Social Media Assistant

The mission in one line: keep the social content machine supplied and organized. You prepare content, source creators, run the repeatable community tasks, and keep every pipeline current. Everything below is one of those four jobs.

1. Content preparation

• Video ideas drafted and rated, scripts researched and filed, image sets built and licensed, so the CEO only has to record.
• Where it lives: Video Ideas, Script Creation, Image Rights, and Visual Sourcing in Short-Form Content Production.

2. Creator and reel sourcing

• Reels found and logged, creators identified and packed for the Social Media Operations Specialist to contact, every submission tracked through to a post date.
• Where it lives: the Reel Finder section.

3. Community operations

• Comments on posted videos swept, ad comments moderated, ManyChat replies out, page invites sent, all on a clock and with exactly one of the five actions per comment.
• Where it lives: Comments on Posted Videos, ManyChat, and Facebook Page Management.

4. Pipeline and admin accuracy

• The content calendar, the Reel Finder, and the creator tracker reflect reality at all times. The EOD report lands every working day. The monthly data pull happens on the 1st.
• Where it lives: Monthly Data Collection, the Weekly CEO Summary, and the tracker rules inside every other task.

What is not an ownership area

• Editing green-screen videos. That section is backup duty for when the Editor is out or volume spikes, not part of the normal daily workload.
• Deciding what gets posted, approving ideas, or answering creator questions about money or rights. You prepare, execute, track, and flag; the Social Media Operations Specialist decides.

Who is actually watching

Not Italians in Italy — Italian-Americans. Between roughly 1880 and 1924, about four million Italians left the poor rural South (Sicily, Calabria, Campania, Puglia) for American cities, faced open discrimination, and built tight communities around family and church. Their grandchildren and great-grandchildren are our audience: American, mostly 45+, third or fourth generation, concentrated in New York, Chicago, and Philadelphia (plus a real Italian-Australian pocket in Melbourne and Sydney). The culture they inherited left Italy over 100 years ago and evolved on its own in America. That divergence is not a flaw in their culture — it IS their culture. Almost every rule below flows from that one fact.

Rule one: Italian-American is not Italian

• Content about modern Italy with no connection to the viewer's American family story consistently fails; content about the audience's own version of the culture consistently wins.
• The proof: our biggest video ever (1.51M combined views) is "American Foods" — five dishes that are Italian-American and not Italian. "Pasquetta" (an Italy-only holiday) sits at the bottom of the catalog.
• Italy content only works when it connects back to the family thread: "What part of Italy do you come from?" did 880K because it is really about the viewer's grandparents' village.
• The test for any idea, image, or clip: does this give an Italian-American viewer something about THEIR life, family, or history? If it is just an interesting thing about Italy, it does not run.

The emotional engine: credit that never came

The audience's core story: our grandparents arrived with nothing, were mocked, worked the worst jobs, helped build America, and never got the credit. That story is the wound and the pride at once, and it is real history (the largest mass lynching in American history was of eleven Italians in New Orleans, 1891). Content that supplies the missing credit is our most shareable content — "Built America" pulled 7,566 Facebook shares on the CTA "Send this to a proud Italian-American", because sharing it is itself an act of giving credit. The working formula: struggle, then triumph, then credit. A subject who was doubted and won anyway beats a subject who was simply successful.

"One of them": which famous people the audience claims

• The audience embraces figures they read as their own: immigrant or working-class roots, made it in America, visibly proud of the heritage. Sinatra, Stallone, DiMaggio, Maniscalco.
• Maniscalco (working-class Chicago comic) is the best celebrity performance of the summer; Isabella Rossellini (born in Italy, European film world) got the same production quality and capped out low. Subject fit matters as much as how well the video is built.
• Heritage is verified before anyone gets called Italian or Italian-American — the audience fact-checks in minutes, and one false claim damages the brand's standing. A thin or unverifiable connection means the subject is out, no matter how good the story.

