Create
Create is where an idea becomes a draft, written in your voice, which then opens in the editor. It does not publish, schedule or post: it ends at a draft, which you revise, export or send to your CMS.
Two ways to start
Create (/create) is one page, whichever way you open it: Create on the rail (or N), Create new on Drafts, the picker (press and hold Create on the rail, or press C, then choose Blog, X, LinkedIn, Newsletter or Just chat with the arrows or 1 to 5), or Ask Viewfield's "Create a blog post" rows. The picker and ⌘K open it with that format already pressed; every other way opens it with none.
The page is New draft and one large box. Say what to write, as briefly or as fully as you like: a single line, or a brief with links (up to 20,000 characters you type). The box grows as you write and then scrolls, and a short paste keeps its paragraphs; a paste from a web page, a Google Doc or Word keeps its headings, lists, bold and links too, as Markdown.
A long paste becomes a card. Paste a document (about 1,500 characters or more, or eight lines or more of text: a whole article, a transcript, a report, a brand document) and it does not fill the box: it becomes a Pasted text card beside your attachments, showing its first lines and how many lines it has, and your words and cursor stay where they were. Press the card to read the whole text (with Copy), and its × to remove it. A second paste is Pasted text #2. A card is sent whole with your words, and the writer reads it the way it reads an attached file; you can send a card with no words at all, and Viewfield writes from it. One paste can be up to 1 MB, and your words and pasted texts together up to 250,000 characters (about 40,000 words); past either, the box says so and nothing is sent or cut.
The box's bottom row holds Attach (up to five attachments in all, files and pasted texts together: text, Markdown, CSV, PDF, Word, Excel or Numbers, 10 MB each, each shown as a chip you can remove), Research first, the model picker (Auto unless you choose a model your plan includes) and the round send button. The formats sit under the box (see The formats). Press Enter or the send button and the draft is created: the page moves to its loader, then to the editor. Shift+Enter is a new line.
The box on Home creates at once, the same way, with the same Attach, pasted-text cards, model and formats, from a smaller box; Just chat there opens a conversation with your words (and what you pasted or attached) instead of a draft.
AI visibility opens Create too: Write it on a question, and a suggestion's Write the post, arrive with the item attached ("Writing for the question ...") and the box is a note to the writer. Nothing is written until you send.
A conversation with Ask Viewfield (from Just chat, ⌘K, or a past chat) also lives at /create, in the same frame: ask anything, attach files and images, and Ask drafts a piece when you ask it to, after you confirm its card.
Research first
Research first sits beside Attach. It reads any link in your words and searches the live web, then brings the findings back as Research notes under the box, each with its source. Keep the notes you want and remove the rest; Research again runs it afresh. When you send, only the notes you keep reach the writer. The writer reads what you sent and nothing else, so a fact it did not get from your brand context, a file or a kept note never arrives on its own. Research spends from the workspace's allowance. Research is a press you choose, never something that runs by itself.
The formats
Every composer offers the same format chips: Blog, LinkedIn, X (one X post) and Newsletter, then any custom kinds your team has added on Brand › Kinds (on Create). None is pressed until you press one, and pressing it again clears it.
With no format pressed, Viewfield decides:
- The format your words name. "Write an X post about our launch", "a LinkedIn post", "a thread on how we cut churn", "Newsletter: what shipped in October" or "an article" each get that format. A format inside the subject is not a request: "a blog post about LinkedIn" is a blog post, and "write about why newsletters matter" names none.
- Otherwise, the best fit for the request, chosen by one small model call (a fraction of a cent, from the workspace's allowance): a single sharp point suits a post, a walk-through of steps suits a thread, an argument with evidence suits a blog post, a roundup for subscribers suits a newsletter.
- If that call fails, a blog post.
A chip you press always wins over the words. The loader says the format once it is decided ("Writing your X post").
The kinds
Every kind is the same editable document. The kind decides what the writer is asked for and the tone it writes in (the rule for that kind on Brand › Kinds); it never changes the editor.
| Kind | What the writer is asked for |
|---|---|
| Blog post | An H1 that states the claim, sections under H2s, usually 800 to 1,500 words. The slug, summary and SEO fields are suggested later, when you send it to your website |
| LinkedIn post | A post whose first two lines earn the click, in short paragraphs, at most 3,000 characters |
| X post | One post, under 280 characters |
| X thread | Numbered paragraphs in one document ("1/", "2/") |
| Newsletter | A subject line, a preview line and the body: one main story and one call to action |
| Custom | A kind your team added on Brand › Kinds (release notes, an investor update), with its own rule |
The draft
The piece is written from one packet, the same for every kind:
- your voice, with the four dials;
- your brand basics and your positioning statement;
- your writing rules: the words you never say, the standing rules, the tone;
- the rule for its kind;
- your docs (always-read docs in full, up to 24,000 characters in all; the rest found by search);
- the notes you kept;
- the text you pasted as cards, whole, each under its label, and the files you attached (each file's text, up to 30,000 characters, under its name); the writer uses their facts and adds no figure they do not state;
- for a piece started from AI visibility: the question, the companies the answers name and the pages they cite.
Sending (from Create, or Go on Home) takes you to the draft's loader at once, titled with your words. It names what the writer is reading (your brief with the notes you kept, the files you attached and the text you pasted, what you sell, your voice, your rules and never-say words, the rule for the kind, your always-read docs by name), one step at a time across the time a draft of that kind usually takes in your workspace, so the last step, Writing, begins a little past halfway. While it writes, the words show as they arrive and the bar keeps moving, and then it hands over to the editor. You can leave: the draft keeps writing, its card on Drafts reads Writing, and a note says "Your draft is ready" with Open when it is done. If a draft does not finish, the loader says so and offers to try again.
Every draft is checked as it lands: the checks on Brand (anti-slop is on by default) lint the draft and run one targeted repair. The checks never outrank your voice: a word your lexicon loves is never flagged. To check the draft's factual claims, press Check claims in the editor.
Iterating
In the editor, Update with AI takes a change for the whole piece ("tighten the hook", "angle this at finance leads", "cut the second section"): the rewrite lands in the document and asks Keep or Revert. Inline revisions on a selection are the editor's own: see The editor.
A question about the draft goes to Ask Viewfield (⌘K), which carries the open draft with it. A change it proposes comes back as a card with Put it in the draft, which lands in the editor with the same Keep or Revert.
What happens next
- A blog post or a custom piece goes to your CMS from Publish in the editor: a draft first, then a separate press to publish. See Publishing.
- An X post or thread, a LinkedIn post or a newsletter leaves as text: its Publish tray copies it (Copy as text strips it to what that network's composer accepts), exports it, and records where it went with Mark as published. Mark it Final with the Draft and Final switch when the team is done with it.
Over MCP and in chat
draft_piece writes one piece from a prompt or from an AI visibility item (item_type of aeo_prompt or aeo_change with its id) and returns the draft's id; leave out kind and Viewfield picks the format the same way the composers do (an item's draft without one is a blog post); get_generation_status says when it is done and get_content reads it. In Ask Viewfield, research is web_research and read_url. See MCP overview and Ask Viewfield.