly when you are done. This is the single biggest quality-of-life feature added to AI tools in 2025.
Projects ChatGPT, Claude Chat Mode Think of a Project as a folder inside the chat tool that holds files, custom instructions, and a long-running thread for one piece of work. Everything inside the Project shares context with the AI automatically. You stop pasting the same background ("I am a PM, the project is the Acme rollout, the team is X, Y, Z…") into every new chat. The AI remembers it for you, in that folder, for as long as you keep it. Any work you come back to more than twice belongs in a Project.
Custom GPTs ChatGPT Chat Mode Think of a Custom GPT as a saved version of ChatGPT with its own instructions, files, and personality. You build one in five minutes without writing code, then reuse it forever. Build one called "Weekly Status Report" that knows your team, your VP’s pet peeves, and your preferred format. From now on, you click that GPT and paste the transcript. No re-prompting. That five-minute setup saves you ten minutes every Monday for the rest of the year.
Gems Gemini Chat Mode Think of Gems as Google’s version of Custom GPTs. Saved Gemini assistants with their own instructions and files. Same idea, just inside Gemini. If your team runs on Google Workspace, Gems give you the Custom-GPT pattern without leaving the Google ecosystem. The instructions you write into a Gem stay private to you.
Agents / Agent Mode ChatGPT, Claude, Gemini Chat Mode Think of an Agent as the AI doing the work itself. You tell it the goal. It opens browser tabs, fills forms, clicks buttons, calls other tools, and chains the steps together until it is done. Magic for repetitive web work like "find the contact for the head of procurement at these 20 companies." Read what the agent did before you trust the result. It will sometimes click the wrong thing or take the wrong shortcut, and the only way you catch it is by checking the trail.
Voice Mode ChatGPT, Gemini, Copilot Chat Mode Think of Voice Mode as a real spoken conversation with the AI. You talk, it talks back, it interrupts when it has a better idea. Use it on a walk or a drive when you want to think out loud and have the thinking captured. Brainstorm an upcoming hard conversation. Rehearse a pitch. Work through a tough message you need to send. The transcript is waiting when you get back to your desk.
Auto / Auto-Max Perplexity Chat Mode Think of Auto as Perplexity letting you off the hook of picking a model. You ask a question, the product picks the right model for that question on its own. You do not need to know whether your question is better suited to Sonar, GPT, or Claude. Auto figures that out. Auto-Max forces it to always use the strongest available models, at the cost of speed and your query budget.
Think Deeper Copilot Chat Mode Think of Think Deeper as Copilot’s version of Thinking mode. It spends extra time reasoning before it answers, instead of replying as fast as it can. Helpful when the question has trade-offs and you want the AI to weigh them, not just respond fast. Use it when "the answer is obvious" turns out to be the wrong answer half the time.
GPT-5.5 ChatGPT Model Think of GPT-5.5 as OpenAI’s current flagship brain inside ChatGPT. Comes in variants like Instant and Thinking that trade speed for depth. When a teammate says "I used GPT-5," ask which variant. Instant and Thinking on the same model produce very different work. The variant is the difference between a 5-second sketch and a 90-second considered draft.
GPT-4o ChatGPT Model Think of GPT-4o as the previous-generation OpenAI model. Still available behind some ChatGPT features and the API, but no longer the default. You will see it in older guides and tutorials. Treat it as a step behind GPT-5. If you are following a 2024 walkthrough and it does not work the same way, the model under the hood probably changed.
Claude Sonnet 4.6 Claude Model Think of Sonnet 4.6 as the brain behind claude.ai for free and Pro users. Anthropic’s daily-driver model: fast enough for chat, smart enough for serious work. In our Issue 01 testing, Sonnet produced the cleanest one-page status report from a real meeting transcript on the first try. If you are picking one model to learn first, this is a strong pick.
Claude Opus Claude Model Think of Opus as Anthropic’s largest, slowest, most careful model. Available on paid tiers. Each message costs more and takes longer than Sonnet. Reach for Opus on the few times a week you need the AI to be its sharpest: a sensitive memo, a hard analysis, a piece of writing that will be read by people whose opinion matters. Default to Sonnet the rest of the time.
Gemini 3 Pro Gemini Model Think of Gemini 3 Pro as Google’s flagship brain inside Gemini, on the AI Pro tier. Big context window, multimodal, tuned for reasoning. When the question is hard and the document is long, Gemini 3 Pro is the version of Gemini you want. Free-tier users get the lighter Flash model instead, which often misses things buried in long transcripts.