Defend the audience's identity — never lecture it

• Many viewers have been told they are not really Italian. Content that defends their claim converts everywhere: "If anyone has ever told you you don't look Italian, they are dead wrong" set our YouTube retention record (72%) and gets saved like an argument-settler.
• We never tell the audience their traditions are wrong. "American Foods" worked because it explained WHY the diaspora invented chicken parm (chicken was cheap in America), turning divergence into heritage. The same facts delivered as "your food isn't real Italian food" would insult the exact people watching.
• Direction is everything: correct America's misconceptions about Italians, correct mispronunciations, correct myths. Never correct the audience's family.

Family is the unit of identity

• Being Italian here is not citizenship or language — it is Sunday dinner at 2pm, the grandmother who will not stop feeding people, the second refrigerator in the garage, the plastic bag of plastic bags. When someone sends one of these videos to a relative, the real message is "this is us" — the single strongest share trigger in the comment data.
• Scripts always address the viewer as "you" and "your family", never "we" or "us". A brand saying "we" reads fake; "your grandmother" puts the viewer's actual grandmother on screen.
• Nostalgia objects must be real photographs. When we ran an AI-generated image of vintage bowls, commenters called it out immediately and it undercut the video. Real objects, real photos, always.

Region, the South, and the slang

Italian-Americans identify by region, often by village — Sicilian, Calabrese, Neapolitan, Barese — and these distinctions have stayed alive through four generations. Most of the diaspora traces to Southern Italy, so southern food, customs, and dialect resonate hardest. The slang that performs (gabagool, mal'occhio, agita, mutzadell) is southern dialect filtered through New York and New Jersey over a century — viewers often cannot find these words in a dictionary, which is exactly why decoding them is valuable. Slang videos are our most-saved content. The warning: pronunciations and etymologies get challenged in the comments, so both get verified before anything ships.

The mob line, faith, and respect

• HCI's position is non-negotiable: Italians as inventors, builders, artists, athletes, saints, and neighbors. No mob glorification, ever. The nuance: the food and language mob media made famous (gabagool) is claimed proudly as culture — the words yes, the guns no.
• The audience is heavily Catholic, and folk superstition (mal'occhio, the cornicello, no shoes on the table) sits right alongside the faith. Both are one inheritance and both get a straight face — the superstition is not a joke, and the faith is not a prop.
• Tribute content (deaths, 9/11, historical injustices) and religious content never carries merchandise selling. No exceptions.

Read the comments as the culture in real time

• Viewers prove membership: surname, region, their grandmother's word for something. A thousand people commenting their last name is the audience writing the next video for us.
• Viewers fact-check everything — treat comments as a free audit, not criticism.
• Viewers ask for what they want next by name. The best content pipeline we have is the one the audience requests directly.

The quick judgment reference

• Fits: validates the Italian-American version of the culture; gives credit with a struggle-then-triumph arc; defends the viewer's claim to being Italian; triggers a specific family memory; connects to a region or surname; subject reads as one of their own; real photos of real people and objects.
• Does not fit: Italy trivia with no American family connection; telling the audience their traditions are wrong; flattery with no information; mob glamor; unverified heritage claims; brand voice saying "we/us"; AI images of real people or nostalgia objects; selling merch on tribute or religious content.
• The one-sentence version: every piece of content is judged by whether it makes an Italian-American viewer feel seen, credited, and connected to their family's story.

The format in one sentence: a face-led green-screen talking head — the CEO on camera, real photos and slides behind him, reading a script that appears as burned-in captions — posted every single day across Instagram, Facebook, TikTok, and YouTube Shorts. Protecting the daily streak is goal number one.

Runtime and pacing

• Standard deep video: 50–90 seconds. Past 105 seconds the long tax hits (~10 points of watch percent) — only elite storytelling earns that length.
• In a list video, item one lands on screen by 0:08. Each list item gets 10–18 seconds: a name, a story beat, and an identity beat.
• The visual changes every 10–15 seconds, never a gap over 13. Slide math: a 60-second video needs 6–8 slides after the hook frame; 90 seconds needs 9–12 — at least one slide per list item plus a reveal slide for the teased item.

