build an ai content pipeline that runs while you sleep

the problem with manual content workflows

you spot a trending topic on monday. you brainstorm on tuesday. you write on wednesday. you design on thursday. you post on friday. by friday the trend is dead and the algorithm has moved on.

this is how most creators operate. and it is why most creators plateau. the bottleneck is never ideas - it is the time between spotting an idea and publishing something about it. shrink that gap to near-zero and you win.

the fix is not "work harder." the fix is a pipeline that detects trends, scores them, drafts content, and queues it for posting - all while you sleep. you wake up, review a notion inbox, approve or tweak, and move on with your day.

here is the exact system we run. every tool is real, every connection is no-code, and the whole thing can be built in a weekend.

the full stack - what you need

before we get into phases, here is every tool in the pipeline and what it does:

• n8n - the orchestration layer. every workflow lives here. self-hosted (free, docker) or n8n cloud ($20/mo). this is the brain that connects everything else
• apify - web scraping platform with pre-built "actors" for tiktok, instagram, x/twitter, reddit, and youtube. pulls raw trend data on a schedule
• claude api - the ai layer. scores trends, generates hooks, writes drafts, and reformats content per platform. uses anthropic's api directly via n8n's http request node
• notion - your content hub. every trend, draft, and approved post lives in a single database with status columns, platform tags, and scheduling dates
• buffer - multi-platform scheduler. supports instagram, tiktok, x, linkedin, threads, bluesky, pinterest, youtube, and facebook. has an api that n8n can push to directly
• make.com (optional) - alternative to n8n if you prefer a visual builder. same logic, different interface. some creators use make for the scheduling leg and n8n for everything else

total monthly cost: $0-50 depending on whether you self-host n8n and which apify plan you use. claude api costs scale with usage but a typical creator pipeline runs under $10/month in tokens.

phase 1 - trend detection (apify + n8n)

this is the ears of your pipeline. every 6 hours, apify scrapers pull trending content from the platforms you care about.

step 1: pick your apify actors. go to apify.com/store and grab the actors for your platforms. the key ones: - tiktok scraper - pulls trending hashtags, sounds, and top-performing videos by niche keyword - instagram reel scraper - extracts public reel data including views, likes, shares, and captions - x/twitter trends scraper - grabs real-time trending topics by country or city - reddit scraper - pulls top posts from subreddits relevant to your niche

step 2: set up scheduled runs in n8n. create a workflow in n8n with a schedule trigger node set to fire every 6 hours. connect it to an apify node that kicks off each actor. when the actor finishes, its dataset flows into the next node automatically.

step 3: filter for relevance. add an n8n "if" node after each scraper. filter by keywords that match your niche. if you are a fitness creator, filter for terms like "workout," "protein," "recovery," "gym." everything that does not match gets dropped. everything that does moves to phase 2.

what you end up with: a raw feed of 20-50 trending topics every 6 hours, pre-filtered to your niche, sitting in your n8n workflow ready for scoring.

phase 2 - ai scoring (claude api)

raw trends are noise. you need signal. this phase uses claude to score every trend on three dimensions so only the best ones become content.

step 1: build your scoring prompt. in n8n, add an http request node pointed at the anthropic api (api.anthropic.com/v1/messages). your system prompt should define three scoring criteria: - virality potential (1-10): is this topic generating high engagement? is it rising or peaking? - brand fit (1-10): does this align with your content pillars and audience? would your followers care? - freshness (1-10): is this new enough that posting about it tomorrow still matters?

step 2: pass the trend data. for each filtered trend, send the title, description, engagement metrics, and source platform to claude. ask it to return a json object with scores and a one-sentence summary of why this trend is worth covering.

step 3: set your threshold. add another "if" node. only trends scoring 22+ out of 30 move forward. this is your quality gate. in practice, out of 50 raw trends you will get 3-8 that clear the bar. that is exactly the right volume for a solo creator.

pro tip: include 2-3 examples of past trends you covered successfully in your system prompt. this teaches claude your taste and makes the scoring more accurate over time.

phase 3 - content generation (claude api + n8n)

winning trends now get turned into actual content. this phase generates platform-specific drafts automatically.

step 1: define your output formats. for each winning trend, claude generates: - 1 short-form hook (under 10 words, pattern-interrupt style for reels/tiktok) - 1 tweet/thread (3-5 posts, first post is a hook, last post is a cta) - 1 carousel script (8-10 slides, slide 1 = hook, slide 2-9 = value, slide 10 = cta) - 1 long caption (instagram/linkedin, 150-200 words, storytelling format) - 1 newsletter angle (2-3 sentence pitch for your next beehiiv or substack issue)

step 2: build the generation prompt. this is separate from the scoring prompt. it receives the winning trend summary plus your brand voice guidelines. include: - your tone (casual, authoritative, funny, raw - whatever fits) - words/phrases you always use - words/phrases you never use - your typical cta style - example posts that performed well (paste 2-3 real ones)

step 3: format the output. use n8n's "set" node to structure claude's response into clean fields: hook, thread, carousel, caption, newsletter_angle, source_trend, score, generated_date.