Gemini 3.5 Flash Gemini Model Think of Flash as Google’s fast, lightweight brain. Default on the Gemini free tier. Fast and cheap. Great for quick lookups and short messages. Less great when you paste a 12-page transcript and need every detail caught. If the result feels shallow, the model is the reason; upgrade or paste a smaller chunk.
Flash Extended Gemini Model Think of Flash Extended as Gemini giving the lightweight Flash model more time to think before it answers. A small quality bump without paying for the bigger Pro model. Worth turning on when you are about to ask Flash a harder question than usual.
Sonar Perplexity Model Think of Sonar as Perplexity’s house family of models, tuned for search and citation rather than freeform chat. It is why Perplexity answers feel different from ChatGPT or Claude even on the same question. Sonar is built to ground every claim in a source it can show you.
Context Window All Limit Think of the context window as the AI’s short-term memory in one sitting. It is how many words it can hold in its head at once before it starts forgetting the earlier stuff. Measured in tokens. A 200k-token window holds about a 500-page book. An 8k window holds about 12 pages. If you paste a long transcript into a small-window model, it quietly drops the beginning and you never see a warning. Use a big-window model (Claude, Gemini Pro) for anything long, so the AI actually reads what you gave it.
Token All Limit Think of a token as a chunk of text the model reads as one unit. About four characters or three-quarters of a word in English. "Hello, world" is roughly 3 tokens. A page of text is around 500. When a vendor brags "200k token context window," that is roughly 150,000 English words at once. A whole novel. Translating tokens to "pages of meeting transcript" is the only way to know if a tool can really hold your work.
Character Limit (per message) Copilot (Free) Limit Think of the character limit as a hard ceiling on a single message you paste in. Different from the context window: this is per-message, not for the whole chat. Free Microsoft Copilot caps at 10,240 characters per message, which a real 60-minute meeting transcript blows past in one paste. Knowing this in advance saves you ten minutes of pasting, getting cut off, and re-pasting in chunks.
Rate Limit / Cooldown All Limit Think of rate limits as a pause the product forces on you after heavy use. The vendor decides "you have used a lot, wait a few hours, or upgrade." When the tool suddenly says "you have hit your limit," that is a cooldown, not a bug. It is also the moment vendors try to convert you to paid. If you keep hitting it, either chunk your work into smaller sessions or accept that the paid tier is worth it.
Knowledge Cutoff All Limit Think of the knowledge cutoff as the date the model stopped learning. Anything that happened after is invisible to it unless web search is on. A model with a 2024 cutoff has no idea who won last week’s election or what your company announced yesterday. Ask "what is your knowledge cutoff?" before you trust any answer about recent events.
File Upload Limit All Limit Think of the file upload limit as the cap on how many files, and how large, you can attach to one chat or one project. Plan ahead. NotebookLM and paid tiers comfortably take dozens of files. Free tiers often choke at three or four. If you have a stack of meeting decks to summarize, check the limit before you start uploading.
Prompt All AI Concept Think of the prompt as the instructions you type into the AI. The single biggest lever on the quality of what comes back. The same tool can produce a useless answer or a brilliant one. The difference is almost always the prompt. Two extra sentences of context, one example of the format you want, and the output jumps a full grade. This is why "prompt engineering" became a job: the upside is huge.
System Prompt All AI Concept Think of the system prompt as the hidden background instructions that shape how the AI behaves before you ever type. Set by the product, by the Custom GPT you are using, or by the Project you are inside. When a Custom GPT "just knows" to write in your voice or use your template, that came from a system prompt the builder wrote once. You can write your own when you build a Custom GPT or set Project instructions: that is where the personality lives.
Hallucination All AI Concept Think of a hallucination as the AI confidently making up a fact, a quote, a citation, or a person who does not exist, and serving it to you in the same calm tone it uses for true things. AI does not know when it is wrong. It is not lying on purpose. It is pattern-matching. The number, the case law, the URL all look real. Always verify any AI-generated name, number, date, or link before you put it in front of an executive, or it will be your name on the mistake.