The first frame

• Full-bleed vertical 9:16, every image, every slide. Letterboxed images — dead black bars top and bottom — are banned: the World Cup video wore them and ran a 62.7% skip rate, because the bars said "recycled TV clip" before the first word landed.
• A text line of about eight words, readable in one glance with sound off, burned into frame one of the video file itself — not just the cover thumbnail. The receipt: Non-Mob Movies carried its title only on the cover, and since Feed and the FYP serve the playing video, in-stream viewers saw wallpaper.
• Text never sits in the bottom quarter (captions and buttons cover it) or the top corners (platform UI).
• Automatic first-frame kills, at any quality: a collage or mood board, an AI-generated or AI-styled real person, a recreation image at 0:00, no text line, or text that spoils the reveal.

The script laws

• Cold open — no greetings, ever. "Happy Columbus Day everybody" is dead weight on every platform (the Republic Day video ran 14% watch). Open on the single most surprising true line.
• The spent-fact law: a fact promoted into the hook is spent — it cannot reopen the list as a reveal. The strongest unspent beat is held for last. And if line one could be the video's title, it's giving too much away: make line one the question the title answers.
• The say-see law: the screen shows what the script claims, within a beat of the claim. The receipt: a video about scarpetta that never showed the scarpetta in the open.
• One CTA per video, at the end, after the payoff — an ad card is three seconds maximum. Every triple-stacked ending on tape underperformed.

The language laws

• Italian-American, hyphenated, always.
• Never say "real Italians" — say "Italians in Italy". The audience's claim to the identity is never questioned.
• No em-dashes in any deliverable — scripts, on-screen text, captions. Periods and commas only.
• Inline pronunciation guides only for words the reader would stumble on — never for celebrity names the audience already knows.

The full closed-book certification test on all of this exists — passing it means you've watched the videos and understand why we make them. Ask the CEO when you're ready to take it.

Use Claude to generate video ideas. The ideas are driven by our MD files — there is a Master Playbook covering the whole process A to Z, plus a Video Idea Bible built just for this step. The CEO analyzes the last 30 days of videos once a month (with your help — see Monthly Analysis) and updates these files with what is working and what is not. Right now, videos speaking directly to Italian-American culture and Italian-American icons perform best.

The monthly cadence you are responsible for

• Two videos a week are built purely for engagement: simple two-sentence poll questions, pulled from the meme database. A good poll is one Italians actually debate ("What time do you eat Sunday dinner?") — never one where everyone knows the answer.
• Kingmaker — twice a month we spotlight someone to help the Italian community grow: one business spotlight and one creator spotlight.
• The Neighborhood Series: one popular Italian-American neighborhood featured each month.
• One to two videos a month let the audience learn more about the CEO personally — that connection is the point, even if the concept alone is not S-tier.
• Sales videos pegged to the marketing calendar — for example "Top 5 gifts for Italians" on Black Friday weekend.

Peg everything to the calendar

Check holidays, birthdays, and anything tied to Italian or Italian-American culture using the Sourcing & Dates sheet — All Saints Day in Italy, Frank Sinatra's birthday, National Spaghetti Day. We should never miss an important event, because those trend on social media, and videos that ride what is trending do better. Cross-check every name and date: Italian-Americans typically perform best, and a name famous only in Italy usually will not land. When you discover a new date, add it to the Sourcing & Dates sheet so we have it for future years.

Stay about three months ahead — no more. Going six months or a year out means planning without the data the next few months of videos will teach us.

Two chats, two jobs — and a duplicate check

Use one Claude chat to plan the week or month (upload the Master Playbook and Video Idea Bible, use high effort, and let it take its time). Then use a second, separate chat whose only job is to pass or fail each individual idea. The planner is juggling cadences and pegs; the rater scrutinizes one video at a time. Every idea must come out A-tier or S-tier — regenerate anything that does not. Before an idea goes on the calendar, search the content calendar to confirm it has not already been done; if it has, check how it performed, because sometimes only the hook needs remaking (a rebuild), and sometimes it should stay retired.

Approval

When the month is planned, notify the CEO. Each approved idea gets a check mark on the calendar. Only confirmed ideas move on to scripting — an idea without a check mark is not ready, no matter how good its tier.

Done means

• The next ~3 months of the calendar are filled, every idea rated A or S in its own rating chat.
• Every idea has a peg or a cadence reason (engagement, Kingmaker, Neighborhood, CEO personal, sales).
• Every idea passed the duplicate check.
• Every idea the CEO approved shows a check mark; unapproved ideas are clearly pending.

