Field-guide course · Getting comfortable
The everyday assistant
Writing, research, documents, meetings, and images — the right tool for each, by name.
6 lessons · For: You chat with an assistant a few times a week and want it woven through your actual work, not just visited. · Product details last reviewed July 2026
The chat box is the front door, but the products have grown far past it: editing surfaces built for drafting, research modes that cite their sources, notebooks that answer from your documents, notetakers that attend your meetings, image tools that read as well as they draw. This course walks through a working week and names the best instrument for each part.
Everything here has a free way to start. Where a paid tier matters, we say so plainly.
- Lesson 1
Writing and editing, properly
The biggest single upgrade to AI-assisted writing is moving out of the raw chat scroll and into an editing surface. ChatGPT's Canvas and Claude's Artifacts open your draft in a side-by-side panel where you can select a paragraph and ask for a change to just that — tighter, warmer, evidence first — instead of watching the whole piece regenerate and drift. Revision becomes surgical, which is what real editing is.
The second upgrade is teaching voice by example rather than adjective. "Professional but friendly" means nothing; three sentences of your own best writing mean everything. Paste a sample, say "match this voice," and keep your best prompts saved. And the standing rule from the study guides applies doubly here: the model drafts, you finish. The judgment, the stakes, and the audience nuance in the final 20% are the part that is actually your job.
In the toolbox
Canvas
OpenAI, inside ChatGPT · Included free
A side-by-side draft editor: select any passage and revise just that, adjust length and reading level, track the piece as a document rather than a chat.
Artifacts
Anthropic, inside Claude · Included free
Claude's workspace panel for documents and drafts (and later, working mini-apps). Paired with Claude's writing quality, the strongest pure-prose setup going.
Gemini in Google Docs
Google · Included with Workspace and AI plans
"Help me write" directly inside the document you are already editing — drafting, rewriting, and tone changes without leaving Docs.
Grammarly
Grammarly · Free tier; Pro ~$12/mo
AI editing that lives in every text box you type in — email, browser, Slack. Less powerful than the big assistants, but present at the exact moment of writing.
Try it this week
Take a real piece you owe someone this week. Draft it in Canvas or Artifacts, give the model three sentences of your own voice to match, then make at least five selection-level edits instead of asking for full rewrites.
- Lesson 2
Research with receipts
Ordinary chat answers come from the model's training plus a quick search — fine for orientation, unfit for anything you will cite. For real questions, use the tools built to show their work. Perplexity made citations the whole product: every claim footnoted, every footnote clickable. The deep research modes inside ChatGPT, Gemini, and Claude go further — give one a paragraph-long brief and it will spend several minutes reading dozens of sources and return a structured, cited report that would have cost you an afternoon.
Two disciplines make these tools trustworthy. Brief them like a person: the question, the decision it feeds, what sources to prefer and avoid, the format you want back. And then open the citations that matter — the study guides call these the load-bearing claims, the ones your decision actually rests on. A cited report you never spot-checked is just a confident report.
In the toolbox
Perplexity
Perplexity · Free tier; Pro ~$20/mo
Search reimagined as cited answers. The right default for quick factual questions where you want sources without commissioning a full report.
Deep Research
OpenAI, inside ChatGPT · Limited free; more on Plus
Commission a multi-source investigation and get a long, structured, cited report in minutes. Thorough to a fault — brief it tightly or it will tell you everything.
Deep Research
Google, inside Gemini · Included with free and paid tiers
Gemini's research agent, with Google's reach across the live web. Plans its approach and shows it to you before it runs — edit the plan, not just the question.
Research
Anthropic, inside Claude · Paid tiers
Claude's take on agentic research, strongest when combined with your connected files and documents rather than the open web alone.
Try it this week
Pick a genuine question from your work. Write a five-sentence brief — question, decision, preferred sources, exclusions, format — and run it through one deep research mode. Then open the three citations behind the report's most important claims before you act on it.
- Lesson 3
Documents, PDFs, and ground truth
The single most reliable way to use AI is to hand it the actual document and ask questions grounded in it. Every major assistant accepts uploads — contracts, reports, slide decks, spreadsheets — and a model that has read the real text beats one recalling the world in general, every time. Sharpen the ask: "answer with quotes and page references" turns a plausible summary into a checkable one.
When the material outgrows a single conversation — a policy binder, a course's readings, a year of board minutes — move to Google's NotebookLM, the quiet superstar of this category. Load up to dozens of sources into a notebook and it answers only from them, citing the exact passages it used. Its signature trick, Audio Overviews, turns your sources into a surprisingly listenable two-host podcast — commute-grade studying from your own materials.
In the toolbox
NotebookLM
Google · Free; higher limits on AI plans
A source-grounded research notebook: it answers from the documents you load, with citations into them, and can generate audio and video overviews of the material.
File upload in ChatGPT, Claude & Gemini
OpenAI · Anthropic · Google · Included free, with size caps
The everyday move: attach the PDF, ask with citations. Claude is notably strong on very long documents; Gemini reads files straight from Drive.
Try it this week
Take the densest document in your life right now — a contract, a benefits guide, a technical manual. Load it into NotebookLM, ask the five questions you actually care about, and follow at least one citation back to the page to confirm the habit works.
