How to automate newsletter curation: selection is the product

The reason people subscribe to a curated newsletter is that somebody read a hundred things so they did not have to. The value is entirely in the rejection — in the ninety-five items that did not make it — and that is precisely the part automation is worst at.

Which does not mean the work cannot be automated. It means the automation belongs on either side of the judgement: gathering candidates so the curator is not searching, and producing the finished issue so they are not formatting. Between those two sits a person deciding, and the newsletters that removed that person are the ones people stopped reading.

What we'll cover

• Curation is rejection
• What automation does well here
• Building the candidate pool
• Building it with CodeWords
• Filtering before a human looks
• The annotation is the value
• Assembly and the weekly rhythm
• Sponsorship, and keeping it legible
• Making it survive
• Limits worth knowing about
• What to build first
• Frequently asked questions

Curation is rejection

Anyone can aggregate. A feed of everything published on a topic is a commodity and costs nothing to produce.

Selection implies a point of view, which is what a reader is subscribing to whether or not they would phrase it that way.

Consistency builds the trust. A reader learns what your inclusion means, and that meaning is the asset.

Volume destroys it. A newsletter with thirty links has made no decisions and is asking the reader to do the work again.

The unpopular inclusion matters most. The item that would not have surfaced algorithmically is the one that justifies the subscription.

An automated feed is not a newsletter, and readers identify the difference quickly even when they cannot articulate it.

What automation does well here

Finding candidates across sources the curator would not check individually.

Removing what is already seen — duplicates, syndicated copies, follow-ups to stories already covered.

Filtering out the obvious noise, which is most of the volume and none of the judgement.

Extracting the facts: publication date, author, source, a summary for the curator to read rather than the whole piece.

Assembling the issue from selected items into the finished format, which is tedious and entirely mechanical.

Scheduling, sending and archiving, which is pure operations.

Building the candidate pool

Cast wider than you would read, since the point of the automation is that breadth costs nothing.

Include the awkward sources: newsletters you receive, forums, mailing lists, release notes, regulatory notices. These are where the non-obvious items come from.

Capture the date properly, since republished old material is a persistent failure of feed-based sources.

Deduplicate on content, not on link, because syndication produces many URLs for one article.

Record where each came from and track which sources actually make it into issues, because most contribute nothing and the pool can be pruned.

Keep a manual add, since the curator will find things themselves and that should be the same pipeline rather than a separate one.

Building it with CodeWords

CodeWords connects to more than 3,000 integrations, so the sources, the review queue, the email platform and the archive can meet in one workflow.

1. Open CodeWords and start a new automation.
2. Describe what should happen in plain language to Cody, the automation builder: which sources, the filters, and where the shortlist appears for review.
3. Authorize the connections — the feed and mail sources, your review destination, and the sending platform.
4. Describe the exceptions: a duplicate, an item already covered, a paywalled source, an item older than a defined window.
5. Run it for two weeks alongside your existing process and compare the shortlist with what you actually chose.

You describe the outcome; Cody builds it, connects it, and deploys it. The free plan covers light use, with Pro at $39 per month and Business at $100 per month as usage grows; details are on the pricing page.

Filtering before a human looks

Remove the duplicates and syndications first, which is the largest single reduction.

Drop anything previously included, which requires keeping a proper archive of what you sent.

Filter by date to your window, strictly.

Remove the formats you never use — job listings, sponsored posts, press releases — as a standing rule rather than a repeated decision.

Rank rather than cut for everything else, since a low-ranked item the curator still picks is exactly the kind of signal you want to preserve.

Target a shortlist a person can actually read, around twenty to thirty items for a weekly issue, which is the number that makes the review a twenty-minute job rather than an afternoon.

The annotation is the value

Why this matters, in the curator's voice, is the part readers quote and the part no summariser produces.

The connection between items — this relates to the thing from three weeks ago — is available only to somebody holding the whole context.

Disagreement is worth including. A link with a note explaining why you think it is wrong is more valuable than an uncritical summary.

Summaries are not annotations. A generated précis tells the reader what the article says and not why they should care.

Keep them short. Two sentences is usually right, and the discipline of two sentences forces the point of view to be explicit.

Write them at selection time, while the reason for choosing is still present, rather than at assembly time when it has faded.

Assembly and the weekly rhythm

One click from shortlist to draft, preserving order and annotations, so the mechanical step disappears.

Consistent structure issue to issue, because readers scan rather than read and a stable shape makes that possible.

Validate the links before sending, since a broken link in a link newsletter is the most embarrassing available failure.

