Claude pitches what your content is missing.

What it is

Pulse is a weekly loop. Claude reads the content you made this week against a catalogue of the components you already have, and logs the moments where a new one would show the idea better than anything in the catalogue. You approve or pass. Approved ideas come back as two or three rough stills, you pick one, and it is built once into your library so every future piece can use it.

It is for anyone with a repeatable content system: Reels and short video built from motion templates, sales decks built from slide layouts, email templates built from blocks, ad creative built from a set of formats. If you keep rebuilding the same kind of thing, this is for you.

We use it for the motion graphics in our own Reels. The idea is not ours: it comes from a Claude “mod” that Joshua Blatch (@joshuablatch) showed in a Reel. This page is how we set it up, written so you can set it up too.

The rule

Three reasons. Building costs time and tokens, and most ideas don’t deserve it. Reading a pitch takes you ten seconds; reviewing a half-built component takes ten minutes. And when Claude builds before you agree, you end up with one-offs scattered across old projects instead of a library.

Three guards keep the pitches worth reading:

• At most three pitches a week. Zero is a fine answer. Weak pitches are noise and cost you attention.
• Never re-pitch a passed idea unless something real changed, and then the pitch has to say what.
• Every pitch names the closest existing component and says why it fails. If Claude can’t name one, it hasn’t read the library.

The loop in six steps

1. Read

   Claude reads this week’s content line by line and asks of each moment: does the component on screen show the idea, or only carry the words?

   You need · the week’s posts, scripts or slides with times or line numbers · a catalogue (one line per component: name and what it shows) · the ledger.

2. Pitch

   Up to three ideas go into the ledger with status pitched: where (line and time), the idea in one or two sentences, and why the closest existing component fails.

   You need · the gap-pass prompt (A below) · a ledger file Claude can append to.

3. Approve or pass

   You read each pitch and answer: approve, pass, or later. Passed ideas stay in the ledger so they are never pitched again. Later ones wait.

   You need · somewhere to answer: edit the ledger’s status column, or give yourself buttons (see below).

4. Two or three still directions

   For each approved idea Claude makes two or three single frames that are genuinely different takes: a different layout, metaphor or motion idea each. Not three colourways. No animation yet.

   You need · the directions prompt (B) · your brand colours and fonts · a way for Claude to render a still (an HTML page it screenshots is enough).

5. Pick

   You look at the stills side by side and pick A, B or C. A note on what to change goes with the pick.

   You need · the stills in one place · the same answer spot as step 3.

6. Build once into the library

   Claude builds the pick as a reusable component, with its changing parts as parameters, a README and a preview, and adds it to the catalogue. From then on every piece can use it, and next week’s gap pass knows it exists.

   You need · the build prompt (C) · a library folder with a consistent structure.

Statuses move in one direction: pitched → approved → directions → picked → built. At any review an idea can also become passed or later.

Copy-paste prompts

These work in Claude Code (best, since it can read and write your files) or in a Claude chat with the files attached. Replace anything in square brackets. Swap “component” for whatever your library holds: scenes, slide layouts, email blocks, ad formats.

A · The weekly gap pass

This is my weekly gap pass. IDEAS ONLY: do not build, draw, render or code anything in this pass.

Read these first:
- CATALOGUE: [path or attachment] — every component in my library, one line each: name and what it shows.
- THIS WEEK: [path or attachment] — this week's [posts / scripts / slides / emails], with times or line numbers.
- LEDGER: [path to ledger.csv] — every idea pitched so far and its status.

Go through THIS WEEK line by line. For each moment ask: does the component on screen SHOW the idea, or does it only carry the words?

Pitch a new component only where it would explain clearly better than anything in CATALOGUE. A good pitch is a picture of the idea that no existing component can make. It is not a restyle, not a layout fix (list those separately under "Notes"), and not a one-week gimmick.

Rules:
1. At most 3 pitches this week. Zero is a fine answer.
2. Do not pitch anything already in LEDGER with status pitched, approved, directions, picked or built.
3. Do not re-pitch anything with status passed unless something real has changed. If it has, say exactly what changed.
4. Every "why" must name the closest existing component in CATALOGUE and say why it fails here.

For each pitch, append one row to LEDGER with status "pitched":
id, week, source (which post/deck/email), at (line or time range), line (the exact words), title, idea (one or two sentences), why.

Then show me the pitches as a short list and stop. Wait for me to approve, pass or say later.

B · Still directions for an approved idea

Idea [id] in LEDGER is approved. Make 2 or 3 still directions for it. Stills only: no animation, no final build.

