What this covers
Gemini’s real brainstorming edge is not that it generates more ideas faster — every LLM does that. It is that it can pull in the half-written Doc, the deck from last quarter, and the customer-feedback Sheet you already have in Drive, then frame the problem against your actual context. This guide turns that integration into a repeatable 30-minute session.
Key tools and concepts:
- Gemini: Google’s multimodal AI assistant, deeply integrated with Workspace and Search.
@mentions: reference a specific Doc, Sheet, or Drive file inline so Gemini grounds its ideas in your material.- Framings: different ways to phrase the same underlying problem; the cheapest way to escape the first-idea trap.
Who this is for
Anyone deciding or planning — PMs picking a feature name, founders sequencing a launch, marketers framing a campaign, writers stuck on the angle for a piece. If you brainstorm alone with a blank Doc more than twice a week, this workflow is for you.
When to reach for it
Strategy, naming, framing, sequencing — anywhere the cost of picking the wrong direction is bigger than the cost of generating ten more options. Skip Gemini for hard analytical problems with a single right answer; reach for it when the space is wide and your first three ideas all feel obvious.
Before you start
- Write a one-sentence problem statement. If you cannot, the brainstorm will be unfocused — the model will pattern-match to whatever sounds adjacent.
- Pull the supporting artifacts into Drive: customer quotes, prior decks, the half-baked Doc. You will
@-reference them, not paste them. - Decide your output format up front: a ranked list of 10, a 2x2 matrix, three framings with a one-paragraph drill-down each. Format constrains the model to useful structure.
- Time-box to 30 minutes. Brainstorming with an LLM has no natural stopping point; the timer is yours to enforce.
Step by step
- State the problem in one sentence, then add a constraints line: “We must ship before September; the audience is mid-market ops leaders; we cannot use the word ‘platform’.”
- Ask for 10 framings, not 10 solutions: “Give me 10 different ways to frame this problem. Vary the unit of analysis — user, system, business, time.” Framings break the first-idea trap better than solution lists.
- Pick 3 framings that feel non-obvious and drill: “For framing 4, what are 5 concrete approaches? For each, name the strongest objection.”
- Push back on the weak spots: “Approach 2 assumes attribution is solved — defend or replace it.” Refusal to defend is a signal to drop the idea.
- Reference your real material: “Now reconcile approaches 2 and 5 with the customer quotes in
@Q3-feedback-summary. Which holds up?” - Close by asking for the inverse: “What would have to be true for none of these to work?” The premortem catches the framing-level error.
First-run exercise
- Pick a real decision you owe someone a recommendation on this week. Not a thought experiment.
- Run all six steps in one sitting, no notes app, just the Gemini conversation.
- Save the conversation transcript into a Doc. Highlight in green the lines that genuinely surprised you, in red the ones that read like LinkedIn filler.
- Re-run only step 2 with a different framing axis (“vary the timeframe instead of unit of analysis”) and compare the green-highlight count.
Quality check
- Did the session produce framings you would not have reached alone, or just polished versions of your starting point?
- Are the strongest two ideas grounded in something specific — a real quote, a real number, a real constraint — or in generic best practice?
- Did you press on the weakest spot of each finalist, or did you fall in love with the first decent option?
How to reuse this workflow
- Save the prompt sequence as a “brainstorm pattern” snippet. The structure travels even when the topic changes.
- For recurring decisions (quarterly naming, weekly campaign framing), keep a “framings I have used” log so you do not converge to the same axis every time.
- Keep the failures: brainstorms where Gemini produced ten variations of the same idea usually mean your problem statement was too narrow.
- Refresh quarterly — model defaults shift, and a framing prompt that worked six months ago may now return cliches.
Recommended workflow
Problem statement (1 sentence) → 10 framings → pick 3 → drill 5 approaches each → reconcile with one @-referenced source → premortem. End with a single Doc that lists: the chosen direction, the two runner-ups, and the assumption that would kill it. That Doc is the artifact you share with a teammate to get a real second opinion.
Common mistakes
- Stating the problem too generically (“how do we grow”) — every LLM will hand you LinkedIn fluff.
- Stopping after the first 3 framings because they sound smart. The 7th-10th are where the non-obvious moves usually live.
- Ignoring the
@integration. A brainstorm grounded in your real customer quotes is 5x more useful than one grounded in averages. - Asking Gemini to “decide” — it will pick the option that sounds safest in writing, not the one that fits your context.
- Skipping the premortem. The framing-level error is the most expensive one, and the only one a fast brainstorm can catch.
FAQ
- Is Gemini better than ChatGPT for brainstorming?: Roughly tied on raw idea generation. Gemini wins when your context lives in Drive; ChatGPT wins when you have heavy custom instructions and persistent memory.
- Should I use Deep Research for brainstorming?: No — it is for synthesis, not divergence. Use plain chat for brainstorming and switch to Gemini Deep Research once a direction is chosen.
- How do I keep Gemini from being generic?: Constraints, examples, and
@-referenced material. Each one cuts the cliche rate roughly in half. - Can I brainstorm with my team using Gemini?: Use a shared Doc with Gemini comments rather than the chat — multiple humans plus one Gemini in the comments is the team-friendly variant.