This AI Setup Runs My Social Media While I Sleep

Every piece of content you post travels further when it shows up in more places. Alston spent a few hours building an AI automation setup inside n8n that takes a single TikTok video and turns it into LinkedIn long-form posts, Instagram image posts with generated graphics, Twitter posts with automatic character-limit checks, and Reddit content. All automatically. All in his voice. All without touching a button after the initial upload.

This is not a theory post. The workflows are live, the LinkedIn posts are already pulling attention, and Alston walks through every stage of the pipeline so you can build your own version. He also shares the roadmap of platforms he is adding over the coming weeks, including Threads, Pinterest, Bluesky, and Facebook. If you have been stuck posting manually to one platform at a time, this setup changes that math entirely.

What You’ll Walk Out With

A clear picture of how one TikTok upload triggers an entire multi-platform content pipeline
The exact tools Alston uses: n8n, Airtable, AWS Transcribe, Banner Bear, and Google Drive
Why he chose AWS for transcription instead of Whisper or OpenAI, and what makes it more reliable
The modularization principle that keeps complex automations manageable and debuggable
How to prompt AI so the output sounds like you, not like a chatbot
Twitter’s character-limit loop fix and Instagram’s image-text limits explained
How to figure out which platform is the right fit for your skills and audience with finder.platformproof.com

One TikTok Video, Five Platforms: The Big Picture

The whole system starts with a single TikTok upload. When Alston posts a video to TikTok in the normal way, a third-party software (linked in the YouTube description) automatically detects the new video, repurposes it as a video file for YouTube, Instagram, LinkedIn, Twitter, and Pinterest, and at the same time downloads a copy to Google Drive. That download is what kicks off the rest of the automation chain.

At this point the video exists in two places: it is live on TikTok and its mirror platforms, and a copy is sitting in a specific Google Drive folder that n8n is watching. That watched folder is the entry point into everything else. The video content becomes text. The text becomes platform-native posts. The posts go live with no manual action required.

What makes this interesting is that the video distribution step (handled by the repurposing software) and the text-content step (handled by n8n) are completely separate pipelines. They run in parallel, not in sequence. By the time the transcription and summarization finish, the video is already live on five platforms. The text posts are just the next layer on top of what is already there.

Step 1: Google Drive Trigger and AWS Transcription

Inside n8n, the first workflow watches the Google Drive folder for a new file. The moment it detects one, it downloads the file and uploads it to Amazon Web Services for transcription. Alston is deliberate about this choice: he does not use Whisper or OpenAI for this step. His reason is that AWS Transcribe accepts both video files and audio files, whereas some alternatives accept only audio and require a separate conversion step. The character limits on other services are also more restrictive. AWS handles longer content more reliably for his use case.

After the upload to AWS, the workflow saves three pieces of data to Airtable: the video title, the filename, and the URL. At this point the transcript is not ready yet, so the Airtable record is created with those three fields and nothing else. The workflow stops and waits. When AWS finishes processing, a separate part of the automation picks up the completed transcript and writes it back to the same Airtable record, filling in the fourth field.

This two-step write pattern matters. If something goes wrong during transcription, the record still exists in Airtable with the metadata intact. Nothing is lost. The transcript just arrives later. This kind of thinking about failure modes is part of why the system is stable rather than brittle.

Step 2: Airtable as the Central Hub

Airtable is where all the data lives between steps. Every downstream workflow, whether it is building a LinkedIn post, creating an Instagram image, or writing a Twitter update, reads from Airtable rather than from the original video file. This means every workflow only needs to know how to talk to one source. If Alston ever changes how transcription works, only the transcription workflow changes. Everything else keeps reading from the same Airtable fields it always read from.

Each Airtable record also tracks the upload status per platform. A field called upload is set to no by default when the record is created. After a workflow successfully posts to a platform, it flips that field to yes. This is how every downstream workflow knows whether it has work to do on a given record. It checks: is there a record where upload equals no? If yes, process it. If no, do nothing. This prevents duplicate posts and keeps workflows from stepping on each other.

