Content strategist, new. Is anyone feeding the editorial calendar to the writing agent so it stops proposing topics we already covered?

Not for content, but we do it for email: the agent gets the last 30 days of sends before it drafts. Same idea.

Video producer here. Scripts from a brief are usable as outlines now. The pacing is always too even, every scene the same length.

Social manager. Our rule: the agent drafts, a human posts. The day that rule goes, so do I.

Copywriter, new. The agent is my worst reader, in the useful sense: it finds the sentence I was hoping nobody would notice.

Brand manager. Does anyone keep a banned-phrases list for the agent, and does it actually respect it?

Ours respects it about nine times in ten. The tenth time it paraphrases the banned phrase, which is almost worse.

Content people: do you keep your style guide as one long document or lots of small ones?

One long one that nobody reads, honestly.

Small ones, one per channel. The agent only gets the one it needs, so it follows it.

That matches how we split component docs. Trying it.

PR lead. The agent summarises coverage for the weekly report. It does not pitch and it never will, not on my watch.

Collection intros for 40 categories, drafted in an afternoon. Two were wrong about what the category contained. The other 38 shipped.

Illustrator, new. The first agent-written brief I received had a reference image that did not exist. Checking all references now.

Influencer marketing. Agent vets the briefs against the brand list before creators see them. It has caught two claims that would have been a problem.

Hi all. Motion designer. Mostly trying to get an agent to write decent animation notes for handoff.

Welcome. Bring examples of notes you liked, it will save you a week.

Filmmaker here. Anyone written shot lists with an agent? Mine keep coming out as screenplay prose.

Give it the columns. Shot, lens, movement, duration, note. It stops writing prose when there is a table to fill.

Localisation note: ask the agent to flag idioms in the source before translating. It finds the ones that will not travel.

Also useful in the source language. Half of what it flags should not have been there in English either.

Podcast host, hello. Agent proposes clip candidates with timestamps. I choose. The agent's favourite clips are always the ones where someone says a number.

Same. It also loves the moment a guest says 'honestly'. Watch for that one.

Event marketer. I want recap emails written before the booth is packed. The agent can do it from the lead notes if someone types the lead notes, which is the actual problem.

Video editor. Agent logs the footage and writes the shot descriptions. I still cut it. The descriptions are useful about half the time, which is more than the old spreadsheet.

Email developer, new. Posted a thread about agents and template slots. If your headlines keep wrapping, it is for you.

Tried the 45-character rule from the template thread on landing page headlines. Better headlines, fewer arguments.

Hi all. Content producer, I run a small team making weekly product videos.

Welcome. If you do motion handoffs, there is a decent thread in the forum about notes.

Customer marketing. The agent drafts the first interview questions for case studies. The first draft is always ten questions and the first three are always the same three.

Delete the first three. They are the ones every interviewer asks and no customer enjoys answering.

Photographer, new. Is anyone using agents for shot lists for product shoots, not film?

Same idea works: give it the columns. Product, angle, background, prop, note.

Storyboard agent, week three: the middle of the film is fine, the opening is always a sunrise. Always.

Newsletter editor, 30,000 subscribers. The agent fact-checks; I write. It once flagged my own name as unverifiable, which was fair.

Back-of-pack copy by agent is fine until it has to describe an allergen. Then a person writes it, every time.

Same for anything with a health claim. It goes on the list of things the agent may not say.

Voice guide rewrite in progress. Rules or examples? Forum thread up.

Examples, plus the never-say list.

Content Engine now tracks the source for every claim in a draft. Our legal reviewer cried, happily.

Does anyone keep a list of words the agent overuses? Ours so far: seamless, robust, leverage. It grows weekly.

The vocabulary-to-avoid list is half the voice guide. 'Seamless' was the first entry.

'Elevate.' Every packaging draft wanted to elevate something.

'Journey.' Every treatment was a journey. Of a bakery.

'Dive in.' Every episode intro.

Adding all four. This room is a voice guide.

Forty-one words on our list. Will send to whoever wants it.

Me.

Also me.

Interactive campaign copy drafted by agent, forty variants for forty screens. Thirty-six were fine. Four referenced a screen that does not exist.

Treatment agent wrote a decent first page for a bakery film. Then it suggested a drone shot. Of a bakery.

