Why do the same corrections keep coming back when Claude writes my content?

Because the correction lives in the post you fixed, and the next run starts from the template you built in month one. Nothing is wrong and nothing warns you — the new posts are fine in the same wrong way. The fix is to make the last approved post of each type the template, and to write every correction into a log, separating one-off fixes from patterns worth keeping.

How long does it take to write a fortnight of posts with Claude?

About an hour, against eight hours by hand: ten minutes taking stock, twenty-five writing all the copy in batches, and twenty-five building the visuals. That is roughly seven hours back per fortnight, or most of two working days a month.

Produce a Month of Posts — 3rd Space AI Workflows

Q: How do I stop Claude inventing examples in my posts?

Name one source file that holds your real anecdotes, cases and figures, and write the rule that nothing may come from outside it. Where the file does not cover a post, the slot stays empty and marked, and Claude writes the question that would fill it. An invented example reads better than an empty slide, which is exactly why it gets through.

Publishing your own content every week is really three jobs, and the third one is the one that breaks.

Job one: decide what goes out. Which subject, which day, which pillar, so the month balances. Job two: write and build it. The copy, the visual, the caption. Most people do these two and think they are done.

Job three: make sure what you learned this month is still true next month. You correct a post after it is built. The correction fixes that post. The next run starts from the template you made in month one, and the same line comes back.

So this workflow does all three, in that order, and the third one is a real step with real files — a log of what went out, a table of which design is the current template, and a numbered list of rules that grows every cycle. Batching the writing saves hours. Recording the corrections is what makes the hours add up to something.

It is written for one person publishing their own practice content, not for a team running a brand account. The difference matters: a reader notices when one voice slips, and does not notice when a brand's does.

# Monthly Post Plan — template Work through this with Claude in conversation *before* the first run. Answer in your own words; let Claude push back and ask for specifics. The output is one setup document that every later run reads instead of asking you again. Most of this is answered once. The last two sections are answered after every run, forever — they are the part that keeps the output sounding like you. --- ## 1. Who is writing Your name and your brand name, as they appear in captions and in file names. The role you write from, in one line. > Example: a therapist who publishes her own practice content. Not a brand account, not a > team. One voice, and readers notice when it slips. ## 2. Your voice, in rules not adjectives "Warm but professional" is not a rule. A rule is something a draft can fail. - Three expressions or constructions you refuse, and why - Punctuation you do not use - One sentence of yours that is *exactly* right, and one that is close but wrong - **Keep these numbered.** Every new rule gets the next number and the example that produced it. ## 3. Two complete reference pieces Not a description of your style — two finished posts you are happy with, pasted in full. > One example is not a pattern. With two, Claude can see what they have in common. With one, > it copies the surface. ## 4. Your pillars and your quotas What subjects you post about, and how many of each per month. Say which are mandatory and which are flexible. Name any cap: how often a single lead magnet may be promoted, how often you may post in a second language. ## 5. Your rhythm Which day carries which post type. Be specific — "three times a week" is not a rhythm, it is an average. ## 6. Your post types, and the structure of each For each type: how many slides or lines, what goes first, what goes last, how it closes. > Example: the series carousel has seven slides — hook, the correction "it isn't X, it's Y", > the mechanism in two parts, an anchor from real work, an "instead of / try this" swap, and > a close with a single call to action. ## 7. What may never be invented **The most important section.** Name the file that holds your real material: the anecdotes, the cases, the personal stories, the numbers with sources. Then write the rule explicitly: nothing comes from outside that file. When the file does not cover a post, the slot stays empty and marked, and Claude writes the question that would fill it. > An invented example reads better than an empty slide. That is exactly why it gets through. ## 8. Your tools and where things live - Where the finished copy gets saved - Where the log lives (see the other tab) - Your design tool, and the account to build in - Your scheduling tool, account and time zone - Output language, and any post type that runs in a second one Do not paste credentials into this document. Say where they live. ## 9. The template rule For each post type, which design is the current template. The answer is always the same: **the last post of that type you approved.** Write the table anyway, and update it in the same run that produces a new post. If you corrected a post after it was built, your corrected version is the template. Not the produced one. ## 10. After the run: what you corrected Answered every cycle. For each correction, decide which it is: - **A one-off fix.** A word changed in one particular context. It goes nowhere. - **A pattern.** A choice you would make again, in any text. This gets written down. The test: > *If you can phrase the change as an instruction for a future text, without mentioning the > post it came from, it is a pattern.* Two occurrences make a rule. One gets recorded as a single occurrence and waits for confirmation on the next post. Do not inflate rules from one case. ## 11. After the run: what was left - The marked gaps, and the question that would fill each - The designs still waiting for your review - What was not covered, and why --- ## Appendix: the first run Give Claude the finished setup document and say: > "Produce the next fortnight. Read the log first and fill only what is missing. Take every > anecdote and figure from the source file — if it does not cover a post, leave the slot > marked and write me the question. Deliver the design links grouped by post type."

