ingridleiria Newsletter Post Writer
The failure is not bad writing. It is the competent, forgettable essay: 1,800 words on the future of hybrid work, or on why culture matters, assembled from things the reader already believed, with no moment in it that could only have happened to this author. Nobody objects to it. Nobody finishes it either, and nobody forwards it, which is the only distribution a newsletter has.
What it costs is specific. A publication with a weekly cadence has roughly fifty chances a year to be worth someone's Sunday morning, and readers decide within the first paragraph. Open rates decline slowly and quietly, and by the time the number is obviously falling the author has spent a year of Sundays producing material that could have been generated by anyone. The expertise was real; it was simply never on the page, because the piece stayed at the altitude where nothing has to be checked and nothing can be wrong.
This skill writes the other kind: a post about one Tuesday, one number, one mechanism, argued by someone who was there.
When to use this, and when not to
Use it to draft a post from notes, to restructure a draft that has three topics in it, to rescue a piece the author says sounds generic, to turn a piece of research or an internal analysis into a public argument, and to plan a run of issues so they do not all say the same thing.
Do not use it for short-form social writing, which is linkedin-post-writer. The two are not scale versions of each other: a LinkedIn post carries one idea and one piece of evidence, while a newsletter post carries one idea, four or five pieces of evidence, and an objection.
Do not use it for an organisation's announcement, a product update or an all-staff note, which is ceo-communications. A newsletter post argues; an announcement informs, and forcing an announcement into this shape reads as manipulation.
Do not use it for an academic introduction, which has different obligations and belongs to introduction-writer. Do not use it to verify sources, which is literature-verification and runs before drafting. Do not use it for the final voice pass, which is human-voice-editor and runs after. Do not use it to decide the visual treatment or the sign-off convention of the publication, which is a brand skill, breaktalk-brand in this library.
What you need before starting
The one thing the post says, in a single sentence. Not the topic. If the sentence cannot be written, the post has two subjects or none, and every hour spent drafting will be spent twice. Missing: write three candidate sentences and ask the author which one they actually believe. That question usually settles it in a minute.
A concrete scene. A conversation, a number that surprised the author, a mistake they made, something they watched happen. It must be real, it must be theirs, and it must connect to the argument before the first section ends. Missing: interrogate the week for one. If there is genuinely nothing, use a number as the scene, which is the variant described in the second scenario below.
The lens, named. The discipline or practice the author brings that a reader could not have supplied: economics, operations research, clinical practice, a particular professional craft. The lens must explain a mechanism, not merely be mentioned. Missing: ask what the author knows that makes them read this differently from an intelligent outsider, and write down the mechanism, not the field.
Two to five sources. A paper, a dataset, a public filing, the author's own figures. Verified before use, because a newsletter that cites a study which does not exist loses the reader who checks, and one reader always checks. Missing: draft with visible placeholders and do not publish until they are filled or the claims are softened.
The reader and what changes for them. One sentence on who is reading and what they could do or think differently on Monday. Missing: propose one and confirm it; a post written for everyone gets read by nobody in particular.
Three to five of the author's previous posts. For rhythm, opening habits, connectives, and the words they never use. Missing: proceed and flag the piece as voice-unmatched so the author reviews the register rather than only the argument.
The publication's constraints. Cadence, typical length, whether there is a paywall and where the fold sits, section conventions, the sign-off. Missing: use the defaults in the length table below and confirm before publishing.
The method
Write the one sentence and test it twice. Test one: would a knowledgeable reader be able to disagree? If not, it is a description, not a claim, and the post will have nothing to argue. Test two: does it name a mechanism rather than a sentiment? "Standups grow because the cost of speaking is zero and the cost of listening is paid by everyone else" is a claim. "Meetings are getting worse" is a mood.
Choose the scene against four rules. It happened to the author. It has at least one number, name or time in it. It connects to the claim by the end of the first section. It can be told without breaching a confidence. If a candidate scene fails the last rule, change the identifying details and say in the text that you have done so; do not tell it as though it were public.
Fix the lens on one mechanism. Write the mechanism in a sentence before drafting: what force acts on what, and what result follows. A lens that is named but never does work is decoration and every reader can tell, because the post reaches the same conclusion with the lens removed.
