TL;DR: A daily AI news digest in your inbox costs $0 to run with n8n and RSS, takes 25 minutes to build, and replaces six paid subscriptions. Below is the exact workflow I run every morning at 7:03 AM.
You don’t need another AI newsletter. You need a daily AI news digest you actually control — one that pulls from the eight sources you trust, drops the rest, and lands in your inbox before you finish your first coffee. I built mine in n8n in under 30 minutes. It replaced six subscriptions and saves about 40 minutes a day on reading and tab-juggling. [test-claim]
What you’ll get
- An n8n workflow that pulls RSS feeds, dedupes, and emails a clean digest
- A scoring prompt that drops 60% of the noise before it hits your inbox
- An HTML email template that reads well on mobile and desktop
- A copy-paste prompt for Claude or GPT-4 to summarize each item in 25 words
Why a Daily AI News Digest Beats Subscribing to Eight Newsletters
Newsletters are someone else’s priorities. Every list on Beehiiv, every ConvertKit blast, every founder Substack — they show up at different times, with different layouts, all competing for attention. A daily AI news digest you own collapses that into one email at one time, with one format you can scan in 90 seconds.
The bigger reason is control. You pick the sources. You set the filter. If a publication keeps shipping AI-generated listicles, you drop the feed. If a small Substack consistently breaks news, you weight it up. No algorithm sits between you and the signal.
There is a cost: you have to build it once. After that, the daily AI news digest runs itself. n8n charges nothing if you self-host. The LLM filter costs about $0.18 per month at my volume. [test-claim] Cheaper than any paid newsletter.
The Make.com vs n8n question comes up a lot. I run both — Make.com for client-facing automations that need a clean visual log, n8n for personal infrastructure I want to own. For a daily AI news digest you’ll touch every few months, n8n wins on cost and on the fact that nothing breaks behind a paywall. {{internal:n8n-vs-make}}
What You Need Before Building Your Daily AI News Digest
Four things. None of them complicated.
1. An n8n instance. Self-host on a $5 Hetzner VPS or use n8n Cloud’s starter tier [verify pricing] [source-needed]. Self-hosting is the right move if you already run Docker. Cloud is fine if you don’t.
2. An LLM API key. Anthropic’s Claude Haiku or OpenAI’s GPT-4o mini both work. Haiku reads cheaper at the volume your daily AI news digest will generate, in my testing [test-claim]. {{internal:claude-vs-gpt-summarization}}
3. An email send method. Gmail SMTP works for personal use. For deliverability on a custom domain, Resend or Postmark are fine [source-needed]. I use Gmail SMTP because the digest only goes to me.
4. Eight to twelve RSS feeds. Most people get this part wrong. More feeds means more noise. Pick the smallest set that still covers your beats. My current list: Anthropic blog, OpenAI blog, Google DeepMind, Hacker News front page filtered for “AI,” Stratechery, The Information’s free RSS, two founder Substacks, and one academic feed. {{internal:rss-feed-list-for-founders}}
The n8n Workflow, Node by Node
Here is the spine of the workflow. Build it in this order and it works on the first run.
Schedule Trigger. Set it to fire at 7:00 AM in your local timezone. Cron: 0 7 * * *. That gives the LLM step a few minutes to finish before the email lands.
RSS Read nodes — one per feed. Yes, this part is repetitive. n8n’s RSS Read node takes a URL and pulls the latest items. Set limit to 10 per feed so you don’t blow up the prompt budget. If a feed is dead, the node fails silently and the rest keep running, as long as you set the workflow to “continue on fail.”
Merge node. Pull all RSS outputs into one stream. Use the “Append” mode.
Code node — dedupe and filter by date. Drop items older than 24 hours. Dedupe by title hash. Twelve lines of JavaScript:
const seen = new Set();
const now = Date.now();
return items.filter(i => {
const t = i.json.title?.toLowerCase().trim();
const d = new Date(i.json.pubDate).getTime();
if (!t || seen.has(t)) return false;
if (now - d > 86400000) return false;
seen.add(t);
return true;
});LLM node — score and summarize. Send each item to Claude or GPT-4o mini. Prompt below.
Filter node. Drop anything with a relevance score under 6.
Sort node. By score descending so the highest-signal items sit at the top of your daily AI news digest.
Set node — build HTML. Wrap the surviving items in a clean email template.
Email Send node. Done.
Cleaning Up Your Daily AI News Digest With an LLM Filter
The filter is what turns a daily AI news digest from background noise into something you read. Without it, you get 80 items a day and 70 are junk. With it, you get 8–12, all relevant.
