If you run a blog, a newsletter, or a content-heavy marketing operation, you already know the grind. You check 15 RSS feeds every morning, read the ones that look relevant, summarize the key points, drop them into a Notion database, and then write an email digest for your audience. That routine eats 60 to 90 minutes a day. Multiply it by five workdays and you are looking at a full work day lost to copy-paste labor every week.
There is a better way. In 2026, the tools to automate this pipeline are cheap, reliable, and do not require a developer on staff. Make.com (formerly Integromat), ChatGPT or Claude, Notion, and a transactional email service like Resend or Mailchimp can be wired together into a content pipeline that runs while you sleep. You wake up to a Notion database populated with AI-summarized articles and a draft email digest ready to review and send.
This guide walks through the full setup. It covers the architecture, the exact Make.com scenario configuration, the prompt engineering that produces useful summaries instead of generic fluff, the Notion database schema, and the email delivery step. It also covers the failure modes, the cost, and the cases where this approach does not make sense.
What the pipeline does, end to end
The pipeline has five stages. Each stage is a module in a Make.com scenario.
Stage 1: RSS trigger. Make.com watches one or more RSS feeds. When a new item appears, the scenario fires. You can watch industry blogs, competitor sites, news outlets, or any source that publishes an RSS feed. Most WordPress sites, Substack newsletters, and major news outlets still publish RSS in 2026.
Stage 2: Content extraction. The RSS item contains a title, a link, and usually a short excerpt. Make.com fetches the full page content from the link using its HTTP module and extracts the article text. You can use the built-in HTML-to-text parser or a service like Diffbot for cleaner extraction.
Stage 3: AI summarization. The extracted text goes to ChatGPT (via OpenAI's API) or Claude (via Anthropic's API). The prompt asks for a structured summary: a one-sentence hook, three to five key takeaways, and a relevance score from 1 to 10 that tells you how interesting this article is for your audience. The relevance score is the part that makes the pipeline actually useful. Without it, you get a flat list of summaries with no signal about what deserves attention.
Stage 4: Notion database entry. The summary, the relevance score, the source URL, the original title, and the publication date get written to a Notion database via Make.com's Notion module. You can review the entries each morning, edit anything that needs a human touch, and mark the ones that should go into the digest.
Stage 5: Email digest. On a schedule (daily or weekly), a second Make.com scenario reads the Notion database for entries above a relevance threshold, formats them into an HTML email, and sends the digest through Resend, Mailchimp, or any SMTP service.
The whole chain runs automatically. Your job shifts from reading and summarizing 15 articles a day to reviewing a pre-filtered, pre-summarized list and hitting send.
Setting up Make.com
Make.com's free tier gives you 1,000 operations per month. Each module execution in a scenario counts as one operation. For a pipeline watching 10 RSS feeds with an average of 3 new articles per feed per day, you are looking at roughly 30 scenario runs a day, each consuming 4 to 5 operations (RSS trigger, HTTP fetch, AI call, Notion write, logging). That is 150 operations a day, or about 4,500 per month. You will need the Core plan at $10.59/month for 10,000 operations, which covers a moderate content operation with room to spare.
Create a new scenario in Make.com and add the following modules in order:
- RSS > Watch RSS feed items. Point it at your first feed URL. Set the polling interval to 15 minutes for near-real-time tracking, or once an hour if you do not need speed.
- HTTP > Make a request. GET the article URL from the RSS item. Set the response type to HTML.
- Tools > Set variable. Extract the article body text from the HTML response. A simple regex or the built-in HTML parser works for most sites.
- OpenAI > Create a chat completion. Send the article text to GPT-4o-mini (cheap, fast, good enough for summaries) with your structured prompt.
- Notion > Create a database item. Map the AI response fields to your Notion database columns.
- Tools > Set variable (optional). Log the run for debugging.
If you want to use Claude instead of ChatGPT, swap the OpenAI module for an HTTP module calling Anthropic's API directly. Make.com does not have a native Anthonic module as of June 2026, but the HTTP module with a POST to https://api.anthropic.com/v1/messages works fine. You will need to set the x-api-key and anthropic-version headers manually.
The prompt that produces useful summaries
The difference between a pipeline that saves you time and one that creates more work is the prompt. A bad prompt produces generic summaries that read like every other AI-generated blurb. A good prompt produces structured, scannable summaries with a clear signal about relevance.
Here is the prompt that works. Copy it into the OpenAI module's system message:
You are a content analyst for a [your industry] audience.
Given an article, produce:
1. HOOK: One sentence (max 20 words) that captures the core news or insight.
2. TAKEAWAYS: Exactly 3 bullet points, each one sentence, focusing on practical implications.
3. RELEVANCE: A score from 1-10 where 10 = directly actionable for a small business owner
in [your industry], and 1 = tangential or purely academic.
4. CATEGORY: One of [your content categories].
Be specific. Cite numbers, names, and dates when available.
Do not use phrases like "in today's fast-paced world" or "it's important to note."
Replace [your industry] and [your content categories] with your specifics. The relevance score is the key feature. It lets you filter the Notion view to show only 7+ rated articles, which turns a firehose into a curated feed.
For cost, GPT-4o-mini processes roughly 1,000 articles per dollar at typical article lengths (2,000 to 4,000 tokens input, 200 to 400 tokens output). At 30 articles a day, your monthly OpenAI spend is about $1. Claude Haiku is similarly priced. This is not the expensive part of the pipeline.