- Each direction is ONE representative frame at [1080x1920 / 1920x1080 / the size of my format].
- Use my real brand: [colours, fonts, or the path to my brand file]. Use the real words from the line the idea was pitched for, not placeholder text.
- The directions must be genuinely different takes on the idea: a different layout, metaphor or way of moving each. Not three colourways, not three font sizes of the same picture.
- Label them A, B and C, each with one line saying what makes it different.
- Build each as a simple static HTML page and screenshot it, or render it directly if you can. Save as [library]/_ideas/[id]/a.png, b.png, c.png.
- In LEDGER, set the idea's status to "directions" and add the three labels to its history.

Then show me the stills side by side and stop. Wait for my pick.

C · Build the pick once

I picked direction [A / B / C] for idea [id]. [Optional: what to change.] Build it once into the library so every future [post / deck / email] can use it.

- First read two existing components in [library] and match their structure and conventions exactly.
- Create [library]/[component_name]/. Everything that changes from one use to the next (words, numbers, colours, timing) is a parameter with a sensible default. Nothing is hard-coded to this week.
- Write README.md: what it shows, when to use it and when not to, every parameter with one worked example, and the closest other component it should not be confused with.
- Make a preview: one still (preview.png) and, if it moves, a short loop.
- Add one line for it to CATALOGUE.
- In LEDGER, set the idea's status to "built", record [component_name], and add "built" to its history.
- Use it in this week's piece only if it genuinely fits.

Do not change any other component. Tell me what you built and where.

D · The ledger

One CSV file is enough. It is the memory that stops Claude pitching the same idea twice, so keep it in the same folder as the library and never hand-delete rows. Mark them passed instead.

id,week,source,at,line,title,idea,why,status,pick,component,history
w07-1,2026-W07,deck-q3-review,slide 4,"Most of our leads come from referrals",Referral web,"One customer in the centre; lines draw out to the people they told, then to who those people told, so the viewer sees one happy customer turn into many.","bar_chart shows a total but not where it came from; icon_row has no connections between items.",pitched,,,"2026-W07: pitched"

# status is one of:
# pitched     Claude logged it; waiting on you
# approved    you said yes; Claude makes 2-3 stills next run
# directions  stills are ready; waiting on your pick
# picked      you chose A, B or C (the pick column says which)
# built       it is in the library (the component column names it)
# passed      you said no; never re-pitch unless something real changed
# later       parked; stays in the ledger

Delete the comment lines before you save the file; they are only here to explain the statuses.

Buttons, not copy-paste

The loop only works if answering is easier than ignoring it. Typing “approve 2026-w07-1” back into a chat every week gets old fast. So we gave ourselves a page with buttons.

Ours is a private Claude artifact page, the kind Claude can publish and later update, with a small database attached. Claude writes the week’s pitches and stills into it. Each pitch shows the line, the idea and the why, with Approve, Pass and Later buttons and an optional note. Ideas with stills show them side by side with A, B and C. A tap saves the answer to the database. On the next run Claude reads the answers back, updates the ledger, and clears the answers it has applied so a stale tap is never applied twice. Nobody pastes anything.

Whether you can build that depends on what your Claude account offers: artifact pages that keep data are not available everywhere. If yours does, ask Claude for it:

Build me a private approval page for my Pulse ledger as a Claude artifact with a small database.

- Two collections you write: "ideas" (one document per ledger row: title, line, at, idea, why, status, and the still URLs when there are directions) and "library" (one per component: name, what it shows, preview still).
- One collection the page writes: "verdicts" ({id, verdict: approve | pass | later | a | b | c, note, at}).
- Show ideas waiting on me first: pitched ideas with Approve / Pass / Later buttons and a note field; ideas with directions as stills side by side with A / B / C buttons.
- Under that, the library as a grid of previews, and the passed and built ideas so I can see the history.
- On every run of my gap pass from now on: read "verdicts" first, apply them to LEDGER, then delete the verdicts you applied.

If you can’t, the fallbacks are honest and fine:

• A table with a status column. Put the ledger in Notion, Airtable or a Google Sheet, give yourself a status dropdown, and point the gap pass at an export of it each week.
• The CSV itself. Change the status cell to approved, passed or a / b / c and save. Less pleasant, works the same.

Whichever you choose, the rule holds: Claude reads your answers from the same place every time, and never acts on an idea you have not answered.

What it changed for us

We added the loop to our weekly Reels in October 2026. The first gap pass pitched three graphics, each tied to a specific line in a Reel and each naming the scene it would replace. I approved all three, picked one still for each, and all three are now built into our motion library: a grid that compares answers side by side, a review card, and a countdown overlay.

The bigger change is quieter. Before, a scene that didn’t quite fit got patched inside one post and forgotten. Now the gap gets written down, I decide once, and the fix is there for every post after it.