How to Build a Personal AI News Briefing with RSS and a Summarizer in 30 Minutes

How to Build a Personal AI News Briefing with RSS and a Summarizer in 30 Minutes Tutorials

By seven in the morning, the interesting things have already been published and buried. A model ships, a paper lands, a regulator opens a consultation, and by the time you sit down with coffee it is three hundred posts down the timeline. The fix is not to read faster. It is to have the reading done for you, from sources you picked, in a format you can finish in four minutes. Here is a build that takes about half an hour the first time and almost nothing to maintain after that.

What you are actually building

There are three moving parts, and it helps to see them separately before you touch any tool.

  • A collector. Something that watches a set of RSS feeds and hands you new items on a schedule.
  • A summarizer. A language model that turns a pile of headlines and excerpts into a short, readable digest.
  • A delivery channel. Usually email, because email is the one inbox you already check without being nudged.

You can wire these together with a no-code automation tool such as Zapier or Make, or with a self-hosted option like n8n if you prefer. All three work; pick the one whose interface you find least annoying, because you will be back in it later to tweak things.

Step 1: Build a short list of feeds worth reading

Start with five to eight feeds. Not thirty. A briefing built on thirty feeds is a briefing you will stop opening by Thursday.

Good candidates: the blogs of the labs and companies whose releases you care about, the announcement pages of the platforms you build on, a research aggregator, and one or two newsletters that already do curation for you. Community feeds are useful too, but treat them as a second tier — they produce volume, not signal.

Most sites hide the RSS link in the page footer or publish it at a predictable address such as /feed or /rss. If a source has no feed at all, a feed-generation service can usually turn its homepage into one.

For each feed, note two things: how often it publishes, and whether its headlines alone tell you enough. Feeds that publish twenty times a day need a tighter filter than feeds that publish twice a week.

Step 2: Filter before you summarize

Summarizing everything is how you get a briefing full of product marketing. Filter first, so the model only ever sees items that passed your bar.

Set your automation to take only items from the last 24 hours, then keep a headline if it contains at least one of your trigger words. A starter set that works well for technical and industry news:

  • release, launch, ships, version, update
  • benchmark, evaluation, results, paper, preprint
  • open source, weights, license, API
  • funding, acquisition, partnership, regulation, lawsuit

Add a deduplication step that drops anything whose link you have already seen. Without it, a story that gets updated three times will appear three times, and the quickest way to kill a daily habit is to make it repetitive.

Step 3: Write the summarizer prompt like a specification

The quality of the briefing comes almost entirely from this step. Vague prompts produce vague paragraph soup. Give the model a role, strict limits, and a rule for what to leave out.

You are editing a short daily technology briefing for one busy reader. You will receive a list of headlines with source names and short excerpts. Return only the items that describe a real event: a release, a result, a funding round, a policy change, or a notable technical claim. Drop opinion pieces, listicles, and anything that is just a product promotion. For each kept item, write one sentence of no more than 25 words describing what happened, using the names, version numbers, and figures exactly as they appear in the input. Do not add context that is not in the input. Do not use adjectives like exciting, revolutionary, or massive. Rank items by how much they change what a practitioner can do this week. Return the output as a numbered list, with the source name in brackets after each sentence.

Two practical notes. First, send all items in a single request rather than one request per item — it is faster, cheaper, and produces a better ranking. Second, ask for the raw headline back alongside the summary. When a sentence looks wrong, you want to check it against the original in the same email instead of hunting through tabs.

Step 4: Format the email and schedule it

Keep the format boring. A structure that works:

  1. Top five stories, one sentence each, with the source and the link.
  2. The skim list, everything else that passed the filter, as bare headlines with links.
  3. A one-line note at the top saying how many items were scanned and how many made the cut. It tells you at a glance whether the filter is too loose.

Then set the trigger. Early morning is the obvious choice, but pick a time you will actually have thirty seconds to read. Send it to a dedicated label or folder so it never competes with the rest of your inbox.

Step 5: Run it once by hand before you trust it

Trigger the automation manually and read the output critically. Look for four specific failures: summaries that contain numbers or names not present in the source, duplicate entries, broken links, and a top five that is mostly press releases.

Fix the filter before you fix the prompt. Most bad briefings are a sourcing problem wearing a summarization costume. Also add a fallback: if the model request fails or returns nothing usable, have the automation send the plain list of headlines anyway. A slightly dull briefing beats a silent one.

Tuning it after the first week

Give it seven days, then make three small edits. Add the one feed you kept wishing was in there. Drop the one that supplied nothing but noise. And raise the bar in the prompt — change "a real event" to your own definition of one.

If daily starts to feel like too much, switch to Tuesdays and Fridays, or move it to a single Sunday digest. Frequency you keep beats frequency you admire.

Keep it small and keep it running

The whole point is compression. A briefing you can read in four minutes gets read; one that takes twenty gets archived. So resist the urge to add feeds when the email feels thin — add better ones instead, and delete the ones that have gone quiet.

Set a reminder to audit your feeds once a quarter. Blogs die, feeds move, and a collector quietly pulling from dead URLs will happily send you an empty email for months without complaining. Ten minutes of maintenance every three months is the entire cost of never being the last person to hear about something again.

Photo: Firmbee / Pixabay