Assemble and verify the evidence, then decide what each source is for. Give each source a job: one establishes that the pattern is general and not just the author's week, one supplies the number the argument turns on, one shows the mechanism operating somewhere else, one is the counter-evidence. A source with no job is padding and gets cut before drafting rather than after.
Build the spine before writing prose. Three to five section headers, each a claim rather than a topic, in the order that makes the argument survive a sceptic. Read the headers alone: they should make the case on their own. If they do not, the problem is structural and no amount of good sentences will hide it.
Write the opening cold, 150 to 300 words. The scene, in the first person, with its specifics. No thesis statement yet, no framing of why the subject matters. The reader should finish the opening wanting to know why this moment is being told to them. Then one paragraph, the turn, which converts the scene into the question the post answers.
Write each section to the same internal shape. The claim in the first or second sentence. The evidence, with the source named in the text and the link on the source's name rather than on "here" or "this study". One concrete instance, preferably the author's own. The lens doing its work on the instance. Vary section length deliberately; four sections of near-identical length read as a template.
Write the counter-case as a real section. Find the strongest objection, not the weakest, and decide by evidence which of two things to do. If the evidence answers it, state the objection in its strongest form and answer it. If the evidence does not answer it, concede it and say what would have to be true for the argument to fail. Conceding is stronger than a weak rebuttal, and readers who work in the field know which one they are reading.
Close by returning to the scene, 100 to 200 words. What that Tuesday looks like now that the mechanism has a name. Then the takeaway in one or two sentences. Then stop. No summary of the post, no promise of a next instalment, no line about what this means for all of us.
Write the title and subtitle last. Titles written first describe the post the author intended rather than the one that exists. Where the publication uses the colon form, the shape is broad concept, colon, specific mechanism. The subtitle carries the specific promise. Then write the preview text, three sentences that stand alone in an inbox and do not repeat the title.
Cut to length by rule, not by feel. Cut in this order: the section that repeats another section's claim, the second example where one would do, the paragraph of background the reader can find elsewhere, the qualifications on claims nobody will dispute. Do not cut the counter-case and do not cut the specifics, which are the two things most often cut and the two things that carry the piece.
Run the voice pass and the link check. Hand the draft to
human-voice-editor, then open every link and confirm each source says what the post says it says. This is a ten-minute step that prevents the two failures readers actually punish.
Length and structure by post type
| Type | Words | Sections | Evidence | Counter-case | | Single idea from the author's own week | 1,200 to 1,800 | 3 | 2 sources | one paragraph | | Argued essay with outside evidence | 1,800 to 2,600 | 4 | 3 to 4 sources | full section | | Researched piece or data-led analysis | 2,600 to 3,500 | 4 to 5 | 4 to 5 sources, one a dataset | full section, plus limits | | Reaction to something public | 900 to 1,400 | 2 to 3 | the thing itself, plus 1 | one paragraph |
Longer is not better. The post is as long as the argument, and the most common length error is padding a 1,400-word idea to 2,400 words because the publication once ran long pieces.
Worked example
Situation. Marta Sende writes a fortnightly newsletter for people who run operations in firms of twenty to two hundred people. She had about 2,100 subscribers, an open rate around 41 percent, and a growing sense that the last four issues had been interchangeable. The trigger was mundane: her own Monday planning meeting had grown from fifteen minutes to fifty over eleven months, with the same eight people and no more work in the business.
Task. One issue, out on the Thursday, seven days away, on why that happens and what to do. Good meant a reader could name the mechanism afterwards and could apply it to a meeting she had never seen.
Action. The first draft opened with a section on the state of meetings, quoting a widely circulated survey figure about hours lost to unnecessary meetings. It was abandoned after 700 words for two reasons. The figure could not be traced past a press release to a method, which failed the verification step outright. More importantly, the opening was interchangeable: any of forty newsletters could have run it that week, which was the exact problem the issue was meant to solve.
The restart used Marta's own calendar. She had eleven months of invitations and could count the durations, so the scene became a specific fact: the same meeting, the same eight people, fifteen minutes in February and fifty in December, with the growth happening in five discrete jumps rather than gradually. Each jump followed a week in which something had gone wrong, and each had added a standing agenda item that was never removed.