Step 3: Modularization, the Principle That Makes This Manageable

Alston uses a principle he calls modularization: each platform gets its own separate workflow rather than one giant workflow that handles everything. He has talked about this in other videos and applies it consistently here. The LinkedIn workflow does LinkedIn and nothing else. The Instagram workflow does Instagram and nothing else. They share the same Airtable data, but they are completely independent of each other in terms of how they run.

The practical benefit is that when something breaks, and something always breaks eventually, you can isolate the problem to one workflow without the others going down. If the Instagram API throws an error, LinkedIn posts keep going out. If you want to add a new platform, you just build a new workflow that reads from the same Airtable source. You do not have to touch any of the existing workflows to do it.

For anyone building their first automation, this is the most important design decision to internalize. One giant workflow that does everything is fragile and hard to debug. Small, focused workflows that share a data layer are much easier to maintain and extend. The extra time spent separating things early pays back quickly once you start adding platforms or running into errors.

Step 4: The LinkedIn Workflow in Detail

The LinkedIn workflow runs on a schedule, checking Airtable approximately every hour. When it finds a record where the upload field equals no, it pulls the transcript summary and sends it to an AI agent inside n8n with a detailed prompt. The AI agent writes a long-form LinkedIn post based on the transcript content.

The workflow then posts that text directly to LinkedIn using the LinkedIn API connection in n8n. After a successful post, it goes back to Airtable and updates the upload field from no to yes. On the next check, that record is skipped. This is how Alston avoids double-posting even if the workflow fires while the previous post is still processing.

The real work in this step is the prompt. Alston describes it as very detailed. It includes instructions to use his tone of voice, to avoid jargon, to write at approximately a sixth-grade reading level, and to avoid using asterisks or other emphasis markers that AI tends to insert. He explains the reading-level choice clearly: the average reading level in the United States is around sixth grade, and for people outside the US whose first language is not English, comprehension drops further. Writing simply does not mean dumbing down content. It means including as many people as possible. The posts that result from this workflow are thorough, readable, and recognizably his voice rather than a generic AI output.

Getting the tone right requires a one-time setup step. Alston’s approach is to feed a few of his existing text samples or audio recordings into an AI tool and ask it to identify what his writing voice sounds like. The output from that exercise goes directly into the prompt as a tone reference. After that, every generated post reflects how he actually sounds rather than how a language model sounds by default.

Step 5: Instagram Image Posts with Banner Bear

Instagram adds a layer of complexity because the platform expects visual content, not just text. Alston’s solution is a tool called Banner Bear, which is an image generation API. He created a Photoshop template with a background image and a white rounded-rectangle text box. Banner Bear takes that template and swaps in whatever text the AI agent generates, producing a finished graphic automatically.

The Instagram workflow runs similarly to the LinkedIn one: check Airtable every hour, find a record with upload equal to no, run an AI agent, get output. The difference is that the AI agent here generates two separate pieces of content at once: the text that goes on the image (currently set to under 100 characters so it fits cleanly in the template box) and the caption that goes in the Instagram post body (set to under 2,000 characters, though Instagram technically allows up to 2,200, Alston sets the limit lower because AI tends to overshoot).

The challenge in parsing those two pieces is that the AI agent returns everything as one block of text. The n8n workflow has to split that wall of text into the image-text field and the caption field before passing each one to Banner Bear and the Instagram API respectively. This is a small but important detail: the AI agent output format has to be consistent enough for the parsing logic to work reliably, which is another reason the prompt design matters so much.

Instagram itself is the most technically demanding platform in this setup. Because Instagram is owned by Meta, posting via the API requires going through the Facebook API. Setting that up is, in Alston’s words, a little bit of a hassle. It involves creating a Meta developer app, getting the right permissions, and connecting it in n8n. The good news is it is a one-time process. Once it works, it keeps working until something on Meta’s end changes, at which point a relatively quick fix gets it running again.

Step 6: Twitter and the Character-Limit Loop

Twitter has a hard character limit of 280 characters per post. AI agents tend to overshoot this limit, especially when given open-ended prompts. Alston’s solution is to build a feedback loop directly into the Twitter workflow. The AI agent generates a post with instructions to stay under 200 characters (giving buffer below the 280 limit). After the output is generated, the workflow checks the actual character count. If it is over 280, the workflow loops back and sends the request to the AI agent again. If it is under 200, the workflow proceeds and posts.