- Lesson 4
Meetings, voice, and ears
Voice turns dead time into thinking time. The voice modes in the ChatGPT and Gemini mobile apps hold real spoken conversations — rehearse a difficult conversation on your commute, think a decision out loud while walking, get a briefing while cooking. For capture rather than conversation, AI dictation tools like Wispr Flow turn rambling speech into clean prose anywhere you can type, at speeds that make typing feel quaint.
Meetings are their own category. Notetakers like Granola and Otter, and the built-in AI companions in Zoom and Teams, produce transcripts, summaries, and action items automatically — which quietly changes what a meeting costs you. Two cautions before you deploy one: recording has consent rules that vary by place and workplace policy, so ask before you bring a bot; and an AI summary of a meeting is a draft like any other — check the action items it assigns you before they calcify into commitments.
In the toolbox
Granola
Granola · Free trial; ~$15/mo
A notepad that listens along on your device during meetings and merges your typed half-notes with the transcript into polished notes. No bot joins the call.
Otter.ai
Otter · Free tier; Pro ~$17/mo
The veteran meeting transcriber: joins calls, produces live transcripts, summaries, and action items, searchable across your meeting history.
Zoom AI Companion / Teams Copilot
Zoom · Microsoft · Bundled with paid workplace plans
The platform-native option — often already approved at work, which for meeting content is the deciding factor.
Wispr Flow
Wispr · Free tier; ~$12/mo
System-wide AI dictation that edits as you speak — pauses, restarts, and "scratch that" all handled. Voice as a first-class input for email and messages.
Try it this week
Do one thinking task entirely by voice this week — a walk-and-talk with ChatGPT or Gemini about a decision you are weighing. Separately, with consent, pilot a notetaker in one internal meeting and grade its action items against your own.
- Lesson 5
Images: making them and reading them
Image generation has crossed from party trick to working tool. ChatGPT's image generation follows instructions unusually well — including rendering legible text, the old giveaway — which makes it the default for diagrams, mockups, and illustrations that must contain specific things. Gemini's image editing is the standout for modifying what exists: "same photo, remove the clutter, warmer light" while keeping the subject recognizably itself. Midjourney remains the aesthetic ceiling for people willing to learn its culture, and Adobe Firefly matters when your legal team asks where the training data came from.
The underused half is reading images. Photograph a whiteboard and ask for structured notes with owners and deadlines. Screenshot an error message, a confusing chart, a parking sign in another language — and just ask. For work use, know your organization's rules on AI imagery, and extend the verification habit here too: a generated diagram can be beautiful and subtly wrong about your own process.
In the toolbox
ChatGPT image generation
OpenAI · Included; more on Plus
Best instruction-following in the category, and reliable text rendering. The default for images that must contain specific elements.
Gemini image editing
Google · Included with free and paid tiers
The standout editor — targeted changes to an existing image while keeping subjects consistent. Google's image stack had a genuine hit here.
Midjourney
Midjourney · From ~$10/mo, no free tier
The aesthetic benchmark for generated art and stylized imagery. A community and a craft of its own; worth it when look is the product.
Adobe Firefly
Adobe · Free tier; bundled with Creative Cloud
Trained on licensed content and designed for commercial safety, with Photoshop integration. The corporate-approved answer.
Try it this week
Two exercises: photograph a real whiteboard or hand-written page and ask for structured notes; then generate three genuinely different candidate images for a document or deck you actually own, and note what you had to specify to get usable results.
- Lesson 6
Teach it who you are
By default, every conversation starts from zero. The products all now offer standing context so they stop asking who you are: ChatGPT's memory and custom instructions, Claude's memory and styles, Gemini's personalization. Five minutes of setup — role, audience, format preferences, "no em-dash overuse" — upgrades every future conversation at once.
For recurring work, go a level deeper: Projects in ChatGPT and Claude (and Gems in Gemini) are standing workspaces that hold instructions and reference files for one job. A "Weekly report" project with your template, last quarter's numbers, and two exemplar outputs turns a fifteen-minute prompt-assembly ritual into one sentence. One hygiene habit: review what memory has accumulated every month or two — these tools remember what you tell them, and the study guides' rule about sensitive data applies to memory features with extra force.
In the toolbox
Projects
OpenAI, inside ChatGPT · Anthropic, inside Claude · Free and paid tiers
Standing workspaces: files, instructions, and conversation history scoped to one recurring job. The single highest-leverage organizational feature in these products.
Gems
Google, inside Gemini · Included with free and paid tiers
Custom versions of Gemini with your standing instructions and files — the same idea as Projects, shareable across your Google life.
Memory & custom instructions
OpenAI · Anthropic · Google · Included
Persistent facts and preferences applied to every chat. Set deliberately, audit occasionally, and keep sensitive details out.
Try it this week
Set your custom instructions in five minutes, then build one Project or Gem for your most recurring task: standing instructions, two reference files, one exemplar output. Use it for two weeks and count the re-explaining that disappears.
Where next
Your week now has the right tool in each slot. The power user course is the next rung: model pickers, reasoning modes, data analysis, building small tools by describing them, and letting assistants act — carefully — on your behalf.
The products above change; the judgment the exam scores does not. The study guides teach that durable layer, and the free check scores you on it in about fifteen minutes.