The lens was economics, and the mechanism was written down before drafting: the cost of adding an item is borne by the person who adds it and is close to zero, while the cost of listening is borne by everyone else and is never counted, so a meeting is a commons and accretes until someone with authority prunes it. That sentence became the spine of the piece.
Evidence was then given jobs. Her calendar export supplied the numbers. A published study on meeting load in knowledge work established that the pattern was general, cited with its authors, year and actual finding rather than a headline. A colleague at another firm who had cut a standing meeting from six items to two supplied a case that was not Marta's own. The counter-evidence was the strongest objection she could find: the meeting had grown because the business had genuinely become more complex, in which case the growth was correct and the mechanism was irrelevant.
She could not dismiss that objection, so she conceded part of it. Headcount had risen by nine over the same period, which explained some of the growth. What it did not explain was that no item had ever been removed, and she said so plainly: if the meeting were tracking complexity, items would leave as well as arrive.
A wrong turn survived until the second draft. A section on tools, comparing three ways of running asynchronous updates, was written and cut. It was useful advice and it belonged to a different post, and its presence meant the piece had two claims and therefore none.
The final structure had three sections with claim headers, 1,750 words, a close that returned to the December calendar entry, and a title in the colon form: the concept, then the mechanism.
Result. The issue went out on the Thursday. Open rate was 46 percent against a baseline of 41, which is one issue and proves little on its own. The more useful signal was 22 replies against a typical three to five, and nine of the replies contained the reader's own count of how their equivalent meeting had grown, which is the response a specific piece gets and a general one does not.
What did not work: the practical section at the end, on how to prune a standing agenda, drew no comment at all and Marta suspects it was skimmed. Her judgement afterwards was that the piece earned its attention on diagnosis and spent it poorly on prescription, and she has not resolved that tension in subsequent issues.
A second scenario, where it goes differently
Three months later she wanted to write about a newly published national dataset on small firm productivity. There was no scene: she had not been anywhere, nothing had happened to her, and the material was entirely public.
Two things changed. The scene became a number rather than a moment. The post opened on a single figure from the release that contradicted what the reader would expect, stated in two sentences with no framing, which does the same job as a moment: it creates a question the reader wants answered. The rule that the scene must be the author's own was replaced by a stricter rule that the number must be one the author computed or checked themselves from the source data, not one lifted from a summary.
The counter-case grew from a paragraph to the longest section in the post, because the audience for a data piece includes people who know the dataset and will test the claim. Limits went in explicitly: what the series does not measure, the definitional change in 2021, and the size of the effect relative to its uncertainty. The piece ran to 3,100 words, at the top of the range, and that length was correct because the objections were real.
What changed: the absence of a personal scene raises the evidential bar rather than lowering it. A post with no anecdote must be more precise, not more general, and the commonest mistake is to compensate for a missing story with a broader argument, which produces exactly the interchangeable essay this skill exists to prevent.
Output
Delivered together, ready to paste:
TITLE: [colon form where the publication uses it: concept, then mechanism]
SUBTITLE: [the specific promise, one line]
PREVIEW: [three sentences that stand alone in an inbox and do not repeat the title]
WORDS: [count] TYPE: [from the length table]
THE ONE THING: [the single sentence the post argues]
THE LENS: [the mechanism, in one sentence]
[The post in Markdown: opening scene, turn, claim-headed sections,
counter-case, close returning to the scene.]
SOURCES
| # | Source, as cited in the text | What it is doing in the post | Link | Checked |
QUERIES
[Any specific the author must supply before publication.]
For an edit rather than a draft, deliver the diagnosis first, in this shape, so the author sees the pattern:
| Element | Present? | Note | | Opening is a scene, not an abstraction | | | | The one thing, as a single sentence | | | | Each section header states a claim | | | | Each claim has a named source in the sentence | | | | Counter-case stated at its strongest | | | | Close returns to the opening and stops | | |
Failure modes
Two posts in one. Recognise it when the one-sentence test produces a sentence with "and" in the middle. Split it and publish the second next issue; a merged post argues neither half.
The lens as label. Recognise it by deleting every mention of the discipline and checking whether the argument changes. If it does not, the lens was never used, and the repair is to write the mechanism sentence and rebuild the middle sections around it.