Temperature All AI Concept Think of temperature as a dial behind the scenes that controls how predictable or creative the model’s answer is. Low temperature: same answer every time. High temperature: different answer every time. You rarely set this yourself in a chat tool, but it is why two identical prompts can produce two different answers. If you want consistent output (say, for a template), regenerate until you get one you like, then save it as a Custom GPT so the temperature is locked in.
Grounding All AI Concept Think of grounding as telling the AI "only answer from this source." Your files, the live web, a specific database. Anything outside that source is off-limits. Grounded answers are dramatically less likely to hallucinate. NotebookLM is the cleanest example: it only answers from your uploaded files, and it refuses to invent things. Any time you can ground an answer in a real source, do it.
Retrieval-Augmented Generation All AI Concept Think of RAG as the AI doing a quick search first, then writing its answer using what it found. Instead of relying on memory, it looks something up, reads it, and quotes from it. This is why Perplexity always shows sources and why NotebookLM only answers from your uploads. When a vendor says "we built RAG on your data," they mean their AI looks at your company’s real documents before answering, instead of guessing from training data.
Reasoning Model All AI Concept Think of a reasoning model as one that thinks step by step before answering, often showing its work. Slower than a standard model. Better on hard problems where one wrong step ruins the whole answer. Use a reasoning model when the work needs to be right, not fast: a calculation, a logic puzzle, a multi-step plan, a hard analysis. Use a standard model for everything else, or you will spend your day staring at progress bars.
Multimodal All AI Concept Think of multimodal as the AI being able to take in more than just text. Images, audio, PDFs, screenshots, video. Most current chat tools are multimodal by default. You can paste a screenshot of a confusing chart and ask "what is this telling me." You can hand it a photo of a whiteboard and ask for a clean typed version. The day you realize you can drag images in is the day AI gets twice as useful.
Fine-Tuning All AI Concept Think of fine-tuning as training extra data into a model so it gets better at a narrow job. Done once, by engineers, with thousands of examples. You will not do this yourself. But when a vendor pitches "we fine-tuned a model on your data," they mean they spent real money and time teaching a model to talk like your company. It is different from a system prompt: deeper, more expensive, harder to change later.
Embedding All AI Concept Think of an embedding as a numeric fingerprint of a piece of text. AI tools compare fingerprints to find content that means the same thing, even when the words are different. This is why "find related docs" inside an AI tool works even when your search words do not appear in the document. The tool is matching meaning, not letters. It is the quiet machinery behind every "smart search" feature you have used in the last year.
Why so many chat modes? All AI Concept Think of every model as making the same three trade-offs: speed vs depth, cost vs accuracy, breadth vs focus. The modes let you pick which trade-off you want, one message at a time. You are not supposed to memorize them all. Default to fast for everyday work, slow and careful for high-stakes work, and web search on when the topic is recent. That covers 95% of real use.
AI All Acronym Artificial Intelligence. Think of AI as software that does jobs once thought to need human thinking: writing, summarizing, translating, recognizing images, holding a conversation. The umbrella over everything else on this page. When someone says "we use AI for that," ask which tool, which model, and which mode. "AI" by itself does not tell you what is actually happening.
LLM All Acronym Large Language Model. Think of an LLM as the kind of AI behind every chat tool on this site. Trained on billions of pages of text. Predicts the next word, over and over, well enough that the result reads like a person wrote it. When someone says "the LLM hallucinated," they mean the chat tool made something up. When a vendor says "we use an LLM," they mean their product is built on the same family of brain as ChatGPT or Claude.
GPT ChatGPT Acronym Generative Pre-trained Transformer. Think of GPT as OpenAI’s family of models, the brand behind ChatGPT. OpenAI also reuses the letters for Custom GPTs, which are saved assistants. You will see these three letters everywhere this year. Now you can read past them: it is just the name of OpenAI’s model family and the assistants people build on top.
RAG All Acronym Retrieval-Augmented Generation. Think of RAG as AI that looks something up before it writes the answer. Search first, write second. Vendors love this term because it is a real fix for hallucination. When a product is "RAG-powered" or "grounded in your data," the answers come with actual sources you can click. Look for it whenever accuracy matters more than creativity.
API All Acronym Application Programming Interface. Think of an API as a way for one piece of software to talk to another. The plumbing. When IT says "we call the OpenAI API," they mean your company’s internal tool quietly uses ChatGPT under the hood. The "AI" in your company’s shiny new product is almost always one of these calls behind a logo.
