You do not pick a mode.
You just start typing.
The newsroom opens on one field. Type a line and it is an article. Press Enter for a second line and it is an assignment. Empty the rows and it folds back. What the desks are handed follows what you actually typed - a blank row you have not filled is still one article, not two.
Everything below runs the same sealed nine-desk pipeline. The doors differ only in how the brief is assembled before it: one article is one run, whichever way it came in. New here? Start with how it works.
What you give it, and what comes back.
Article
- Give
- One keyword or headline
- Get
- One article
Type a line. That is the whole interaction - committing a topic does not open a second screen, it opens the brief in place: the suggested headline, what the competition is doing, the secondary keywords, the docket.
Assignment
- Give
- A list of topics, each with an outline if you want one
- Get
- One article per row
Press Enter for a second line and the front door becomes the assignment where it stands. Delete back to one line and it is a single topic again. Supply your own outline per row and the writer keeps it.
Blueprint
- Give
- A reusable structure crossed with your own values
- Get
- One article per combination
Save a structure once, save a table of values once, then cross them. Values paste in or arrive as CSV and load into any Blueprint run.
From a page
- Give
- A URL you already have
- Get
- That page refreshed, rewritten, or beaten
Point it at something you already published. Nothing here is sent unread: the pages are fetched and shown to you first, with the headline we found, the domain, the word count and the date, and a page we could not read is named and left out rather than generated against.
Head to head
- Give
- Two subjects
- Get
- One comparison
A shaped format over the same pipeline. This is one of two ways in that has been run all the way to a finished article in verification, rather than only up to dispatch.
Ranked list
- Give
- A subject and a count
- Get
- One ranked list
Say what to rank and how many. The shape is handled; the sourcing is the same nine desks.
Explainer
- Give
- A term, or a list of them
- Get
- One definition piece per term
It used to ask you to classify first, with a switch above the field. It does not: type a term and it is one piece, press Enter for a second and it is one article per term.
Look before you spend an article.
Find sources
LiveType a few words and get back real sources from the web, ranked and judged, without spending an article. It is the same sourcing the nine desks do before they write, exposed as its own door - so you can see what the desk would write from before you commit. Each result says why it ranked. Zero sources is an answer, not an error, and it says which reason it was.
Research
On a paid planOn-demand research reports: hand it a subject and it comes back with what is worth covering, backed by sources it actually read. A different thing from Find sources, deliberately kept apart - this finds you subjects you did not have, that finds sources for a subject you already have. Acting on a suggestion is the normal article flow; nothing is spent from inside a report.
Upload a list you already have
LiveSeveral files at once, in .txt, .csv, .xlsx, .docx, .pdf or .json. On From a page the same upload reads a sitemap: give it your sitemap address and it lists what the site publishes, narrowed to a path and to pages untouched in the last 3, 6 or 12 months. It reports how many the sitemap listed, how many survived the filters and how many are being taken, because a filter that quietly ate the list looks exactly like one that worked.
Five angles a topic keeps coming back to.
Not an SEO field
Each topic a domain covers can carry up to five keywords, set beside the topic list. They steer the search queries first, then the outline, then the prose, in that order. Nothing is sprinkled in afterwards to hit a density.
Any angle the sources cannot support is dropped rather than padded in. That is the whole reason they sit in front of the search instead of behind the draft.
Per topic, never per domain
Finance keywords are not useful in a garden section, so an article is written against one topic and sees only that topic's keywords. A topic with none is written from the topic alone, which is a full state rather than a gap.
Renaming a topic starts its keywords over. The name is the topic's identity everywhere else too, so the two cannot drift apart quietly.
An article you ask for by hand has no topic behind it, so it uses the per-run keyword panel on the writing surface instead. Topics themselves belong to a domain's newsroom setup.
Share it, and get paid every month they stay.
Your team gets a referral code and a page of its own in the dashboard. Someone who signs up through it gets 10% off their first payment. You get 30% of their plan's monthly article-credit allowance, in credits, every month, for up to twelve months.
It reads their current plan every month, so an upgrade mid-term raises what you earn and a downgrade lowers it. Nothing is granted until money actually moves, and only a subscription payment starts the twelve-month clock - a credit-pack purchase does not. Capped at fifty paid referrals per team. Referring yourself does not work, and we check properly: tags and dots are collapsed before the comparison.
The dashboard shows your clicks, your signups, who subscribed and what you have earned. You are told in-app each month a grant lands.
What is in the rail marked Soon.
These are named in the product and are not clickable. Naming them is a stated intent, not a delivery date. They are listed here for the same reason they are listed there: a door you can see and cannot open is honest, and a door nobody mentions is not.
From a video
SoonTurning a video into an article. Named in the rail, marked Soon. The open question is the transcript source, so the next step is a measurement rather than a build.
One thing is deliberately absent rather than marked Soon: AI-search visibility tracking. That is someone else's product, and Mastheads is not an SEO tool suite - no keyword research, no topical maps, no rank tracking. That is a boundary, not a gap.
- Do I have to pick a mode before I start typing?
- No. The app opens on one field. Type a line and it is an Article; press Enter for a second line and it is an Assignment; empty the rows and it folds back. What the desk is handed follows what you actually typed, not what the editor looks like.
- Are the different ways in different products?
- No. Every one of them runs the same sealed nine-desk pipeline, and they differ only in how the brief is assembled before it. One article is one run of the pipeline, whichever door it came through.
- Do I need a website to use any of this?
- No. On-demand writing belongs to your profile and needs no web address at all - you sign up and write, and your article defaults live on your profile. A domain is a different thing: a newsroom around a real website, which starts from your web address and a scan.
- What happens to something I typed but did not send?
- It is not thrown away by navigating. Leaving a writing surface with work on it asks, and keeping it saves it as a list of its own, then lets you go where you were going. Each surface offers back only what it parked, because a list of addresses is not a list of topics.
- Is a run of 30 articles cheaper than 30 single articles?
- No. A run of N articles is N full generations, not one cheap batch - the desks run per article, so it costs and takes N times a single article. Runs are capped so a typo cannot start hundreds, and what the cap holds back stays in the box as your next run.
Pick a door, or just type.
Your first week is free - 10 articles, no card required. Signup takes no payment details at all.