Which tool to choose
Three free options. Each works best in a specific situation.
Gemini (Google) is the best choice if you brainstorm by talking or on the move. The Gemini Live voice mode is available in the mobile app in over 45 languages: open the app, turn on Live mode, and start talking. You can reason about ideas while walking, change your mind mid-sentence, interrupt, jump topics. The free plan runs Gemini 2.5 Flash, Google's lightweight and fast model, which handles everyday questions, writing, and brainstorming well.
ChatGPT (OpenAI) is the all-rounder if you also want concept images and web research in the same place. Free accounts can search for up-to-date information on the web, analyze data, upload images or files, use GPTs, and generate images. A real limit to know: the free plan allows about 10 messages every 5 hours, then the chat falls back to a reduced model until the window resets.
Claude (Anthropic) is the strongest when you start from your own material (notes, briefs, documents). The free plan uses Claude Sonnet 4.5 as the default model and includes projects, artifacts, web search, and file uploads, up to 20 files per chat and a maximum of 30 MB. Practical difference: in brainstorming Gemini draws on a wider range of sources on its own, while Claude works better with what you give it.
Quick rule: brainstorming by voice or tied to Google → Gemini. Ideas plus images → ChatGPT. Ideas from one of your documents → Claude.
How to do it
The problem for those who come up empty-handed: they ask "give me ideas for a post" and get banalities. The AI answers generically because you were generic. Here is the sequence that works.
- Load the context before the request. Write who the audience is, what you sell or talk about, where you publish, what tone you use. Without this the AI is guessing.
- Ask for quantity, not quality. Fifteen ideas, not three. The first ones are always the obvious ones; the value is in the tail of the list.
- Force different angles. Compel the AI to vary perspective: contrarian, behind the scenes, common mistake, surprising data point. Without a constraint it gives you fifteen variants of the same idea.
- Iterate on the best one. Pick an idea and ask it to develop it in three directions. This is where the real content is born.
From browser or app the path doesn't change. The operational syntax for the first round:
You are a content strategist. Here's how I work:
- Audience: [e.g. small artisans who sell on Instagram]
- What I offer: [e.g. a basic course on phone photography]
- Channel: [e.g. Instagram carousels]
- Tone: practical, direct, no jargon
- What I've already published: [e.g. "3 lighting mistakes", "how to clean the lens"]
Give me 15 post ideas, grouped by angle:
3 contrarian, 3 behind the scenes, 3 common mistakes of the audience,
3 surprising data/number, 3 micro tutorials.
For each, write only the title and a one-line promise.
Avoid the most obvious and predictable ideas of the sector.
Feedback: if it gives back titles that are interchangeable from one sector to another ("5 tips for...", "the ultimate guide to..."), the context was too thin. Add a concrete detail about the audience and try again.
Once you've chosen the idea that convinces you, the second round:
Take this idea: [paste the chosen title].
Develop it into 3 different versions by angle:
one emotional, one based on data, one with a real case.
For each, give me: opening hook, 3 central points, closing with CTA
(Call to Action, the final invitation to act).
Concrete example
Marta runs a small plant shop and her Instagram calendar has been empty for two weeks. She opens Gemini for free and pastes the first prompt: audience = city people without a green thumb, product = easy-to-care-for plants, channel = reels and carousels, tone = ironic and reassuring.
The first five ideas are predictable ("the 5 most resilient plants"). But in the contrarian group one pops up that strikes her: "Stop watering your plants — 70% die from too much water, not too little." She develops it in the second round by asking for the data-based version.
Result: a six-slide carousel with the overwatering hook, three signs to recognize a "drowned" plant, and the CTA to save the post. Total time: eleven minutes, two requests. From an empty calendar to a post ready to pass to the designer.
When it does NOT work (and how to fix it)
If the ideas are all generic and interchangeable
It means the AI doesn't have enough of your material. Don't ask it to "make them more original": that's a vague command. Instead, give it a constraint a competitor couldn't have: an anecdote of yours, a data point from your sector, a phrase your customers often say to you. Originality comes in through specific context, not from the adjective "original."
If it keeps proposing the same things
You're exhausting the vein in a chat that's now long. On Claude, long conversations burn through the quota faster, because the model keeps the entire history as context, up to 200K tokens. Open a new chat and restart from the prompt with clean context: by changing the starting point, the outputs change. On Gemini use the temporary chat, which explores an idea outside your usual style without saving it in the history.
If you run out of free messages mid-session
On ChatGPT free the wall comes early. When you hit the limit, or during high-traffic hours, OpenAI falls back to a reduced model: you notice because the answers become less detailed. Solution without paying: keep a second tool open (Gemini or Claude) and move the brainstorming there when one gets blocked. The limits reset on different windows, so by alternating them you work continuously.
If the ideas are creative but unworkable
The AI doesn't know how much time and budget you have. Add the constraint in the prompt: "each idea must be doable by one person alone in two hours, with no budget for equipment." It turns abstract creativity into things you can actually produce tomorrow.
A tip from someone who really uses it
Save your base prompt with the context already written and reuse it every week, changing only the last block. On Claude you can create a Project (a workspace folder that keeps instructions and context files) dedicated to your content; on Gemini there are Gems (reusable personalized assistants) and on ChatGPT custom GPTs. You configure audience and tone once, and every subsequent request starts already calibrated. The difference between mediocre ideas and good ideas isn't in the model: it's in not starting from scratch every time.
Frequently asked questions
What's the best prompt for getting original ideas?
There's no magic formula: originality comes from the specific context you provide, not from the word "original" in the prompt. The most effective prompt describes audience, goal, and what you've already done precisely, and forces the AI to generate ideas for different angles. Use the one in the "How to do it" section as a skeleton.
Can I do this for free or do I have to pay?
For generating ideas, free is more than enough. The free plan handles quick questions, rough drafts, and single sessions. The real limit isn't quality but volume: if you brainstorm for an hour straight you run out of free messages. Alternate two free tools and you'll rarely need to subscribe just for ideas.
Is it better to ask for a few curated ideas or many ideas to skim?
Many, always. Asking for three ideas leads to three cautious, average answers. Asking for fifteen forces the AI to get past the obvious and reach the unusual angles, which sit at the bottom of the list. You do the culling in ten seconds: that's your job, not its.
Is it worth brainstorming by voice instead of writing?
If you freeze in front of the blank page, yes. Gemini Live allows a real-time dialogue where you can interrupt the AI, and for many it feels like a phone call with a colleague: useful for getting out a complicated idea when you don't feel like writing paragraphs. Talking freely brings out associations that you'd censor while typing.
Aren't the AI's ideas already seen by everyone, since it uses the same model as the competitors?
That's the serious objection, and the answer is clear-cut: the raw idea maybe yes, but the final product no. Two people who give the AI the same vague prompt get similar ideas. Two people who give different context — their audience, their anecdotes, their tone — get incomparable results. The AI doesn't replace your specificity: it amplifies it. Whoever stays generic gets the generic; whoever puts themselves into it comes out with something no competitor can replicate.