NLP All Acronym Natural Language Processing. Think of NLP as the older branch of AI focused on understanding and producing human language. LLMs are the modern, much more powerful version. You will see "NLP-powered" on older product pages. Today, it usually means an LLM under the hood. The branding caught up; the underlying tech jumped.
OCR All Acronym Optical Character Recognition. Think of OCR as the AI pulling text out of an image or scanned PDF. The step that turns a picture of words into actual editable words. Why uploading a screenshot of a slide into AI still works. The model OCRs the image first, then reads the text. Without OCR, every image upload would be a black hole.
MoE All Acronym Mixture of Experts. Think of MoE as a model design where only part of the brain fires on any given question. Different "experts" handle math, code, language, etc. You will hear this in vendor pitches. The practical version: MoE models are faster and cheaper to run than equally smart single-block models, which is why so many flagship 2025 models use it.
RLHF All Acronym Reinforcement Learning from Human Feedback. Think of RLHF as training a model with thumbs-up and thumbs-down from people: "this answer good, this answer bad, do more of the first." Why modern chat tools sound polite, helpful, and refuse to do dangerous things by default. It is the human polish on top of the raw model. Every time you click thumbs-up or thumbs-down in a chat, you are feeding the next round.
SLM All Acronym Small Language Model. Think of an SLM as a stripped-down LLM that runs on a phone, laptop, or device without sending your text to a data center. Why some AI features now work offline. Why Apple Intelligence and Microsoft’s on-device Copilot features stay private: the SLM is running on the chip in your machine, not on a server somewhere.
M365 Copilot Acronym Microsoft 365. Think of M365 as the suite of Microsoft apps your company already pays for: Teams, Outlook, Word, Excel, PowerPoint, SharePoint. When someone says "M365 Copilot," they mean the paid Copilot that lives inside those apps, with access to your inbox and your files. Different product from free Copilot at copilot.microsoft.com.
Incognito mode Claude Feature Think of incognito as a private chat that does not save. The little ghost icon in the top-right of Claude opens one. You chat, you get your answer, you close the window, and the conversation disappears. Nothing lands in your chat history. Nothing is used to train the model. Sometimes you want to ask AI a sensitive question (a coworker situation, a personal thing, a draft you want to throw away cleanly) without it living in your account forever. Incognito lets you ask and forget. Most people do not know it exists and quietly accumulate chats they would rather not have.
Customize Claude Feature Think of Customize as standing orders for Claude. A sidebar item where you tell Claude how you want it to write to you every time: “Always write in clear language,” “I am a PM in transportation,” “Never use em dashes.” Once set, every new chat starts with those rules already in place. The same five sentences you find yourself adding to every prompt? Put them in Customize once and stop typing them forever. The quality bump is immediate. Most readers do not know this menu exists and re-paste their context into every chat.
Action chips Claude Feature Think of the chips under Claude’s composer (Write, Learn, Code, Life stuff, Claude’s choice) as starting points. Tap one and Claude pre-loads itself for that kind of task before you type a word. Blank-page paralysis is real. The chips are five doors into the same room. Tapping Write tells Claude “we are drafting.” Tapping Learn tells Claude “we are explaining.” Same prompt, different shaped answer. You can always skip the chips and type, but using them is the fastest way to set tone.
Artifacts Claude Feature Think of Artifacts as a side panel where Claude builds a real document, table, or piece of code you can edit live. The chat stays on the left, the artifact stays on the right. Click the Artifacts item in the sidebar to see every artifact across every chat. No more scrolling chat history hunting for the latest draft. The artifact stays put while Claude revises it in place. You copy it out clean when you are done. This single feature is why Claude is loved for serious writing.
Audio Overview NotebookLM Feature Think of Audio Overview as a podcast NotebookLM generates from your uploaded files. Two AI hosts have a 10-to-20-minute conversation about your documents. Listening is easier than reading. A 200-page benefits PDF becomes a 12-minute drive-home podcast. Especially useful for onboarding new team members or revisiting old projects on a walk.
Mind Map NotebookLM Feature Think of Mind Map as NotebookLM drawing the structure of your uploaded files: topics, sub-topics, how things connect. Click any node and it dives in. A 300-page document is hard to summarize but easy to navigate as a tree. Use Mind Map when you need to find your way around a notebook, not when you need a final answer.