Building the Notion database
Create a Notion database with these columns:
| Column | Type | Purpose |
|---|---|---|
| Title | Title | Original article title |
| URL | URL | Link to the source |
| Source | Select | Which RSS feed it came from |
| Hook | Text | One-sentence AI summary |
| Takeaways | Text | Three bullet points |
| Relevance | Number | 1-10 score from the AI |
| Category | Select | Content category tag |
| Published | Date | Original publication date |
| Added | Date | When the pipeline added it |
| Status | Select | Pending / Approved / Published / Discarded |
The Status column drives the workflow. Each morning, you open the Notion database, filter to Status = Pending and Relevance >= 7, review the summaries, and change the status of the ones you want in the digest to Approved. The email scenario reads Approved items, sends the digest, and marks them as Published.
A Notion database with this schema takes 10 minutes to set up. Make.com's Notion module maps directly to these columns. The mapping is straightforward: each field from the OpenAI response goes to one Notion column.
The email digest scenario
The second Make.com scenario runs on a schedule. Set it to fire at 7am or 8am on the days you want to send the digest. The scenario:
- Notion > Search database items. Filter: Status = Approved, Added = today (or this week, for a weekly digest).
- Array aggregator. Collect all matching items into a single bundle.
- Tools > Compose a string. Build the HTML email body from the aggregated items. Each item becomes a section with the hook as a bold header, the takeaways as bullet points, and a "Read more" link.
- Email > Send an email (or HTTP > Make a request to Resend API). Send the formatted email to your list.
Resend's free tier gives you 3,000 emails per month, which is more than enough for a daily digest to a small list. If you already use Mailchimp, use its Make.com module instead.
The email template does not need to be fancy. A clean, text-heavy format with clear section breaks outperforms a designed newsletter template for this kind of curated digest. People are reading it for the content, not the layout.
Cost breakdown
Here is what this pipeline costs per month at moderate volume (30 articles/day, daily digest to 200 subscribers):
| Component | Plan | Monthly cost |
|---|---|---|
| Make.com Core | 10,000 operations | $10.59 |
| OpenAI (GPT-4o-mini) | Pay-as-you-go | $1 to $2 |
| Notion | Free plan (personal) | $0 |
| Resend | Free tier (3,000 emails) | $0 |
| Total | $12 to $13 |
At higher volume (100 articles/day, 2,000 subscribers), Make.com scales to the Pro plan ($18.82/month for 10,000 operations) or the Teams plan ($34.12/month), OpenAI costs rise to $5 to $8, and Resend moves to the $20/month plan. Total: roughly $45 to $65 per month.
Compare that to a human content assistant spending 5 hours a week on the same task at $25/hour. That is $500 per month. The automation pipeline costs 3% of that and runs seven days a week without calling in sick.
<figure> <img src="/blog/img/ai-workflows-make-com-chatgpt-claude-gemini-content-pipeline-3.webp" alt="Vintage tin-toy robot at a whiteboard drawing a workflow diagram while smaller robots watch from a conference table" width="2000" height="1125" loading="lazy"> </figure>When this pipeline does not make sense
The automation works well for curation and summarization. It does not work well for original content creation, nuanced editorial judgment, or audiences that expect a personal voice.
Do not use this pipeline to auto-publish. The AI summaries are a starting point, not a finished product. Every digest should go through a human review before it hits inboxes. The pipeline saves you the reading and summarizing time. It does not replace the editorial judgment about what matters and why.
RSS feeds are shrinking. More publishers are moving to newsletter-only distribution, paywalled content, or platform-native publishing (LinkedIn articles, Substack notes). If your key sources do not publish RSS, the pipeline has nothing to watch. You can work around this with Make.com's web scraping modules, but scraping is fragile and raises legal and ethical questions that RSS does not.
AI summaries hallucinate. GPT-4o-mini and Claude Haiku are good at summarization, but they occasionally invent details, misattribute quotes, or produce takeaways that the original article does not support. The relevance score can also be miscalibrated, especially in the first two weeks before the model sees enough examples. Review the summaries. Trust but verify.
API costs are real. The per-article cost is low, but it scales linearly. If you are watching 50 feeds with 10 articles each per day, you are processing 500 articles a day. That is 15,000 articles a month, which pushes Make.com into a higher tier and OpenAI costs into the $15 to $20 range. Still cheap compared to a human, but no longer pocket change.
Extending the pipeline
Once the basic pipeline runs, there are three extensions worth considering.
Slack or Teams notifications for high-relevance articles. Add a Make.com module after the Notion write that sends a Slack message when the relevance score is 9 or 10. This gives you a real-time alert for must-read content without waiting for the daily digest.
Multi-model comparison. Run the same article through GPT-4o-mini and Claude Haiku, compare the summaries, and keep the better one. This costs twice as much per article but produces higher-quality output. Make.com's router module makes this straightforward.
Automatic social media drafts. After the digest is sent, take the top three articles by relevance score and ask ChatGPT to write a LinkedIn post or X thread for each one. Save the drafts to a separate Notion database for review. Do not auto-post them. AI-written social content needs a human pass to avoid the uncanny tone that audiences recognize and ignore.
The bottom line
A content pipeline built on Make.com, ChatGPT, and Notion takes about 2 hours to set up, costs $12 to $15 per month at moderate volume, and saves 6 to 10 hours of manual reading and summarizing per week. The tools are mature, the APIs are stable, and the pricing is predictable. For any solo operator, small marketing team, or agency that publishes curated content regularly, this is one of the highest-ROI automations you can build in a single afternoon.
The pipeline does not replace editorial judgment. It replaces the grunt work that eats your morning. You still decide what matters. You still write the intro and the commentary. You still hit send. The AI just makes sure you are starting from a filtered, summarized, structured list instead of a wall of open browser tabs.