claude code for marketing work you can check

start with a campaign packet

give the bot the approved product facts, audience, offer, deadline and writing rules as files. ask for a list of missing facts before asking for copy. if the price, launch date or terms are absent, the right next step is a question, not an invented sentence.

in dusk bot, use Attach files or drop the material into the chat. what the bot makes appears in the chat and lives in its folder, or a project folder you provide. keep the source packet separate from the working draft so you can compare each claim against the material that supports it.²

three bounded jobs to try

• brief review: ask it to identify contradictions between the campaign brief and the approved product facts. require the relevant passage for each concern.
• copy review: ask it to mark claims in a supplied draft that are missing a source, and to separate factual concerns from style suggestions.
• content inventory: provide a file of existing titles and ask it to group overlapping topics before proposing another article.

check the first answer closely. pick one job per thread. for wider discovery, the market research workflow is a different task from reviewing supplied material.

turn a repeated review into a routine

after a useful manual run, make the repeat job precise. for a weekly content review, name the source file, the day, the destination and when to stay quiet. tell the bot to report missing material rather than fill gaps, and to ask before changing anything.

dusk routines are made by describing the repeating job. a card shows the schedule and next runs with Edit, Pause and Undo. Automations has the calendar and run history. the newsletter workflow is a related place to separate drafting from the final send.³

claude code has its own scheduler too

if you prefer working directly with claude code, its routines are now called cloud scheduled tasks. as of 12 november 2026, they save a prompt, repositories and connectors, and can start on a schedule, api call or github event. they run on anthropic-managed infrastructure and count against plan usage.⁴

that may fit a marketing site managed as a code project. it is not the same setup as a dusk routine created in a bot chat. the scheduled-task explainer helps distinguish cloud tasks, local desktop tasks and /loop before you choose one for recurring work.

give publishing its own decision

for reviewing source files, the bot needs no social account. if you later connect an app, choose access for that bot only. in dusk bot, reads and drafts can happen freely. sending messages, posts, emails or payments asks first unless you choose a daily amount the bot may send on its own. a bot on Full access with no sending rule of its own may send up to 25 a day without asking, so set Ask every time for it.⁵²

the sending setting doesn't cover changes like editing a page or changing app records. to keep them supervised, give the bot read-only access or write "ask me before you change anything" into the job. the social posting workflow covers a separate publishing task. approval of one draft is not a reason to remove review from the whole campaign.

what remains an editor’s job

check names, prices, links, dates and promises against your source files. a bot cannot know an unpublished launch decision you never gave it. it also cannot establish customer results, testimonials or search demand from a blank brief.

a draft that reads well can still say the wrong thing. keep unsupported claims out, preserve uncertainty and review the exact version that will be published. when a task needs fresh research rather than supplied material, define that research separately.