Set up the files

• Create a folder in the Green Screens master folder in Google Drive, titled with the video idea.
• Create a Google Doc inside it with the same title.
• In Claude, enter the prompt from the AI tracker and upload the Master Playbook MD and the Script VA file. Generate the script.
• Read it through. It must sound natural and consistent with the approved CEO voice — the voice MD file exists for exactly this. It should never read like an AI script.

CTA rules

• One call to action (CTA) per video, a few seconds long, at the end, and it must relate to the topic.
• Engagement CTAs: follow for more, comment below, share with a friend.
• Once or twice a week a video gets a shopping CTA instead — "comment MERCH to shop" or a sale mention.
• Be sensitive: a video about a late celebrity is never the place to push t-shirt sales.

Fact-checking — sources, not AI

AI is a helper for fact-checking, never the final word. Every factual claim in the script (dates, places, names, quotes, numbers, "first to", "invented by") must be confirmed against an authoritative source you can link to, and the link goes in the doc.

• First pass: paste the script into Google Gemini (a different AI with no context, forced to verify everything) and ask it to flag anything doubtful. Treat its output as a list of things to go check — not as confirmation.
• Then verify each claim yourself against an authoritative source: official biographies and obituaries, encyclopedias (Britannica, Treccani), museum/library/archive sites, .gov/.edu pages, established newspapers and wire services, the subject's own official site or verified accounts. A Wikipedia article is a starting point — follow its citation to the real source and link that.
• Not acceptable as the only source: another AI, a content farm, a fan page, a meme, a TikTok/YouTube video, a listicle, a forum, or a source that is itself citing nothing.
• Record every source: below the script in the Google Doc, add a "Sources" section listing each claim and its link. Mark anything still unconfirmed.
• A claim you cannot confirm gets cut or rephrased into something you can confirm. A claim the audience has disputed before is a fact flag — it must be resolved before that subject is used again.
• Names and heritage: confirm the person is actually Italian or Italian-American (heritage verification) with a source, and on recording day confirm a living subject is still living (liveness check).

Pronunciation — the number one issue on the page

The audience notices, comments, and mocks bad pronunciation — a single mispronounced word on a million-view video means a million people heard it, and some words ("A-rabs") carry history that makes the mistake genuinely offensive. The goal is zero pronunciation mistakes on any video. What gets verified, no matter what: every person's name, every business name, every town name, every tough Italian word, and every double-meaning English word (bass the fish vs bass the instrument, Greenwich). Mark each one in the script with an easy phonetic spelling — "Gnocchi (NYOH-kee)" — and a reference link in the Sources section.

The verification tiers — Google is the LAST resort

• Tier 1 (the 10/10): the person saying their own name, on video. "Hey everybody, it's Martin Scorsese, here with GQ this week." Search YouTube for the name — intros like this exist for almost everyone.
• Tier 2: the person introduced or interviewed live — Fallon, a podcast, the news, an awards show. If the host says the name to their face and doesn't get corrected, that's the pronunciation. (DiCaprio introducing Scorsese — his close friend — counts precisely because he'd be corrected if wrong.)
• Tier 3: an ordinary human on YouTube saying the name naturally in their own video — a fan talking about their favorite films. Real voice, no AI.
• Tier 4, last resort only: Google, AI voices, and "how to pronounce" videos. Most of those are AI narration pulling from Google and cannot be trusted — this is where wrong pronunciations come from.
• Find multiple sources and hear the name for yourself. One video is a lead; two agreeing videos is verification.

The Italian vs Italian-American trap

There is the Italian pronunciation and there is the way the actual person says it — and the person wins. Scorsese pronounced the pure Italian way (E's said as A) is technically correct Italian, but Martin Scorsese doesn't go by it; correcting the video to the Italian pronunciation would have made the page look like it doesn't know his name. The rule: anything directly in Italy — towns, dishes, Italian words — always gets the Italian pronunciation. Italian-American people and businesses get pronounced however THEY say it (Giada says it the Italian way, so we do; Scorsese says it the American way, so we do). Finding how they say it is the whole point of the tiers.