# Content log — template This is the file that stops the workflow from drifting. Keep it in the project your content runs from, and update it **in the same run** that produces a post. A run that produces posts and does not update this file has not finished: the next one will repeat content and remake corrections you already made. Read it first, write it last. --- ## 1. Published | Date | Type | Subject | Hook used | Link | |---|---|---|---|---| Everything that went out. The hook column matters as much as the subject: a spent hook coming back is harder to spot than a spent topic. ## 2. Scheduled | Date | Type | Subject | State | Link | |---|---|---|---|---| State is one of: copy written · visual ready · scheduled. This is what the next run reads to know which days are already handled. ## 3. Current templates | Post type | Current template | Produced | Approved by you | |---|---|---|---| **The template for a type is the last post of it you approved.** Not the first one designed. Update this row in the same run that produces a new post of that type. If you corrected a post after it was built, the corrected version goes here, not the produced one. ## 4. Style rules Numbered, consecutive, never renumbered. Each rule gets the concrete example that produced it, and the sentence of yours that it came from. Write the rule as an instruction, not as a prohibition where you can: > **01. Unpack nouns into verbs.** "the effort you feel", not "felt effort". Nominalised > forms sound like a manual; this copy is read aloud. A rule with one occurrence is marked as such and waits for the next post to confirm it. ## 5. Writing benchmark Not rules — construction patterns. What goes on the second slide, what a close looks like, what a swap label has to say. Point at the two reference pieces and say what they have in common. This section is what a new run reads before writing a carousel. ## 6. Rejected hooks | Hook | Rejected because | |---|---| Cheap to keep and it stops the same bad hook arriving three times in different words. ## 7. Spent material Anecdotes, cases and personal stories already used, with where they were used. A good story used twice in four months reads as a thin practice. ## 8. Open gaps | Date | Type | What is missing | The question that would fill it | |---|---|---|---| Gaps are left marked on purpose. The question is the point: it is what turns an empty slot into two minutes of your attention instead of an invented example. ## 9. Designs awaiting review Designs that inherited a background photograph from the post they were copied from. Listed so they can be reviewed in one sitting rather than one at a time. ## 10. Unestablished sources Figures that went into a post without a source being established. Kept visible rather than quietly accepted.

Batch the writing, batch the design, and the part that quietly undoes bothProduce a Month of Posts

What this does

Three things that kept going wrong

How it works, in three steps

Take stock before you write a wordno writing yet

Write the fortnight in batchesall of one type at once

Build each visual from the last approved onecopy, then replace the words

What you need

The mistake that cost the same hour, every month

Your corrections don't stick, and you won't notice until the third time.

It will invent the example

The character limit does not beat the sentence

The preview lags the text

A new text box loses your font

Group the links by type, not by date

Batch a fortnight, not a month

Copy this and use it today

Copy the template

Paste it into Claude

Answer the setup questions

Run it, then record what you changed