Study Guide NotebookLM Feature Think of Study Guide as NotebookLM turning your uploaded files into flashcards, quiz questions, and a glossary. One-click generation. Originally built for students. Wildly useful for adults learning a new domain at work: vendor docs, compliance training, a new platform. Skip the highlighter; let NotebookLM build the study deck.
Stakeholder All PM Term Think of a stakeholder as anyone whose work, money, or reputation is on the line for your project. Sponsors, users, vendors, legal, IT, the VP whose name is on the slide. When you ask AI to “write a status update,” the answer depends entirely on who reads it. A status for the VP is one paragraph and a number. A status for the working team is a list with owners. Naming the stakeholder in your prompt is the single biggest quality lift on PM writing.
Transcript All PM Term Think of a transcript as the typed-out version of a meeting, call, or interview. Teams, Zoom, Meet, and Otter all produce them automatically now. Transcripts are the raw material AI is best at. A 60-minute meeting becomes a 90-second summary, a list of decisions, a list of action items, a draft email. Drop the transcript into Claude or Gemini and ask for any of those four outputs by name.
RAID Log All PM Term Think of a RAID log as a four-column table PMs keep to track Risks, Assumptions, Issues, and Decisions across a project. One row per item, owner and date next to it. AI is great at filling and updating RAID logs from meeting transcripts. Drop the transcript in, ask “extract any new risks, assumptions, issues, or decisions and add them to this table.” What used to take 30 minutes of skimming takes 30 seconds.
Executive summary All PM Term Think of an executive summary as the one paragraph at the top of a long document that tells a busy person what to do or know without reading the rest. Five sentences max. Decision, status, ask, risk, next. The single most useful AI output for working with senior people. Every long memo, deck, and report should start with one. Ask Claude to write it last, after the body is done, so the summary actually reflects the final story.
Tier All Limit Think of a tier as which version of a tool you are paying for. Free, Plus, Pro, Max, Team, Enterprise. Each step up unlocks more models, bigger files, more messages per day, and steadier access during busy hours. When someone says “Claude is amazing,” ask which tier they are on. Free Claude and Pro Claude behave differently enough that they are almost different products. Same logic for ChatGPT (Free vs Plus vs Pro) and Gemini (Free vs AI Pro vs AI Ultra).
Effort level (Low / High) Claude, ChatGPT Chat Mode Think of effort level as a dial on the model. Low means “answer fast.” High means “think harder before answering.” In Claude it sits next to the model name (“Sonnet 4.6 Low”). In ChatGPT it lives under model picker as Standard, Thinking, and Thinking longer. Same model, different answers. Low on a contract review will miss things. High on “rewrite my email” burns your daily quota for no gain. Match the dial to the stakes: Low for everyday, High for the answers that need to be right.
Soft cap vs Hard cap All Limit Think of a soft cap as a warning (“you are using a lot today, please slow down”) and a hard cap as a wall (“stop, you cannot do this until tomorrow”). Free Copilot’s 10,240-character per-message limit is a hard cap. Free Claude’s daily message limit is also hard. Most paid tiers move toward soft caps you can lean on. Knowing which kind of cap you are about to hit changes your day. A soft cap is fine to push through; a hard cap means breaking your work up or upgrading the plan.
Cursor External Tool Think of Cursor as VS Code (the popular code editor) with AI built in. It writes, edits, and refactors code alongside you. Not in our six-tool comparison because it is a developer tool, not a generalist chat. We mention it on the home page because PMs often hear “we use Cursor” from engineers. Now you know what they mean. If your work is mostly code and pull requests, Cursor or one of its peers (Windsurf, Zed) is closer to what you want than a chat tool.
Beehiiv External Tool Think of Beehiiv as the publishing platform AI PM Weekly runs on. It is to newsletters what Substack is to blogs: signup, sending, archives, and analytics in one place. We name it because subscribing happens through Beehiiv, not through us directly. If a confirmation email shows up from “Beehiiv,” that is not spam; that is the subscription working.
M365 Copilot Copilot Tool Think of M365 Copilot as the paid Copilot that lives inside Word, Excel, Outlook, Teams, and PowerPoint. Different product from free Copilot at copilot.microsoft.com, which is the web chat anyone can use. When someone at work says “use Copilot to summarize this meeting,” they almost always mean M365 Copilot inside Teams. It has access to your inbox, calendar, and files. The free Copilot does not. We test on M365 Copilot for any task that involves real work documents.