Search-led content in this machine
The method is /session/work/create-search-content/README.md; read it first. You do the thinking
and the writing. OpenSEO, on the person’s own account, supplies the search evidence and keeps the
project’s record. The Mac makes the film. Each opportunity becomes one brief, a TOML file checked by
content.py, and nothing is drafted until the person approves it.
cd /session/work/create-search-content
mari-openseo tools # what the person's OpenSEO account offers
cp brief-template.toml briefs/<id>.toml # one brief per opportunity
python3 content.py check briefs/<id>.toml # until it says ready
python3 content.py produce briefs/<id>.toml # an approved film: outputs/<id>/<id>.mp4
- Connect. Run
mari-openseo tools. If it is refused, tell the person what it said: OpenSEO is a switch in this agent’s inspector, under Reach, Accounts & folders (on by default, and never at the Restricted level), and it needs their OpenSEO key added once in Mariwave, where it stays. Run no paid call yet. - Agree the business. Ask, with
mari-ask, for the website, audience, actual service area, offer, next step, voice, proof and claims to avoid. Never infer the business from its name. Separate what they confirmed from what you assume, and save it to the OpenSEO project’s context only once they approve. - Research. Every
mari-openseo callspends the person’s OpenSEO credits: say which calls you will make and ask before the first, and again before widening the research or saving keywords. Use the keyword research, search result, competitor and clustering tools the account lists. Read the ranking pages themselves withmari-web fetch. Write down each metric with its market and the date it was read; one that did not come back is “unavailable”. - Brief. One
briefs/<id>.tomlfor each opportunity worth making: the question, the evidence, what this business can show that the ranking pages do not, one answer, one format and one next step. Each claim is an[[evidence]]with the address it was read at and the date. Never invent a customer, a case or a figure.content.py checkuntil it says ready, then ask the person to approve it, and setstatus = "brief approved"only on their word. - Script. For a film, four openings first (the line, the first picture, the promise, the
payoff), then 30 to 45 seconds of
[[scenes]]: the line on screen, the narration, the picture with nothing written in it, and the evidence it rests on. Show the scenes and get them approved before anything is made. - Make it.
content.py producemakes each picture withmari-image, reads each narration in the Mac’s own voice, scores it withmari-musicand has the Mac cut it withmari-media, all on the Mac at no charge. It also writes the transcript.mari-offer-filethe film and transcript to the person. A page, post or FAQ you write in the chat, from the same brief. - Review. The person watches the whole film. Check the scene text against the brief’s evidence and list what blocks release, what could be better, and what only their playback can check. Never say you watched the film, and never predict views.
- Publish by hand. Prepare the title, caption, description, call to action and transcript. The person publishes it; nothing here posts, schedules or says anything is live. When they give you the live address, record it, and the brief’s status, in the OpenSEO project report.
- Learn. On the next round, read the saved keyword positions through OpenSEO and compare the same keyword, market and window. A ranking change is not a view, a lead or a sale, and a film is not its cause without evidence. Say repeat, revise, expand or stop, with the reason.
What this does not do
- No realistic presenter, cloned face or cloned voice, and no footage of real events.
- No words drawn in the pictures: the film’s line is spoken, and the transcript carries it as captions where the person posts it.
- No paid video unless the person asks for one by name:
mari-videois charged to their video provider account. The pictures, voice and score are free on the Mac. - No publishing, scheduling or social analytics, and no ranking promise.
