HOW IT WORKS

From a keyword to a content plan

No jargon required - here's what happens, in order, when you type in a keyword.

The 10 steps

Think of it as four specialists on a content agency team, each handing their work to the next

1
Clean up your keyword
We tidy up what you typed - fix spacing, lowercase it, figure out which country/language it's for.
Behind the scenes — Bang Maulana types “Padel Jakarta”, picks Indonesia + Local Service. We clean it up to padel jakarta and resolve the market to Indonesian (id).
2
Ask Google what people actually search
Google Suggest
We ask Google's autocomplete for real queries related to your keyword - the same suggestions you see typing into the search box, just gathered in bulk.
Behind the scenes — ~10 parallel requests go out: padel jakarta, sewa padel jakarta, harga padel jakarta, padel jakarta terdekat... Google hands back real queries like sewa lapangan padel jakarta selatan and harga sewa padel per jam.
3
Remove duplicates & rank by popularity
Duplicate queries get merged. The more often a query shows up across our searches and the higher Google ranked it, the more weight it gets.
Behind the scenes — sewa lapangan padel jakarta came back from 3 different searches and ranked #2 each time - it rises to the top of Bang Maulana's keyword universe.
4
Spot obvious buying signals
Words like "harga" (price) or "beli" (buy) are flagged by a simple word list, before any AI gets involved - this part is just pattern matching, not guessing.
Behind the scenes — sewa lapangan padel jakarta matches “sewa” → tagged transactional. cara main padel matches “cara” → tagged informational. No AI has run yet.
5
Work out what people want
AI - SEO Strategist
The AI reads every query plus the word-list hints and decides: are people mostly researching, comparing, or ready to buy? It also explains any time it disagrees with the word list.
Behind the scenes — The AI reads all of Bang Maulana's queries with their tags attached and estimates the mix: ~50% transactional (booking a court), ~30% local (nearest venue), ~20% informational (how padel works).
6
Group into topics
AI - SEO Strategist
Related queries get bundled into a handful of topics worth writing about - grouped by what the searcher is trying to get done, not by shared words.
Behind the scenes — Queries get grouped into “Court Booking & Pricing”, “Nearest Venues”, and “Learning to Play” - plus a note that nobody's query addresses court size or racket rental, a gap worth filling.
7
Figure out who's searching and what to publish
AI - Content Strategist
For each topic: who is likely searching, how close they are to a decision, and what kind of page (guide, comparison, pricing page) fits best.
Behind the scenes — Persona: “a Jakarta professional booking a court with friends after work.” “Court Booking & Pricing” gets mapped to bottom-funnel - a venue/booking landing page, not a long-form guide.
8
Brainstorm content ideas - and reject the boring ones
AI - Creative Director
The AI pitches several content ideas with a hook and a title, scores them, and also lists ideas it deliberately rejected and why - the same way an agency creative director would. It closes with a plain-English "Director's Note": one or two sentences, no jargon, saying which single idea it would bet on and why.
Behind the scenes — Pitched: “Padel Court Booking Guide: Jakarta's Best Indoor Venues.” Rejected: “What Is Padel? A Beginner's Guide” - killed because this keyword set is overwhelmingly people ready to book, not people who don't know what padel is.
9
Write the brief
AI - Content Leadonly when you pick one
Only happens once you pick one idea. Produces a full brief a writer can start from today: outline, title, FAQs, links to include.
Behind the scenes — Bang Maulana picks the booking guide and clicks Generate Brief. Out comes an outline, meta title, FAQs like “Berapa harga sewa lapangan padel per jam di Jakarta?”, and the entities a writer must cover.
10
Save your work
Every step is checked and saved as it finishes, so if anything fails partway, you keep everything that already worked and can retry just the broken part.
Behind the scenes — Every step above was saved to the database the moment it finished. If Bang Maulana's wifi drops right after step 6, reloading the page picks up exactly there - nothing re-runs, nothing gets billed twice.

What we refuse to make up

We don't show search volume, CPC (cost per click), or “keyword difficulty” scores - the numbers most keyword tools show you. We don't have real access to that data, and a confident but made-up number is worse than showing nothing, because you'd plan around it as if it were true.

Instead, everything on screen is labelled with where it actually came from:

SUGGESTReal data - an actual query Google suggested, and how often/high it showed up.
RULESimple word matching - flagged by a fixed list of words in code, no AI involved, so you can check it yourself.
AIAI's opinion - a judgment call, clearly marked as one, not a fact.
MAUSCOREA calculated score - worked out from a formula we show you, not hidden inside a model.

Why doesn't the AI just decide the ranking itself?

The AI is good at judging three things: how strong the buying intent is, how different an idea is from what already exists, and how much work it'd take. It does not get to pick the final order - AI models are unreliable at doing consistent math across a list, and “why is this idea ranked first?” deserves an answer that's the same every time you ask.

Opportunity Score = 0.30 x demand signal (Google Suggest) + 0.30 x intent value (AI) + 0.25 x differentiation (AI) + 0.15 x (100 - effort) (AI)

Almost a third of every ranking is grounded in real Google data, and each idea card shows exactly which part.

What's actually running this

The “thinking” steps are handled by an AI model called Nemotron?, reached through a routing service called OpenRouter?. Right now this workspace is running on the live AI model

Saved runs and briefs live in a real database. The keyword search itself (step 2, Google Suggest) always uses live data either way - it doesn't need an AI model or a database to work.