n8n shows a morning AI digest workflow: scoring an article costs $0.00004

The n8n team has published a recipe in its blog for an automated morning AI digest. Every weekday at 8:00, the workflow searches for the day's news via You.com and scores each article for "materiality" at about $0.00004. It then posts one digest per topic to Slack, with links to the sources. If a topic has no real news, no post appears.

How the pipeline works

The announcement implies three steps: search, scoring, publishing. Search runs through You.com. Scoring judges whether an article matters on substance rather than merely mentioning the topic. Publishing produces one digest per topic, citing the underlying articles. The announcement does not explain the role of Jev, who is named in its headline, so we won't guess at it.

The most useful detail here is the silence rule. Most news bots post everything they find, and within a week the channel turns into a feed nobody reads. Here, an empty topic means no message. A morning notification from such a bot carries information simply by appearing.

The economics: $0.00004 per article

The cost of scoring barely affects the budget. Take 1000 articles a day as an example:

• 1000 × $0.00004 = $0.04 per day;
• about 260 weekdays a year come to roughly $10.4 per year.

At 200 articles a day, that is about $2 per year. The money goes elsewhere: search queries, generating the digests themselves, and the time of whoever maintains the system. The digest most likely costs more than scoring, since it is a longer text. The announcement gives no such figure, so you will need to measure it at your own volume.

Key parameters

• Schedule: weekdays, 08:00.
• Orchestration: n8n.
• News source: You.com search.
• Scoring: each article gets a materiality score, ≈ $0.00004 apiece.
• Output: one digest per topic in Slack, with links.
• Empty topics: no post is published.

Weak spots

The "materiality" score remains subjective. The announcement does not name the model, the cutoff threshold, or the scoring accuracy on labeled data. Without those numbers, there is no way to tell how often the filter misses important stories or lets noise through.

There are also general risks. The same story arrives from a dozen sources, and without deduplication it can inflate a topic's score. Generation can distort a digest, and the links act as insurance here: a reader can check every statement. Search results are not neutral either, and skew toward the sources the engine indexes best.

How to replicate it

Start small. Pick three to five topics that genuinely affect your decisions, such as model releases, GPU prices, or regulatory news. Connect one Slack channel and one source. For the first two weeks, save every score to a spreadsheet along with the headlines, including the discarded articles. That sample will let you tune the threshold and see what the filter lets through.

After that, manually check ten or so rejected articles once a week. If important ones show up among them regularly, the threshold is too high. Track the cost of the whole digest separately: scoring is almost free, while generation and search will make up most of the bill. Keep the "no news, no post" rule as a principle. It is what keeps the team's attention on the channel.