Here is the prompt I use, tuned over about two weeks of edits:
You score AI news for a solo founder building a SaaS.
Score 1-10 on relevance to: practical AI tooling,
model releases, pricing changes, agent frameworks,
indie SaaS case studies. Skip: enterprise deals,
geopolitics, opinion pieces, listicles.
Return JSON: {"score": N, "summary": "max 25 words"}
Title: {{ $json.title }}
Description: {{ $json.contentSnippet }}Two notes from the field. First, the scoring is more consistent if you give the model a tight list of “skip” categories. Vague rules (“avoid filler”) produce uneven cuts. Second, ask for JSON output explicitly and parse it in the next node. Free-form text breaks the workflow more often than you’d think. [test-claim]
Score threshold is a knob. Six is my default. If your inbox feels thin, drop to five. If it feels noisy, push to seven. I retune mine once a month based on what I actually clicked.
Sending an Email You’ll Actually Read
An ugly digest is one you’ll archive without scanning. Three rules:
One column. Mobile-first. Inline CSS only. No images in the body — they slow render and trip spam filters.
Title + summary + source link. Nothing else per item. No author photos, no share buttons, no “read time.”
Group by source type. “Model releases” first, “Tooling” second, “Long reads” last. The brain skims better when it knows which bucket it’s reading.
Keep the whole email under 1,200 words. That’s the line where a digest stops being a digest and starts being homework.
What 30 Days of My Daily AI News Digest Looked Like
Concrete numbers from my own run in June 2026 [test-claim]:
- Newsletters unsubscribed: 6
- Average daily reading time: 4 minutes, down from about 35
- LLM cost: $0.18 across 30 runs (Claude Haiku)
- Workflow breakage: twice — both times a feed URL changed
- Items per digest: 9.2 average, range 5–14
The two breakages were dead RSS URLs. Adding a Slack notification node on workflow failure fixed that. I now know within five minutes if the morning run didn’t fire.
One surprise: I started saving items I wanted to revisit by forwarding them to a Notion inbox. Notion‘s email-in feature handles this without an extra integration. Two months in, that inbox has become my actual reading queue for deep work, not the digest itself.
Bottom Line
Build your daily AI news digest in n8n. Self-host if you can, use n8n Cloud if you can’t. Use Claude Haiku for the filter. Send via Gmail SMTP. Keep the feed list under twelve. Run it at 7 AM your time.
Skip the “AI news platform” SaaS products. They’re newsletters with a UI on top. You’re paying $15/month for someone else’s filter. Yours will be sharper because you set it.
FAQ
How much does it cost to run a daily AI news digest in n8n?
About $5/month if you self-host on a small VPS, plus around $0.20/month for LLM calls at Claude Haiku rates. n8n Cloud’s lowest paid tier runs higher [verify pricing] but removes the server work.
Can I use Make.com instead of n8n?
Yes — the same workflow translates. Make.com is easier for visual debugging. n8n is cheaper at this volume and self-hostable. For a personal daily digest, n8n wins. For client work, Make.com is fine.
What if a source doesn’t have RSS?
Use a service like RSS.app or Feedity to generate one from a page. Their free tiers are usually enough for a personal digest [source-needed].
Why Claude Haiku and not GPT-4o mini?
Both work. In my testing Haiku followed the JSON output format more consistently across 30 days, with fewer parse errors. GPT-4o mini was slightly faster per call. Pick whichever API key you already have.
How many RSS feeds is too many for a daily AI news digest?
Twelve is the soft ceiling. Past that, the LLM cost climbs and the filter starts cutting items you wanted to see. If you’re tempted to add a 13th, drop one first.
Can I get the digest as a Slack DM instead of email?
Yes. Swap the Email Send node for a Slack node and post to yourself. Same workflow, different channel. I tested both. Email won because I read it during coffee, not while context-switching.
What to do in the next 10 minutes
- Sign up for n8n Cloud’s free trial, or spin up the Docker image on a server you already have.
- Open a notes file and list the 8–12 RSS feeds you’d actually read if the daily AI news digest worked. Skip anything you currently skim past.
- Grab a Claude or OpenAI API key. You’ll spend less than a dollar testing this end to end.
Build the spine of the workflow first — Schedule + one RSS + Email Send. Get it landing in your inbox. Then add the LLM filter, then the rest of the feeds. If you try to build everything at once, the first failure stops you. Build it ugly, then make it sharp.