The scene that never returns. An opening anecdote followed by a general essay is the most common structural failure, and readers experience it as a broken promise. The close must come back to the specific thing.
The unverifiable number. A striking figure that traces back only to a press release or a secondary summary. Recognise it when the search for the method stops at a blog post. Cut it, however good it is; one reader will follow the link.
The straw counter-case. Recognise it when the objection is answered in a single sentence. Real objections take a paragraph, and answering a weak one signals that the strong one was avoided.
The uplift close. Recognise it by any final sentence containing "as we look ahead", "at the end of the day", or a first-person plural the rest of the post did not use. Delete it. The correct final sentence is usually the one before.
Padding to a habitual length. Recognise it by background paragraphs the reader could find elsewhere and by the second example that adds nothing. Cut by the order in step 11.
Confidential material as colour. Recognise it when a scene identifies a client, an employer or a colleague who did not consent. Change the details and say so, or use a different scene. A newsletter that burns a source loses more than an issue.
Edge cases
Nothing happened this fortnight. Do not manufacture a scene. Use the number-as-scene variant from the second scenario, or write the shorter reaction piece from the length table, or skip the issue. A visibly invented anecdote costs more than a missed week.
The evidence contradicts the author's intended claim. Change the claim and write the more interesting post about being wrong, with the original position stated honestly. This is the single best piece a newsletter can run and authors reliably avoid it.
A paywall sits in the middle. Structure so the free portion contains the scene, the turn and one complete claim with its evidence, and place the fold at a section boundary rather than mid-argument. A reader cut off inside a claim does not subscribe, they leave.
The subject is contested or sensitive. Raise the evidential standard, attribute every claim, state the author's position and interest explicitly, and answer the strongest objection rather than the most convenient one. Where the author has a financial or professional stake, disclose it in the post rather than in a footer.
The material is genuinely technical. Keep the technical precision and move the derivation into an appendix section after the close. Simplifying a mechanism until it is wrong loses the readers who matter most.
The piece is a commission with an editor's brief. The brief overrides the length table and the structure, but not the verification standard or the counter-case. Say which constraints came from the brief when handing the draft over.
Quality bar
- One sentence states what the post says, and no section argues something else.
- The post opens on a specific scene or a specific number, within the first two sentences, with no preamble on why the subject matters.
- Every claim carries a named source in the sentence that makes it, and every source has been opened and checked.
- Every section header states a claim that a reader could dispute.
- The strongest objection appears, stated at its strongest, and is either answered or conceded with its consequence.
- The close returns to the opening and the final sentence is the point.
- The lens does work: removing it would change the argument.
- Length matches the type in the table, and nothing was padded to reach it.
Adapting this to your context
The defaults come from a fortnightly essay newsletter of about 2,100 subscribers, written for operations readers through an economics lens.
- The length table. 1,200 to 3,500 words across four types. A weekly cadence, a daily briefing or a technical blog needs its own bands, taken from where your readers stop.
- The reply signal. The example calls 22 replies a success against a baseline of three to five. On a list of 200 or 20,000 that reads differently. Set your own baseline over four issues.
- The lens. Economics here. Substitute whatever explains a mechanism an intelligent outsider could not: clinical practice, law, epidemiology, teaching. The test holds either way: delete the lens and see whether the argument changes.
- The source standard. Two to five sources, each with a job. Where the evidence is a case file, an archive, an interview transcript or a code repository, keep the jobs and change what fills them.
- Publication conventions. The colon title, the preview text and a paywall at a section boundary are Substack-shaped. Check where your platform actually cuts.
- What not to change. One idea per post, a specific opening the close returns to, and the strongest objection stated at full strength and then answered or conceded.
Related skills
linkedin-post-writer takes the strongest single idea from a finished issue and turns it into short form; the two share evidence but not structure. literature-verification runs before drafting on anything citing research, and its standard applies to every source here. human-voice-editor is the final pass and this skill hands every draft to it. breaktalk-brand supplies the title convention, sign-off and visual treatment where the publication is that one. ceo-communications covers the organisational announcement this deliberately excludes, and external-insights supplies the raw research a data-led issue is built from.