The first thing I asked Gemini was whether a 1998 Honda Civic needed premium fuel. It answered in about four seconds, got the octane rating right, and then offered to look up the car’s MOT history if I pasted the registration. Slightly unnerving. Genuinely useful.
That was six weeks and roughly 140 conversations ago. I started on the free tier, upgraded to Google One AI Premium on day 12, and kept a notes file the whole time. What follows is the review I wish I’d read before signing up: what a beginner can realistically do with Gemini, where it quietly falls over, and whether the £18.99 a month is justified.
What Gemini actually is, without the marketing
Gemini is both the name of Google’s AI model family and the name of the chat app you use to talk to it, which is unhelpful and worth getting straight early on.
You’ll find it at gemini.google.com, as a mobile app, and in a side panel inside Gmail and Google Docs. Two models do most of the work: a fast one called Flash that handles free requests, and a slower, stronger one called Pro reserved for paying subscribers. Don’t confuse any of it with Google Assistant. Assistant still does “set a timer” and “call Mum”. Gemini does “read this 40-page PDF and pull out every deadline mentioned.”
The free version is the real starting point
Plenty of beginner guides skip straight to the paid plan. That’s a mistake, because the free tier is generous enough to tell you whether you even like the thing.
What you get at zero cost
Text chat, image generation, photo and file uploads, voice conversation, and a small allowance of Deep Research runs. I used it for eleven days before paying and only hit a ceiling twice, both times on image generation. If your needs are “draft this email” and “explain this concept”, free is genuinely enough.
What £18.99 a month adds
- The stronger Pro model with a far higher usage ceiling
- Deep Research, which browses 30 to 40 sources over 20 minutes and returns a structured report with citations
- Gems: saved custom assistants with their own instructions and reference files
- A 1 million token context window, which in practice means a 600-page PDF or an entire codebase
- 2TB of Google One storage and higher NotebookLM limits
The jump from 15GB to 2TB of Drive space matters more than people admit. If you already pay for extra Google storage, the extra AI features cost you roughly £5 a month on top, not £19.
Four things that genuinely impressed me
- Long email threads. I forwarded a 60-message chain about a delayed kitchen delivery and asked for a timeline. It built a dated sequence, spotted two contradictions between what the supplier promised on the 4th and what they said on the 19th, and drafted a follow-up that was firm without being rude.
- Video questions. Paste a YouTube link, ask what was said about a specific topic, and it answers with a timestamp you can click. That alone saved me an hour on a two-hour planning committee recording.
- Document extraction. I fed it a 214-page local planning document and asked which policies applied to a two-storey extension. It answered in about 90 seconds with section numbers I could check myself.
- Directed image edits. “Remove the wheelie bin from the foreground, keep the shadow where it falls” worked on the second attempt. On the first it removed the shadow too.
Where it stumbles
The invented citation is the big one. It quoted a council policy number that looked completely plausible and did not exist. I only caught it because I went looking, which is the point: verify anything factual before it leaves your drafts folder.
Long conversations also drift. Past about 30 exchanges it starts answering an earlier version of your question. Starting a fresh chat with a short summary works better than correcting it in place, even though that feels wasteful.
And left to its own devices it writes in a rounded, faintly promotional register full of triads and “not only… but also” constructions. You have to instruct it out of that habit, or it edits everything into the same voice.
The privacy setting almost nobody turns off
Gemini Apps Activity is on by default, which means your conversations can be reviewed by humans and are retained for up to three years, disconnected from your account. You can switch it off at myactivity.google.com under Gemini Apps Activity. Do it before you paste anything sensitive.
I’d go further and treat the free tier as public. Client contracts, medical details, anything with someone else’s name in it: keep out. There’s a practical privacy walkthrough for AI tools that covers the account settings most people never open, and it’s worth 15 minutes before you hand over work documents.
Gemini vs ChatGPT for a total beginner
If your life already runs through Gmail, Docs and Calendar, Gemini wins on convenience by a wide margin. Summarising a thread without copying anything anywhere is a real advantage, and it’s the reason I kept paying.
ChatGPT is still ahead on memory and custom instructions, and its free tier is arguably more capable for pure writing tasks. I wrote up an honest pros and cons list after using ChatGPT daily, and most of the weaknesses it identified apply in reverse here: Gemini feels more integrated but less personal. I also ran the same six-week experiment with Claude, and that longer Claude review for beginners is a fairer comparison for anyone doing heavy document reading.
If you’re weighing up seats for a small team rather than yourself, the per-person maths changes quickly, and this breakdown of what $25 a seat actually buys is a useful sanity check before committing.
A sensible first week with Gemini
Don’t try to learn the whole tool at once. This is roughly the sequence I’d follow if I started again:
- Days 1 and 2: one task only, something you already do weekly. Meeting notes, a recurring report, replies to a certain kind of email.
- Day 3: upload a real document instead of describing it. This is where most beginners have their first “oh” moment.
- Day 4: build a Gem for that recurring task, with three or four examples of output you like pasted into its instructions.
- Day 5: run one Deep Research query on something you’d normally spend an hour Googling, then check three of its sources.
- Day 6: turn off Gemini Apps Activity and read the data controls properly.
- Day 7: decide. If you haven’t used it in three days, cancel. The free tier will still be there.
The habit that decides whether Gemini helps or wastes your time
Context is the whole game. Three-word prompts produce three-word-quality thinking. The people who get value out of Gemini are the ones who paste the actual email, attach the actual report, name the audience, and say what a good answer looks like.
Ask for the answer first, then ask it to show its reasoning, rather than the other way round. Keep a note of prompts that worked so you’re not rebuilding them each Monday. And treat every factual claim as a draft until you’ve checked it, because the tool’s confidence has almost nothing to do with its accuracy.
Six weeks in, my honest position: Gemini is the best AI assistant Google has shipped, it’s worth £18.99 a month if you live in Gmail and Docs, and it rewards people who treat it like a fast, tireless, slightly overconfident junior colleague who needs a brief and a proofreader.