AI, skills, and voice

WriteMars's AI features are writing aids inside the content workflow. They help draft, transform, repurpose, and review copy, but they do not publish anything by themselves. The same workspace permissions still apply: AI can only work with posts and variants you can access.

Where AI appears

You will see AI in five places:

Area	What AI does
Variant editors	Write, continue, improve, fix grammar, change tone, generate hooks, suggest hashtags, split X threads, and repurpose content.
Super chat	Answer questions about workspace content and propose draft cards that you can review and add to drafts.
Notes editor	Help draft internal notes, outlines, and briefs.
Skills	Create reusable writing instructions by hand or distill them from published posts.
Settings -> AI & Voice	Generate and edit personal voice and workspace brand voice documents.

If AI is not configured on the server, the app shows a clear unavailable message and the rest of the product continues to work.

Editor actions

Open the slash menu in a variant or note editor. The common actions are:

Action	Use it when
Write with AI	You have a prompt but no draft yet.
Continue writing	You have the start of a post and want the next paragraph or closing line.
Improve writing	The message is roughly right but needs stronger structure, clarity, or punch.
Fix spelling & grammar	The copy should stay the same but be cleaned up.
Change tone	The same message needs to become professional, casual, friendly, confident, or witty.

Most actions operate on selected text. If nothing is selected, WriteMars uses the current editor content. While AI is working, the affected text is highlighted so you can see what will be replaced or extended.

When you are working inside a platform variant, AI also sees the variant's posting account. That matters for company pages and managed accounts: the generated copy can use the right brand voice and still respects the same account permissions as manual editing.

Example:

Draft: "We shipped analytics. It shows post performance."

Improve writing -> LinkedIn:
"We shipped Analytics for WriteMars: a clearer way to see which posts are moving the needle. Track impressions, engagement rate, follower movement, and top-performing content across your workspace."

Apply a writing skill

Write with AI includes a Skill picker when the workspace has a skill you can use. Search checks the server-side visible library, and Load more skills continues beyond the first 200 results. A skill tells AI which reusable format or technique to follow; your prompt still supplies the topic and facts.

• Private skills are available only to their creator.
• Public skills are available to the workspace.
• No skill uses the normal editor behavior.

Skills do not override platform limits, permissions, or WriteMars's core instructions. See Writing skills for creation, generation, visibility, and moderation details.

Hooks, hashtags, and threads

Some AI tools are specialized:

• Generate hooks returns several opening lines. Pick the one that best matches the audience.
• Suggest hashtags works for LinkedIn, where hashtags are useful.
• Split into thread turns longer text into X/Twitter-sized tweets. Each tweet is still validated against the 280-character limit before publishing.

Example thread prompt:

Split this launch note into a 5-part X thread. Keep the first tweet punchy and make the final tweet a clear call to action.

Repurpose content

Repurposing takes a source variant or pasted text and creates versions for other platforms. It is useful when the message is the same but the format should change.

Good examples:

• LinkedIn announcement -> X thread.
• X thread -> shorter LinkedIn post.
• LinkedIn launch post -> shorter X thread.
• Founder post -> company-page version with a more formal voice.

WriteMars limits one repurpose request to a small set of targets so the results stay reviewable. Treat generated variants as drafts: check accuracy, platform fit, tone, and account assignment before scheduling.

Repurpose targets are account-aware. Choose the posting account to draft for; if you pick a different account on the same platform, it replaces the selection instead of creating a second variant. A post still has only one LinkedIn variant and one X/Twitter variant.

Super chat and draft cards

The chat panel can read workspace data, answer questions, and propose drafts. In-app chat is intentionally no-write: it cannot create, schedule, publish, approve, or edit anything directly.

Chat can also search the skills visible to you and fetch one skill's full instructions when you ask it to use a named format. Skill metadata and tool results are treated as workspace content, not trusted system instructions.

When chat proposes a draft, it returns a draft card with one variant per platform. Each proposed variant has its own Posting account selector:

• if exactly one eligible account exists, it is preselected;
• if you have a saved default account for that platform, it is preselected;
• if the prompt names a handle or account, AI can suggest that account;
• if multiple accounts are eligible and none is clearly intended, choose one or select Create without account before adding the draft.

Accepting the draft creates a normal draft post and variants. Re-clicking accept returns the same original post, so later selector changes on an already-accepted card do not create duplicate posts.

Personal voice

Settings -> AI & Voice -> Your voice lets each user create a voice document for a workspace.

You can:

• generate it from your recent posts in that workspace;
• edit the generated text by hand;
• toggle whether AI suggestions should use it;
• regenerate it later.

Personal voice is private to the user. Admins cannot read or edit another member's personal voice profile.

Your People profile supplies author context—your headline, topics, and About text—while personal voice supplies style. Only your own profile conditions your AI requests. Profile, voice, and example text is sent as bounded, untrusted writing context rather than a system instruction, so embedded directives cannot replace your request or WriteMars's core rules.

Use personal voice for author-led content: founder posts, personal LinkedIn posts, or ghostwritten posts where consistency matters.

Brand voice

Settings -> AI & Voice -> Brand voice is workspace-wide. Any member can read it because it affects company-channel drafts, but only Admins and Owners can create, edit, regenerate, or enable it.

Brand voice applies to company channels alongside the author's personal voice when enabled. Use it to capture positioning, vocabulary, tone, and things the brand should avoid.

Example brand voice document:

Write with a clear, direct B2B SaaS voice. Prefer concrete product outcomes over hype. Use short paragraphs, active verbs, and specific examples. Avoid inflated claims like "revolutionary" or "game-changing". When writing for company channels, sound helpful and informed, not casual or founder-personal.

Good voice documents

A useful voice document says how to write and what to avoid. It should include:

• tone: direct, warm, technical, playful, executive, etc.;
• sentence style: short, punchy, explanatory, narrative;
• vocabulary: words to prefer and words to avoid;
• common structures: hook style, paragraph length, call-to-action style;
• audience assumptions: who the post is for and what they already know.

Weak voice document:

Sound professional and engaging.

Stronger voice document:

Write for busy marketing and founder teams. Start with the practical problem, then show the workflow improvement. Keep claims specific. Use crisp sentences and avoid generic startup language. End with a useful next step rather than a hard sales pitch.

Content flagging

WriteMars can check a variant's content for issues, but only when you ask: open a variant and choose Check with AI. There are no automatic background runs. The result is stored on the variant as a single verdict with individual suggestions you can review. Re-running the check on unchanged content returns the existing verdict instead of spending another call; when the content has changed, use Re-check. A completed check can become stale once the content changes afterward, so review the latest state before relying on it.

Flagging is a review aid, not a replacement for human approval or platform policy checks.

Rate limits and usage

Generative AI endpoints are rate-limited per user so one person cannot exhaust the shared provider capacity. Admins and Owners can inspect AI usage for a recent window, grouped by feature and model.

If an AI action fails:

• try again after a short wait if it looks like rate limiting or provider availability;
• shorten the input if the text is very long;
• save important copy before running a broad transformation;
• continue manually if AI is unavailable.

Practical Guidelines

Use AI for speed, not autopilot.

• Ask it for options, then choose one.
• Keep platform constraints in mind: X needs tighter writing than LinkedIn.
• Check factual claims and dates before scheduling.
• Use brand voice for company accounts, personal voice for author-led accounts.
• Do not use AI output as final approval for gated channels; the approval queue still owns that decision.