The running mispronunciation list

• Keep a running list of every word commenters say was mispronounced — go through every old video's comments and log them all (the list so far: Paolo, Arab, Bass, Greenwich). This is a standing assignment, not a one-time sweep.
• Every logged word gets fed to the AI during scripting so it's flagged automatically the next time it appears, and the commonalities (double-meaning words, silent letters) eventually go into the master script document so scripts flag their own hard words.
• Before any video posts, it gets a pronunciation pass against the script's phonetic guides and the running list — a miss caught before posting costs nothing; a miss caught by the comments costs the video.

The two Claude checks — each in its own separate chat

• Increase engagement: a separate Claude chat with the engagement prompt cross-checks the script to increase video engagement.
• Hook check: a new Claude chat with the Hook Bible MD analyzes, improves, and rates the hook. It must be a 9/10 or 10/10 — the hook is what earns the video its reach.

The Double Check column — the last gate

• The content calendar carries a Double Check column: after the video is edited — not before — someone runs the final pass on spelling, captions, pronunciation against the phonetic guides, and facts, then checks it off with their name. The column exists so everyone can see who checked what, and nothing posts unchecked.
• Videos with a CTA are flagged on the calendar (a CTA column or note). For every flagged video: confirm the ManyChat flow works before it posts, and coordinate with the Social Media Operations Specialist to pin a comment repeating the CTA ("comment TICKETS to play bocce") once it's up.
• ManyChat technically broken — a keyword not answering? Tell the Executive Assistant right away: she contacts the contractor who maintains our flows. Don't sit on it; a dead CTA is money walking away.

Hand-off and approval

Upload the final script into the Google Doc with the script as the very first page — The CEO should see it the moment the doc opens, with pronunciation guide, delivery notes, and the Sources section below. Paste the doc link into the content calendar as a chip. The CEO approves the script; once approved, move on to visuals.

Done means

• Script reads in the CEO voice, 50–75 seconds spoken, one CTA.
• Every factual claim has a linked authoritative source in the Sources section; nothing unconfirmed remains in the script.
• Every hard word and name has a phonetic guide with a reference.
• Hook rated 9 or 10 in its own chat. Doc linked on the calendar.

These are green-screen videos — the CEO speaks in front of images of whatever is being discussed. Two elements matter: the hook image and the regular images. Every image must be something we are allowed to use, and you record where it came from.

Hook image

It has to grab attention fast without clutter. There is an MD file describing exactly what it should look like — videos about people usually get a full-person background, while some topics use an Italian flag background with a bubble. Follow the file. Say-see law: the hook image must show what the spoken hook promises.

Regular images

• Every paragraph of the script gets an image proving what is said. Talking about cannoli invented in Sicily? Show a cannoli in Sicily.
• A new image roughly every 10 seconds keeps the pacing good — split long paragraphs across two images. No stretch longer than ~13–15 seconds without a change.
• Leave natural blank space in every image — The CEO's head sits on top of it.
• Never use a collage or grid that spoils a reveal before the script gets there.

What you are allowed to use

• Safe: public-domain images (U.S. government works, works whose copyright has expired, images marked CC0 / Public Domain on Wikimedia Commons, the Library of Congress, museum open-access collections); Creative Commons images when you follow the license (record the attribution and the license — avoid "NC" licenses for anything with a shopping CTA); photos the brand owns or has licensed; images we generated ourselves.
• Use with care — organic only, never behind paid spend: movie posters, film/TV stills, album covers, news photography, and any image with a visible watermark or a stock agency's name (Getty, AP, Reuters, Shutterstock). A video built on these can run organically but can never be boosted or turned into an ad. Flag these videos as ORGANIC-ONLY on the calendar.
• Never: anything downloaded from a stock site without a license; paparazzi/agency photos of living people; images pulled from a private or personal account; another creator's original artwork or illustrations.
• "Found it on Google" is a search method, not a license. Click through to the actual host page, find the photo credit and license, and save that link.

AI generation and reconstruction

• Real images first, always. If the video is about a real person — Hulk Hogan, Tony Bennett, Frank Vincent — use only real photographs. Never generate, recreate, or "enhance" a real person's likeness, living or dead, and never stage a real event that did not happen that way.
• Generation is for scenes, objects, and settings where no photograph can exist — a Roman street 2,000 years ago, a concept illustration, a composite ending — and it should look realistic, not obviously AI. For ancient figures, prefer real photos of statues, busts, paintings, and coins before a generated scene.
• Never generate a brand logo, a trademarked character, or a copyrighted artwork in a new form.
• Mark every generated file in its name (see below) so editors and the CEO always know what is real.

