A researched LinkedIn draft each day, waiting for a person to approve it
Each day a Make scenario takes the next queued topic from an Airtable content calendar, has Claude plan the angle, researches it twice with Perplexity, has Claude write the post in the founder's voice, generates a matching image and saves the draft to Airtable for review. Nothing is published from this scenario.
Where this applies
Founders, consultants and small firms that publish thought-leadership posts and want a researched first draft, not an automatic post.
- AI models, one to research and one to write
- 2
- modules, topic to draft
- 11
- drafts, never auto-posted
- Daily
The problem
Posting regularly on LinkedIn is a commitment most founders make and few keep, because every post is research, writing and an image. Handing the whole job to one model produces posts that read like everyone else's and sometimes state things that are not true.
The useful split gives the machine the legwork (finding the facts, a first draft, an image) and keeps the decision to publish with the person whose name is on the post.
What we built
The calendar is the queue
Topics wait in an Airtable table. The daily run searches it and a filter lets through only a queued topic, so a day with nothing queued costs one search and stops.
Plan, research, then write
Claude first turns the topic into an angle and an outline. Perplexity then runs twice, once for research and once for facts and sources. Only then does Claude write the final post in the founder's voice from that material. The research model and the writing model are kept apart, so the facts are gathered before the voice is applied.
An image by polling
The image comes from an image-generation API that works asynchronously: the scenario posts the request, sleeps, polls for the result and downloads the file. A skip handler on the download means a failed image does not lose the post.
Back to the calendar, not to LinkedIn
The post text and the image are written onto the topic's record in Airtable for review. Publishing is a separate stage, so nothing reaches LinkedIn without a person choosing it.
The outcome
Built in November 2024 and switched on, running daily. It used 60 credits in the current billing period, about two a day, with no errors seen; recent runs have mostly found no topic queued.
The companion stage that publishes approved posts is switched off, so drafting is the part that runs.
Where it stops
The scenario drafts and does not publish, and the publishing stage is switched off. Nothing checks the post against the sources Perplexity returned; that is the reviewer's job. A failed image download is skipped, so a draft can arrive without its image. Three earlier versions are archived and not described.
Built with
- Make.com
- Airtable
- Anthropic Claude
- Perplexity AI
- Image-generation API
Patterns used
- Calendar as queue
- Research model, writing model
- Async job by polling
- Draft for approval
Page copy drafted with AI assistance. Skofi is accountable for it. Editorial policy
Have something shaped like this?
If any part of the above resembles a problem you are living with, a short call is the fastest way to find out whether it is the same problem underneath.
What happens on the call
- 15 minutes, no deck
- We map the process on the call
- A written scope, or an honest no