This self-correcting loop is a practical example of what makes n8n powerful for content automation. You can add conditional logic that validates the AI output before it ever touches an external API. Instead of catching errors after a bad post, you prevent the bad post from going out at all. The same pattern can be applied to any platform-specific constraint: character limits, image dimensions, caption formatting rules, and so on.

The result for Twitter is clean, short, punchy posts that read as intentional rather than truncated. They stay within the limit, they match the content of the original TikTok video, and they go up automatically without any manual review needed.

Not sure which platform is worth your time to automate first?

The free Finder tool matches you to the right income stream based on your actual skills and schedule. Try it at finder.platformproof.com.

Prompt Engineering: Getting AI to Sound Like You

The quality of every output in this pipeline depends on the quality of the prompt. Alston’s prompts share a few consistent elements that are worth pulling out explicitly because they apply to any AI content workflow, not just this one.

Tone of voice instruction: Feed the AI a description of how you sound. Do not just say “write like me.” Give it specific examples or a written profile. Alston generates this by feeding his own content into AI and asking it to characterize his voice, then copies that characterization into the prompt as a standing instruction.

Reading level: Setting a target of sixth-grade readability forces simpler vocabulary and shorter sentences. This is not about the audience being unsophisticated. It is about being accessible to as many people as possible, including non-native English speakers who are a significant portion of any online audience.

Explicit formatting rules: AI defaults to bolded phrases, asterisks for emphasis, and structured lists. For platforms where that formatting renders as literal asterisks (plain text posts on LinkedIn or Twitter), those defaults create visual noise. Telling the AI specifically not to use those markers produces cleaner output. Hard character limits, output structure requirements (image text first, caption second), and field separators all belong in the prompt as explicit rules rather than implied expectations.

Good prompt engineering is the difference between a content automation system that requires daily fixes and one that runs without supervision. The time spent refining the prompt up front saves hours every week afterward.

Honest Drawbacks

Alston does not oversell the ease of this setup. He estimates it takes around five hours the first time through, and that is for someone who is already comfortable with the concepts. For someone new to n8n or API connections, it will take longer. There is also a predictable cycle of frustration: things will break during setup, the AI agent will misbehave, API permissions will be confusing, and there will be moments of real self-doubt.

The Facebook and Meta API setup for Instagram is the single most time-consuming piece. The developer app creation, permission scoping, and token management are not intuitive, and the Meta developer documentation is not always easy to follow. This step alone can add two or three hours to an initial build.

The ongoing maintenance commitment is real too. APIs change. n8n updates sometimes affect existing nodes. The repurposing software that handles video distribution is a third-party dependency that could break or change its pricing. None of these are deal-breakers, but anyone building this should go in with realistic expectations about the occasional troubleshooting session.

That said, Alston’s framing is accurate: once you have built and debugged your first few AI agent workflows, the core skills transfer directly to every new workflow you build. The fifth workflow is dramatically faster to build than the first. The troubleshooting gets faster too because you have already seen most of the failure modes before.

Alston’s Roadmap: Platforms He Is Adding Next

At the time of the video, these platforms and content types were on Alston’s list to add to the pipeline:

Threads (text posts and image posts, similar structure to Twitter and Instagram)
Pinterest (unique pin creation, with strong relevance to the pet niche he is building out)
Twitter images (separate from text-only Twitter posts)
Blog posts (using the same transcription and summarization pipeline to generate written articles)
Facebook text posts and Facebook image posts
Bluesky images and Bluesky text posts
TikTok image posts (uploading a static image on TikTok an hour after the original video goes up, using key points from that video as the image text)
Faceless YouTube content, both long-form and short-form, using the same upload pipeline

He also mentioned building an AI agent to collect performance data automatically, so he can track impression and view growth across platforms without manually pulling numbers from each one.

The broader point here is about the reusability of the framework. Once you know how to watch a Google Drive folder, transcribe a file, summarize it, and post to one platform, adding a second platform is mostly a copy-paste operation with platform-specific adjustments. The heavy conceptual work only happens once.