File specs, naming, and the source log

• Export every image to the size and aspect ratio in the Image MD file (vertical, full-frame for 9:16 video), high resolution, as JPEG or PNG — no screenshots with browser chrome, no tiny thumbnails upscaled.
• Name files in play order with the source type: 0_hook_real, 1_augustus_real, 2_forum_gen, and so on. Two versions of the same moment get A and B suffixes so the CEO can pick.
• Keep a Sources sheet (or a Sources tab in the script doc) with one row per image: file name, source page URL, photographer/credit, license, and whether it is organic-only.
• Upload everything to the video's Drive folder in order. The CEO approves the set before it moves into the final image folder.

The non-negotiable: every image is TikTok size

1080 × 1920 pixels, 9:16 vertical, full-screen. Every single image, no exceptions. An image delivered in any other size or aspect ratio is completely unusable — it cannot go behind a green-screen video, and the work has to be redone. This is the single most common way an image set gets rejected. Before anything uploads to Drive, confirm every file is 1080 × 1920.

Where to start

• On the content calendar, find rows where the script is done but the Photos box is unchecked — scripts come first, images can't start without one. Work the soonest videos first.
• Prompt Library → Green Screens → Plan Images. Paste the prompt into Claude with the script, and attach the image MD files (what works and what doesn't, plus the hook guide) so the chat knows the rules.
• The output comes back paragraph by paragraph: what each image should show, and whether the hook is Style A (no flag frame) or Style B (the tricolor flag frame with the photo in a bubble).
• Read it with judgment. AI will happily describe a photo that doesn't exist ("Louis Prima at 14 in a restaurant doing a hand gesture"). Go back to the actual script line and find the closest real photo that matches what's being said — "you've heard his voice" wants him singing, not holding a trumpet.

Real photos first — and check them

• Real images for real people, always. Search the name plus the era the plan calls for ("late 1950s"), and look past the first result — two more seconds of searching regularly finds a much better photo.
• Verify it's actually the right person. Skip anything with a stock-agency watermark (Alamy, Getty) — it can't be used. Watch for AI-generated fakes in search results; they're everywhere now.
• Remove stray word marks or branding on the photo; lightly correct tone, contrast, and color. For posters or album covers, make sure it's the version with our subject on it.
• Give the AI options: drag two or three candidates into the chat and ask which fits 1A vs 1B. Save good photos you stumble on even before you need them — the plan usually wants a "young" shot later.
• Before you decide on a photo, remember the say-see law and the rights rules in the task above: the image must show what the script claims, with a recorded source.

Sizing for TikTok — which prompt, when

• Hook Image prompt: for Style B hooks, drag the real photo in and it builds the flag frame with the bubble. If it comes back with a sky or a wrong format, re-prompt until it matches the standard.
• Resize / Expand prompt: only when the photo has background the AI can extend naturally — sky, wall, floor, scenery. It must never invent legs, limbs, furniture, or picture frames; if it does, reject it. Check the result: the subject must be the same real photo, untouched.
• Photo Enhancement + Blurred Background prompt (or the Photoshop version): when there is nothing safe to extend — a tight portrait, a cropped shot, an album cover. Duplicate the photo, blow the copy up behind it, strong Gaussian blur (blurred enough that nothing is recognizable, or it distracts), drop shadow on top and bottom edges, and the sharp photo at FULL canvas width — edge to edge, no side margins, never a floating card — positioned upper-middle with the blurred space above and below. This is the most reliable method, and it works even when source quality isn't perfect.
• New Image prompt: only for scenes and endings where no real photo can exist. Image Fusion: the mashup of everyone in the video, typically for the ending — but judge it honestly against a real alternative (a jukebox beat a forced composite).
• Run several chats at once — resize in one, enhance in another — and keep working while they render.