the key insight: you are not asking claude to write finished posts. you are asking it to write 80%-done drafts. your voice and judgment close the last 20%. this is what separates slop from content that actually sounds like you.

phase 4 - content hub (notion)

every generated draft lands in a notion database automatically. this is your review inbox.

step 1: create your notion database. set up a database with these properties: - title (text) - the hook or headline - status (select) - "draft," "in review," "approved," "scheduled," "published" - platform (multi-select) - instagram, tiktok, x, linkedin, threads, newsletter - content type (select) - hook, thread, carousel, caption, newsletter - source trend (text) - what triggered this piece - trend score (number) - the claude score from phase 2 - body (rich text) - the full draft content - scheduled date (date) - when this should go live - generated date (date) - when the pipeline created it

step 2: connect n8n to notion. use n8n's native notion node. map each field from your generation output to the corresponding notion property. every time the pipeline runs, new rows appear in your database.

step 3: build your review workflow. every morning, open your notion database filtered to status = "draft." read each piece. edit what needs editing. change status to "approved" and set a scheduled date. this should take 15-20 minutes for 3-8 pieces.

optional: add a kanban view. create a board view grouped by status. drag cards from "draft" to "approved" to "scheduled." visual and satisfying.

phase 5 - scheduling (buffer)

approved content gets pushed to buffer automatically. no copy-pasting. no manual scheduling.

step 1: connect notion to n8n (again). create a second n8n workflow triggered by notion database changes. when a row's status changes to "approved," this workflow fires.

step 2: format for each platform. use n8n's "switch" node to route content by platform. each branch reformats the content for that platform's requirements - character limits, hashtag placement, mention formatting.

step 3: push to buffer via api. use n8n's http request node to hit buffer's api. create a post for each platform with the content body and scheduled time. buffer handles the actual publishing at the right time.

alternative: use make.com for this leg. if you prefer make.com's visual builder, you can use it just for the notion-to-buffer connection. make has native modules for both notion and buffer, which means less api configuration.

posting time optimization: buffer analyzes your audience engagement patterns and suggests optimal posting times per platform. use these suggestions when setting your scheduled dates in notion.

phase 6 - the feedback loop

this is what separates a pipeline from a system. after content goes live, you feed performance data back into the pipeline to make it smarter.

step 1: track performance. once a week, check which posts performed best. note the trend scores, content types, and platforms that drove the most engagement.

step 2: update your scoring prompt. add recent winners to your claude scoring prompt as examples. "trends like [this] scored well. trends like [that] underperformed." this tunes the ai to your specific audience over time.

step 3: refine your filters. if certain keywords consistently produce low-scoring trends, add them to your n8n filter's exclusion list. if new keywords emerge, add them to your inclusion list.

after 4 weeks: your pipeline gets noticeably better. trend detection is tighter, drafts are closer to your voice, and your approval rate goes up. you spend less time editing and more time doing the work that actually requires a human - being on camera, talking to your audience, building relationships.

the daily workflow (after setup)

here is what your day looks like once the pipeline is running:

• morning (15-20 min): open notion. review 3-8 drafts. edit, approve, set dates. done.
• rest of the day: create original content, engage with your audience, pitch brands, live your life.
• the pipeline runs 4x/day in the background. trends get scraped, scored, and drafted without you touching anything.

before the pipeline: 3-4 hours/day finding topics, writing, formatting, scheduling. after the pipeline: 15-20 minutes/day reviewing and approving.

common mistakes to avoid

• skipping the human review step. never auto-publish. always review. the pipeline generates drafts, not finished content. your taste is the product.
• setting the score threshold too low. if you approve everything, you flood your feed with mid content. keep the bar high. 3-5 great posts per day beats 15 forgettable ones.
• ignoring the feedback loop. the pipeline is only as smart as the examples you feed it. update your prompts monthly at minimum.
• overcomplicating the setup. start with one platform and one content type. get that working. then expand. a pipeline that runs beats a pipeline that is "almost ready."

the complete stack (with costs)

• n8n - workflow orchestration ($0 self-hosted / $20/mo cloud)
• apify - trend scraping ($0-49/mo depending on volume)
• claude api - scoring + generation (~$5-15/mo for a typical creator pipeline)
• notion - content hub (free tier works fine)
• buffer - scheduling ($0-6/mo per channel)
• make.com (optional) - visual automation alternative ($0-9/mo)

total realistic cost for a solo creator: $15-50/month. compare that to 80+ hours/month saved.

who this is for

solo creators and small teams (1-5 people) who post on 2+ platforms and want to stop reinventing the wheel every morning. you need basic comfort with connecting tools via apis - no coding required, but you should know what an api key is and how to paste one into a settings field.

if you can follow a recipe, you can build this pipeline. the initial setup takes one focused weekend. the compound returns last as long as you keep creating.