Archive publicly with stable URLs, which also becomes the previously-included list the filter uses.

Same day, same time, since a curated newsletter is a habit and an irregular one is not.

Skip an issue rather than padding it. A week with four good items and a note saying so is better than four good items and six filler ones.

Sponsorship, and keeping it legible

Most curated newsletters eventually carry sponsorship, and the handling of it decides whether the trust built by good selection survives the revenue.

Label it unmistakably, which is a legal requirement in most markets and a trust requirement everywhere.

Keep it structurally separate from the curated items rather than styled to blend in, since a sponsored item sitting in the selection undermines every unsponsored one.

Do not let it into the selection. The moment a sponsor can buy a link in the curated section, the section means something different.

Disclose affiliate links individually, not in a footer nobody reads.

Automate the placement, not the decision about whether to accept, which is an editorial judgement about the audience.

Report sponsored performance separately, because mixing it into overall engagement figures makes both numbers harder to interpret.

Making it survive

Handle a source going away, which happens constantly with feeds, by reporting it rather than silently losing coverage.

Respect robots and terms on the sources you read, and link rather than reproduce.

Do not republish substantial extracts, which is both a copyright question and the thing that removes the reason to click.

Keep the archive as the source of truth for what has been sent.

Report the outcome: candidates gathered, filtered, shortlisted, selected, and which sources contributed.

Limits worth knowing about

Feed quality varies enormously, and many sources provide partial content or unreliable dates.

Paywalled sources need a decision about whether to include items most readers cannot open.

Automated summaries misrepresent with some regularity, which matters more when you are pointing readers at somebody else's work.

Relevance filtering has a blind spot for the genuinely novel, since novelty by definition does not match prior patterns.

A high-volume topic overwhelms a weekly format, which is a scope decision rather than a technical one.

What to build first

The candidate pipeline: gather, deduplicate, filter out anything previously included, and present twenty to thirty ranked items with dates and summaries. Nothing is selected and nothing is sent. It removes the searching, which is the largest part of the time cost, and leaves the judgement exactly where it belongs.

Two habits make the difference. Keep the archive as a real previously-included list, since re-sending an item is the failure readers notice immediately. And write the annotation at selection time rather than during assembly, because the reason you chose something is vivid for about a minute and reconstructing it later produces the bland summary you were trying to avoid.

Frequently asked questions

Can newsletter curation be fully automated?

The gathering and the assembly, yes. The selection is the product — readers subscribe because somebody read a hundred things and rejected ninety-five — and an automated feed is identifiable as one quite quickly.

How many items should a curated issue have?

Few enough that the selection is visible. Thirty links means no decisions were made and the reader has to do the work again, which is the thing they subscribed to avoid.

What should the automation filter out before a person looks?

Duplicates and syndicated copies, anything previously included, items outside the date window, and formats you never use. Rank everything else rather than cutting it.

How large should the shortlist be?

Around twenty to thirty for a weekly issue, which makes the review a twenty-minute job. Much larger and the curator is doing the filtering again.

Are automated summaries a substitute for annotations?

No. A summary says what the article contains; an annotation says why it matters and how it connects to something from three weeks ago. The second is what readers quote.

What sources are worth including?

Wider than you would read yourself, including the awkward ones — newsletters, forums, mailing lists, release notes, regulatory notices. Track which sources actually make it into issues and prune the rest.

Should I skip an issue with too little material?

Yes. Four good items and a note saying it was a quiet week is better than four good items and six filler ones, which is how readers learn the selection no longer means anything.

Is there a copyright issue with curation?

Link rather than reproduce, and avoid substantial extracts. It is both the safer position and the one that preserves the reason for a reader to click through.

How should sponsorship be handled in a curated newsletter?

Labelled unmistakably, structurally separate from the curated items rather than styled to blend in, and never inside the selection itself — the moment a sponsor can buy a curated link, the whole section means something different.

Do affiliate links need individual disclosure?

Yes, next to the link rather than in a footer. It is a requirement in most markets and it is also the version readers actually see.

How do I know which sources are worth keeping?

Track which sources actually produce items that reach an issue. Most contribute nothing over a quarter, and pruning them makes the shortlist better rather than smaller.

What does automation miss when filtering for relevance?

The genuinely novel, since novelty by definition does not match prior patterns. That is the argument for ranking rather than cutting, and for keeping a manual add in the same pipeline.

Related reading

• Automating blog post distribution
• Automating PR monitoring
• Automating a weekly team digest