Find Your X

Building a multi-platform content automation system is one way to grow an online presence without grinding manual posting. But it is only worth the setup time if you are posting content in a niche that matches your skills and has real income potential. The free Finder tool at finder.platformproof.com helps you identify which platform and income stream fits your actual situation right now, so you build the automation around the right foundation from the start.

Frequently Asked Questions

What is n8n and why does Alston use it for this?

n8n is an open-source automation platform that lets you connect apps, APIs, and AI models using a visual node-based interface. Alston uses it because it supports AI agents natively, has direct integrations with the tools in this stack (Google Drive, Airtable, LinkedIn, Twitter), and gives him full control over the logic without requiring custom code for most steps. It is available as a self-hosted option, which matters for anyone concerned about data passing through a third-party cloud service.

Why does he use AWS Transcribe instead of Whisper?

AWS Transcribe accepts both video files and audio files directly, so there is no need to strip audio before uploading. Whisper and OpenAI’s transcription endpoint have lower size and length limits, which can cause problems with longer videos. AWS handles the volume and file format flexibility that Alston’s workflow requires. If you are working with shorter clips and prefer a simpler setup, Whisper is a reasonable alternative, but it requires an extra audio-extraction step.

What is modularization and why does it matter for automation?

Modularization means splitting a large process into small, independent workflows that each do one specific thing. In this context, it means LinkedIn has its own workflow, Instagram has its own workflow, Twitter has its own, and so on. They all read from the same Airtable data source but run independently. When one breaks, the others keep working. When you want to add a new platform, you add a new workflow without touching the existing ones. For anyone building multi-step automations, this design principle dramatically reduces debugging time and maintenance load.

How do you make AI write in your specific voice?

The most reliable method is to feed a few examples of your own writing or a transcript of your speech into an AI tool and ask it to describe what your voice sounds like. The output is a short profile: your sentence length, vocabulary choices, how you use questions, whether you use humor, your pacing. You then paste that profile into your automation prompts as a standing instruction. Pair that with explicit rules about what not to do (no jargon, no asterisks, no overly formal phrasing) and the output will sound noticeably more like you than generic AI text.

What is Banner Bear and what does it actually do here?

Banner Bear is an API-based image generation tool. You create a template in Photoshop or Canva with a text placeholder, upload it to Banner Bear, and then make API calls to swap in different text values. In Alston’s workflow, the AI agent generates the text that should appear on the Instagram image, that text gets sent to Banner Bear via n8n, and Banner Bear returns a finished image with that text rendered inside the template. No manual design work after the initial template is built.

Is the Facebook API setup for Instagram really that complicated?

It is the most technically demanding single step in the whole pipeline. You need to create a Meta developer account, set up an app with the correct permissions for Instagram Graph API access, generate access tokens, and connect those tokens inside n8n. The process works, but the Meta developer portal has a lot of settings and the terminology is not always clear. Alston’s honest assessment is that it takes time and causes frustration on the first attempt. Once it is done, you typically do not need to redo it unless Meta changes something or your tokens expire.

Does this setup work if you are just starting out with no audience?

Yes, and that is actually when it is most valuable. When you are building an audience from scratch, consistency matters more than perfection. Showing up every day on multiple platforms is extremely difficult to sustain manually. An automated pipeline means you post consistently even during weeks when life is busy, without burning out. The posts being in your genuine voice also means they do not read as obviously automated, which is important when you are still building trust with a new audience.

How long does it realistically take to build this from scratch?

Alston estimates around five hours for a first build, and that is for someone who is not brand new to automation tools. If n8n and APIs are completely unfamiliar territory, budget closer to eight to ten hours spread across several sessions. The first workflow is the slowest because you are figuring out concepts, not just configuration. Once you have LinkedIn working end-to-end, adding Twitter will take a fraction of the time. Adding Instagram will be slower because of the Meta API, but the n8n workflow logic itself will be fast to build.

Sources

Alston Godbolt, “This AI Setup Runs My Social Media While I Sleep,” YouTube, youtu.be/xHrEa1qfNNc
n8n automation platform: n8n.io
Airtable database platform: airtable.com
Banner Bear image generation API: bannerbear.com
AWS Transcribe documentation: aws.amazon.com/transcribe
Meta for Developers (Instagram Graph API): developers.facebook.com