Most people’s first week with Google Gemini goes the same way. You type “write me a blog post”, get something bland, shrug, and go back to doing it yourself. The tool isn’t the problem. The problem is treating Gemini like a search box with better grammar. It behaves more like a capable junior colleague: fast, tireless, occasionally overconfident, and only as good as the brief you hand it.
What follows is the practical version. A repeatable workflow, prompts you can copy, and the specific places where Gemini earns its keep versus where it will quietly waste your afternoon.
Pick the surface before you pick the prompt
Gemini isn’t one app. It’s several, and they behave differently. Choosing the wrong one is the most common reason people conclude it “doesn’t work”.
- Gemini app on Android or iOS — best for voice, camera input, and quick questions while you’re out. On most Android phones, holding the power button gets you there.
- gemini.google.com — the full workspace: long documents, file uploads, Deep Research, and canvas-style editing.
- Gemini inside Workspace — Docs, Sheets, Gmail, Slides. Use it when the output needs to land in a document you’re already working in, because copy-paste is where quality dies.
Rule of thumb: if the task involves pasting something long, move to the browser. If it needs your eyes and ears, use the phone.
Brief it like a freelancer, not a genie
Weak prompt: “Plan a trip to Porto.” You’ll get a listicle you could have found on any travel blog in ten seconds.
Strong prompt: “You’re a travel planner who specialises in trips with young children. Plan three days in Porto for two adults and a two-year-old, arriving Friday evening, total budget £600 excluding flights. We walk slowly, need a nap window each afternoon, and one of us is vegetarian. Give me a day-by-day plan as a table, with an indoor backup for each block in case it rains.”
Same model. Wildly different output. The second brief contains five things the first one lacks: a role, context, hard constraints, a format, and an edge case.
The five-part brief
- Role — who should it sound like?
- Task — the single outcome you want.
- Context — the facts only you know.
- Constraints — budget, length, tone, what to avoid.
- Format — table, bullets, 200 words, three options.
If you remember one habit, make it this: after the first answer, don’t start a new chat. Reply with what’s wrong. “Too generic. Cut the guidebook stuff and give me specific neighbourhoods.” Iterating in the same thread is where the quality jump actually happens.
A worked example: turning a messy week into a plan
Here’s a workflow I’ve used often enough to trust. It takes about four minutes.
Step one: dump everything. Paste in raw notes, half-finished to-dos, three email threads, the thing your manager mentioned in passing. Don’t tidy it. Gemini handles noise well.
Step two: ask for structure, not answers. “Group these into themes. Don’t suggest solutions yet. Flag anything that’s actually three tasks wearing a trench coat.”
Step three: add the calendar. “Now build me a Tuesday and Wednesday with two-hour focus blocks. Put shallow work in the 30-minute gaps. Nothing before 9am.”
Step four: verify the load-bearing details. Any deadline, price, address, or legal claim in that output needs checking against the source. Gemini drafts; you sign off. That distinction matters more than people expect, and the widely covered story of hikers who had to be rescued after trusting an AI-planned route is a decent reminder that a confident answer is not a verified one.
Give it your files, not just your words
Two features change the ceiling on what Gemini can do: uploads and connected apps. In gemini.google.com you can drop in PDFs, spreadsheets, slide decks, and images, then ask questions across all of them at once. “Compare these three contracts and list every clause about notice periods” becomes a 40-second job instead of a 40-minute one.
The consumer integrations have quietly got genuinely useful too. If you’ve ever wanted to search your own photo library in plain English, Gemini Spark can now handle Google Photos library tasks directly, from finding images to organising them. And if you’re generating visuals for work rather than publishing them as-is, you can switch off Gemini’s visible watermarks so the output drops cleanly into slides and mockups.
Match the model to the job
Gemini ships in tiers, and picking the wrong one is a quiet tax on your day. Flash models are built for speed: summaries, reformatting, first drafts, high-volume grunt work. Pro-tier models are for reasoning-heavy tasks like multi-step analysis or code, where you’d otherwise want to think out loud with a smart friend.
That line moves fast. Google has said its newer Gemini 3.8 Flash model works harder but might cost more, which tells you the fast tier keeps absorbing work that used to require the expensive one. Start on the fastest model. If the answer comes back shallow or wrong, escalate. Don’t default to the biggest model for everything; you’ll just wait longer for the same result.
If you’re studying, the loop looks different
Students often get the worst version of AI because the obvious use, “write my essay”, is both against the rules and useless for learning. The good loop is Socratic.
Ask Gemini to quiz you on a chapter, one question at a time, and refuse to give the answer until you’ve attempted it twice. Ask it to explain a concept at three levels: to a 12-year-old, to a first-year undergraduate, and to an examiner. Ask it to mark your practice response against a rubric and tell you exactly where the argument sags. Google has been building for this audience specifically, and its dedicated student hub pulls those tools together rather than scattering them across the app.
Prompt patterns worth saving
- The rewriter: “Keep every fact. Rewrite this at a year-9 reading level. Cut 30% of the words.”
- The devil’s advocate: “Argue against my plan. Find the three weakest assumptions and rank them by how expensive they’d be if wrong.”
- The rubric: “Give me a six-point checklist I can score this against before I hit send.”
- The translator: “The client is non-technical and impatient. Rewrite this email in plain language, under 120 words.”
- The second opinion: “Here are two drafts. Tell me which is stronger and why, then merge the best parts.”
Five habits that make Gemini look stupid
- Treating the first answer as the final one. Almost every good output is iteration three.
- Asking for facts without asking for sources, then acting surprised by a confident wrong number.
- Pasting 40 pages and asking one vague question. Give it a job, not a pile.
- Using it for work that needs your taste: a message to a friend, a resignation letter, anything where the point is that you wrote it.
- Ignoring the settings. Personalisation, connected apps, and model choice all change results, and most people never open them.
Build a small prompt library, then prune it
The highest-return habit here isn’t learning more features. It’s keeping a notes file with the eight or ten prompts that reliably work for you, with your context already baked in: your job title, your audience, the tone your boss actually likes.
Keep it short on purpose. Every time a prompt gives you something you’d have been happy to send, paste it in. Once a month, delete the ones you never reuse. Within a few weeks Gemini stops feeling like a novelty you occasionally open and starts feeling like a colleague who already knows your standards, which is the only version of this that saves real time.