Ask Gemini to “summarise this” and you get a tidy paragraph you will never use. Give it a job title, an audience, a word limit and one example of the tone you want, and the same request comes back as something you can paste straight into a client email. That gap between the two outcomes is what this tutorial is about.
What follows is a walkthrough of the Gemini chatbot from first login to a repeatable workflow: the settings worth changing on day one, the prompt shape that works, and the habits that keep it from confidently inventing numbers on you.
What you’re actually working with
Gemini is Google’s chatbot. You get it as a web app at gemini.google.com, as iOS and Android apps, and as a sidebar inside Gmail, Docs and Sheets if your account has it switched on. Which model answers you depends on what you pay.
- Free tier. A Flash-class model with a daily prompt cap, limited file uploads and basic image generation. Fine for drafting, rewriting and quick questions.
- Google AI Pro (about $19.99 a month). The Pro model, a context window long enough to hold entire reports, Deep Research, and the ability to build your own Gems.
- Workspace accounts. If you log in with a work Google account, an admin decides whether Gemini is available and whether your prompts are shielded from human review.
One quirk worth knowing early: chats don’t transfer between accounts. Start a project on your personal Gmail and you cannot open it on your work login. Pick the right account before you type anything long.
Ten minutes of setup that pays off later
Decide what happens to your chat history
Gemini Apps Activity is on by default and keeps your conversations for 18 months. Google states that reviewed conversations are decoupled from your account, but the option still exists in your myactivity settings, and turning it off takes about thirty seconds. Do it before you paste anything sensitive, not after.
Install it somewhere you’ll actually open it
The mobile app is genuinely useful for photo prompts, which is where the chatbot stops feeling like a text box. Snap a printed receipt, a handwritten page from a notebook, or a whiteboard after a meeting and ask what changed. Add the web app to your dock or home screen so it isn’t a browser tab you forget about.
Name your chats like files
“New chat” twenty times over is how people conclude the tool is useless. Rename threads to “Pricing page rewrite” or “Q3 board update” so search can find them, and keep one thread per project. Context carries within a conversation, and that continuity is most of the value.
The prompt shape that gets usable output
Most weak answers come from under-specified requests. Four ingredients fix the majority of them:
- Task. A verb with a boundary: rewrite, extract, compare, outline, critique. “Help me with” is not a task.
- Context. Who reads the output, what they already know, what decision it feeds into.
- Format. Bullets, a 120-word paragraph, a three-column table, a numbered list of seven items.
- Constraints. Word count, tone, banned phrases, facts that must not change.
Here’s what that looks like in practice: “Rewrite the paragraph below as a client update email. The reader is a non-technical marketing director who has already approved the budget. Keep it under 90 words, no jargon, one clear ask in the final line. Don’t invent any dates.” That’s a fifteen-second edit to a request you were going to make anyway.
Make it ask before it answers
Add “Before you answer, list your assumptions and ask me up to three questions” to any request where you’re unsure you’ve explained yourself. Gemini is often better at spotting the gap in your brief than you are, and two clarifying questions cost less than a full rewrite.
Ask for two versions, not one
“Give me a 60-word version and a 200-word version” forces a real editorial choice instead of a single averaged-out blob. Same trick works for headlines: ask for ten, then ask it to rank them and explain the ranking.
Feeding it your own material
The upload button accepts PDFs, images, and files from Drive including Docs, Sheets and Slides. A 40-page PDF summarises comfortably on the free tier. A 400-page annual report usually doesn’t, and you’ll hit a length error rather than a bad answer.
For anything long, ask for a section-by-section index first, then request summaries of the two or three sections that matter. It’s more prompts, but you get a map instead of a blurry average. Screenshots work surprisingly well too, provided the image is legible: hand it a photo of a spreadsheet and ask which line item grew fastest between Q2 and Q4.
Gems: saved instructions you don’t retype
A Gem is a reusable set of instructions wrapped around a chat. Build three and you’ll stop writing the same preamble every morning:
- An editor. House style, banned words, a preference for short sentences, instructions to flag any claim that needs a source.
- A weekly report builder. Takes your rough bullet points and returns a structured update in the format your manager expects.
- An explainer. Assumes zero background knowledge, defines every acronym, ends each explanation with one worked example.
Custom Gems sit behind a paid plan. The free tier gets the preset ones, which are useful but generic.
Where Gemini still slips
Arithmetic is the obvious one. Ask it to show its working line by line and check the middle steps, because final answers can be wrong while the reasoning looks clean. Citations are the second: Deep Research does return sources, and those sources don’t always say what the summary claims, so open two or three at random.
The subtler failure is agreeableness. Tell Gemini your plan is good and it will find reasons your plan is good. Flip the request to “argue against this as sharply as you can, and name the strongest objection a sceptical finance lead would raise” and you get something worth reading.
The privacy rules people skip
Consumer Gemini conversations can be reviewed by humans to improve the model. That makes it the wrong place for contract clauses, salary data, API keys, patient details or anything covered by an NDA. Workspace accounts under a business agreement are a different arrangement, and it’s worth confirming which one you’re signed into before your first paste. If you want a wider checklist for vetting AI tools before you hand over data, this privacy review walkthrough for AI tools covers the settings most people never open.
Is the paid plan worth it?
If you use long documents, Deep Research, or custom Gems more than once a week, the subscription earns its keep quickly. If you mostly rewrite emails and ask quick questions, the free tier holds up, and the money is better spent elsewhere in your stack. Our breakdown of which AI tools are worth paying for and which to skip goes through the trade-offs with actual numbers.
Where a chatbot stops and an agent starts
Gemini answers. It doesn’t reliably run a five-step process in the background, check its own work and come back with a finished deliverable. That’s a different category of tool, and if you’re tempted to hand off recurring tasks, start with the AI agents that suit beginners and the ones to leave alone. Gemini also isn’t the only option: Claude alternatives worth testing covers who each chatbot genuinely fits, which matters if you write long documents for a living.
A twenty-minute drill that teaches you more than any video
Pick a real document you already know well. Something with numbers in it, like a supplier quote, a research summary or last quarter’s sales notes. Then run this in a single thread:
- Ask for a 100-word summary aimed at someone who has ninety seconds.
- Ask for every action item and who owns it, as a table.
- Ask it to rewrite the same content for a different reader: a customer, an auditor, an intern.
- Ask which single paragraph is the weakest and what evidence would fix it.
- Ask it to state, in one line, anything it guessed rather than read.
Step five is the one that changes how you use the tool. Save the prompt that produced the best output in a note, because your personal library of four or five good prompts will outperform any generic list you find online. Repeat the drill with a fresh document next week and you’ll start recognising which requests Gemini handles cleanly and which ones you should break into smaller pieces before trusting the result.

