TL;DR
Anyone running a blog, newsletter, or channel hits the topic wall within six months. The fastest fix in 2026 is a three-model relay: research the niche in Gemini 3.1 Pro (live search), generate 30 angles in ChatGPT (GPT-5.5, fastest divergent output), then have Claude (Sonnet 4.6) pressure-test the shortlist. The prompt below produces 30 working titles grouped into six intent buckets in one pass. Your only real job is picking the five you would be proud to ship — and that part AI cannot do for you.
The problem AI actually solves here
You sit down to write and realize you have already covered the obvious posts, and the next ones feel forced. AI is genuinely useful at this exact moment: not for picking the topic, but for surfacing the 20-some candidates you would not have reached in a solo 20-minute brainstorm. Solo brainstorms cluster around three or four mental ruts; a structured prompt forces breadth across reader intents you would otherwise skip.
When AI is the right tool — and when it is not
| Use AI when | Skip AI when |
|---|---|
| You have a defined niche (broad enough for an audience, narrow enough to picture one reader) | You are still deciding what the publication is about |
| You have published 10+ pieces, so you can feed real examples of your angle | You have no body of work yet — write a few posts first, then return |
| You feel stuck and the well is dry | Topics are already flowing; do not interrupt momentum |
Two failure modes to plan around:
- AI regresses to the niche mean. Models lean toward the content their training data is densest with, so you will see titles you have already seen elsewhere. Filter aggressively for the only-you-could-write-this angles and discard the rest.
- AI is blind to timing. It does not know a topic spiked this week. Before you finalize, layer in two or three angles from real-world signal: a recent news event, a question a subscriber emailed, a thread that blew up in your community.
Which model for which step (June 2026)
The strongest 2026 workflow is not one model — it is a short relay that plays to each model’s edge.
| Step | Model | Why | Cost to start |
|---|---|---|---|
| 1. Map the niche, surface trends and gaps | Gemini 3.1 Pro | Live Google search built in; fastest way to see what is already covered and where the white space is | Google AI Pro $19.99/mo (free tier exists with tight limits) |
| 2. Generate the 30 angles | ChatGPT (GPT-5.5) | Quickest divergent output; reliably the most “out of the box” titles for the volume phase | ChatGPT Free $0 (limits), Go $8, Plus $20 |
| 3. Pressure-test the shortlist | Claude (Sonnet 4.6, 1M context) | Strongest at honest critique and tightening; paste your past posts plus the shortlist and ask for overlap and weak angles | Claude Free $0 (limited), Pro $20 |
If you only have one subscription, run all three steps in it — the prompt still works. The relay just sharpens each phase. All three free tiers are enough to test the workflow before paying.
What to feed the AI
Give the model signal about your angle, not just your topic:
- Niche in one sentence
- Audience in one sentence, with one concrete detail (location, role, life stage)
- Tone (calm, contrarian, warm, technical)
- Three past posts that performed well (title + one-line summary each)
- One post that under-performed, so the model learns what to avoid
Copy-ready prompt
Replace each [bracketed] placeholder with your own detail, then paste into any of the three models:
Brainstorm 30 content topic angles.
Niche: [niche]
Audience: [audience]
Tone: [tone]
My best-performing past posts:
- [post 1]
- [post 2]
- [post 3]
A post that underperformed: [underperformer]
Group the 30 angles by reader intent, with these exact buckets:
1. Educational ("what is X")
2. How-to ("how to X")
3. Comparison ("X vs Y")
4. Opinion / contrarian ("why most X is wrong")
5. Story / case study ("how we did X")
6. Resource / list ("10 X for Y")
Five angles per bucket. Each angle is one line, written as a working title.
After the list, mark the 3 that most overlap with my best-performing posts,
and the 2 that feel most generic (likely seen everywhere).The six-bucket shape forces variety across intents, which is the entire reason to use AI here. The closing instruction makes the model flag both its strongest matches and its own filler, which saves you a pass.
Filtering 30 down to five
The model hands you raw material. The shortlist is yours:
- The competitor test. Would you click this title if a competitor published it? If no, cut it.
- The SERP test. Search each surviving title. If the top 10 results already cover it well, you need a sharper angle or a personal slant they cannot copy.
- The lived-experience test. Keep only what you have actually done or thought hard about. Forced posts read forced, and readers can tell.
Aim to ship five to seven of the 30 over the next quarter. The rest is not waste — it sharpens your sense of the niche and feeds the next round.
Common mistakes
- Asking for 100 angles. Quality collapses past 30; you get padding.
- Picking topics you have no lived experience for. The reader can always tell.
- Generating everything in one tone. You want range across intents, not five variations of the same voice.
- Treating the list as a content calendar. It is raw material; you still need to sequence and pace it.
Keep a topic graveyard
Keep a single file of angles you almost wrote but didn’t. Re-read it monthly. Some topics need six months to become writeable, because you have to live the experience first. AI will not notice that — your graveyard will. When you have a stack of survivors, move them into a real plan with the content calendar workflow and group them into themes using content pillar planning.
FAQ
- Which AI is best for brainstorming in 2026? For raw idea volume, ChatGPT (GPT-5.5) is the fastest at divergent output. For seeing what already exists in your niche, Gemini 3.1 Pro’s live search wins. For honest critique of a shortlist, Claude (Sonnet 4.6) is the most useful. Pick by the step, not by overall ranking.
- How many of the 30 angles do I actually use? Aim to ship five to seven over the next quarter. The rest informs your sense of the niche and seeds the next session.
- Should I tell the AI my competitors? Yes — one line per competitor describing their angle. The model then pushes toward the white space they are not covering.
- Can I run this for video as well as written content? Yes. Add the format constraint to the prompt (length, platform, hook style) so the working titles fit the medium.
- Does the paid tier help for brainstorming? Not much for this task. The free tiers of ChatGPT, Claude, and Gemini all generate 30 angles fine. Paid tiers matter more once you move into long-form drafting against full context.