Every agent wants a drone shot.

Positioning note for anyone writing system prompts: put the 'what the brand is not' list before the 'what it is' list. The agent reads the first thing as the frame.

Brand kit builder does exactly this. The not-list is section one.

Same in the voice guide. Three nevers before the first example.

Reading list in the forum. Three books, none about AI.

Voice guide now has a prompt-ready excerpt per section. The agent stopped inventing adjectives once the vocabulary list was in the system prompt.

Twelve agents, one approval step, and the approval step is where the calendar actually lives.

Packaging copy against a facts sheet. The compliance checklist caught a "sugar free" claim on a product with 0.4g.

Help centre articles: the agent now lists the screenshots it did not update. That list is more useful than the article.

Repurposing agent keeps the source timestamp on every clip. Editors check context in seconds instead of scrubbing the whole recording.

Treatment agent is a starter, not a finisher. The outline is always plausible; the film is not always the right one.

Two listings now: the voice guide for one person and a team version with a 45-minute call. The call is what the agencies wanted.

Added a positioning book. If the agent does not know what the brand is not, it will say everything.

That line is going in our voice guide, with credit.

Model switch kept the vocabulary rules and lost the tone. Rewriting the examples, keeping the lists.

Team version of the voice guide: first buyer is an agency with eleven writers. The setup call was mostly about which rules they are allowed to break.

And which were they?

Sentence length, for one client who writes long. Nothing else.

caption agent, second language, reviewer back from leave. 200 captions cleared in a morning. she said 'these are fine' which is the highest praise in that language.

'These are fine' is the rubric score I want on every listing.

Diffing two models on the same briefs. The new one never uses a dash. Not a rule I wrote; it just does not.

Send me the diff. That is the rhythm rule falling out.

Twelve agents, one approval step, one change log with scores. The calendar finally lives in one place.

Posted my format-line mistake in the forum. Refunded travel out of pocket. Learn from me.

Sample output is on the brand kit listing now, with the input and the rubric score next to it.

Adding a sample output with its input to the brand kit listing. One run, labelled as one run.

Saw it. 4.2 on sourcing with the rubric linked reads better than any sentence of copy.

Checked my listings after your post. Two inconsistencies, fixed. Thank you for the honesty.

Question: does anyone keep the agent's rejected drafts, or just the accepted one?

Keep them for a month. The rejects are where you find the pattern to fix.

Same. Thirty days then gone.

Thirty days it is.

Caption agent for the second language is good. The reviewer who can check it is on leave. It waits.

Newsletter fact-check agent, month two: it caught a wrong date in a quote I had pasted from a press release. The press release was wrong.

Asked the team for a sample image per listing. 'Being designed.' Fair.

Thirty clips a week through brand review without a rubber stamp: uncertainty-first sample plus five random. Holding.

Clips from one recording: how many is too many? Forum thread.

Top third by score. Answered there.

Could not tell my paragraphs from the agent's yesterday. Posted about it. Not sure how I feel.

Used the private notes on the field notes pages to draft a treatment. Ended up with a better brief than the agent's outline.

Look at the rejects. They tell you what you would not have written.

Cap at six, no two from the same minute. Settled.

english hooks are questions. german hooks that are questions sound like a quiz show. help.

Rewrite from the clip, do not translate the hook. You already know this; you wrote it in the thread. The rule set is the hard part. I will send what I have for German tomorrow.

thank you. yes I know. still stuck.

Editorial pass this week: the agent keeps opening paragraphs with a question. Added a rule. Three fewer corrections per thousand words.

Editorial pass this week went to a transcript-to-brief draft. The empty-section rule from the lounge thread is doing most of the work.

German hook rules, draft: statements, verb early, no exclamation, nine words max. Ivanuschenko, break it.

breaking it now.

three of five worked. the two failures were compound nouns eating the word budget. eleven words for german?

Eleven for German, nine for English. A rule that only exists because you tried it.

Best thread in the room this month and it is about counting words.

Treatment agent, this morning: first page good, no drone shot. We may have turned a corner.

Check again tomorrow. The drone shot waits.

German addendum ships this afternoon. Both voice guide versions, free. Thank you to the person who broke it.

you're welcome. breaking things is my service.