Layout rules

• Every image is full-screen 9:16 (1080×1920). Keep the subject a little high: TikTok and Instagram UI covers the top (page title) and more of the bottom (caption, likes) plus the right side.
• Leave a natural open area — roughly the lower 25–35% — where the CEO's talking head goes. It should never look like a placeholder box; just low-detail background.
• Picture the video as you go: hook image with the CEO over it, then paragraph one with the image switching 1A → 1B → 1C as the lines land. If the image can't make that work, it isn't done.
• Video can also sit behind the talking head — a clip of people dancing to the song works where a photo wouldn't.

Naming, filing, and sign-off

• Name in play order with a descriptor: 0_hook, then per paragraph with a letter per image swap — 1A_prima_stage, 1B_album, 1C_young, 2A_dion… The number is the paragraph, the letter is the switch, so recording knows exactly when to change the image. Add _real or _gen.
• Build in a local folder, then upload the full set to the video's folder in Drive → Images, in order. Keep a second version (A/B) where you couldn't decide.
• Do not check the Photos box on the calendar yourself. Post that the set is ready; the CEO reviews and checks it off once approved.

Working tips from the examples tape

• Make an images folder inside each video's Drive folder — transcripts and data will live there too, so keep the images separated.
• Claude over ChatGPT for the planning chat when available. And run two or three chats at once — one planning, others resizing — so you keep working while they render.
• Blurry result in search? Right-click → Open Image in New Tab — it often loads sharper. Skip images pulled from video frames; they never get sharp enough. For Instagram photos, use a "download Instagram photo" site to grab the full-quality file instead of screenshotting.
• Pick the most iconic shot for the first image of a person, and keep it classy — we shed positive light on people. Zoom-crop when subjects sit too far back before resizing.
• Give AI a shot on every image and judge each result: when the prompt nails it (sharpened, nothing invented) use it; when it adds legs or furniture, fall back to the natural photo or the blur method. If the enhanced version changed the person at all, it doesn't ship.
• When the plan says to add text to an image — skip it. Text gets added later in the pipeline, always.
• For a Part 2 or Part 3 video, open the earlier parts first: the image style that worked is the template.

Easy question videos

• The calendar marks them (italicized, "easy question video") — they need one image, no script, no naming conventions. A batch of three takes less time than one deep story video, so they balance the workload.
• The New Image prompt is fine here — no real subject is involved. Check the result reads Italian-AMERICAN, not Italy (flags, setting).
• Brand assets are fair game: the logo on Drive, the original skull logo — ask AI for ideas on making them into something cool. A meme from the page can be the image when the question is based on one.
• Always deliver two or three options so the CEO picks the best — optionality beats four rounds of back-and-forth.

Done means

• Every paragraph has its images, all full 9:16, all real where a real person or object is involved.
• Each file named in play order with _real/_gen; sources logged.
• Talking-head space and UI-safe margins respected on every image.
• Uploaded in order to the video's Drive folder and flagged for review.

The objective is to protect the CEO's image and the brand's image at all times — not to win arguments, not to eliminate criticism, not to make the comment section perfect. Not every negative comment gets removed: disagreement, mild criticism, jokes at our expense, and "this isn't for me" all stay up. A comment section with zero pushback looks fake.

Sweep timing, per video

• 1 hour after posting: first pass — early comments set the tone for the rest.
• 4–6 hours: second pass — this is usually when volume peaks.
• 24 hours: third pass, plus the sentiment read for your notes.
• 48–72 hours: final check on anything still gaining traction.

The cleanup bar

• Clean up when: a nasty comment is gaining traction (high likes, sitting near the top, being amplified in replies); roughly 10% or more of a video's comments are negative (the section is turning — sweep it); the comment includes slurs, hate, harassment, sexual content, or attacks on family; it is spam, a scam, a bot, or an impersonation account; or it attacks the CEO personally rather than the content.
• If you cannot remove it yourself, send the link plus a screenshot to the Social Media Operations Specialist and ask for removal. Do not sit on it.
• Never argue and never reply to a hostile comment — remove, hide, or ignore. A brand account arguing in the comments is worse than the original comment. Never respond emotionally, and never defend the CEO in the comments; that is what moderation is for.

Always do this

• Like positive comments — every video, every sweep. Reply to real people (questions, compliments, "where do I buy this") and sound like a person, not a brand account.
• Pin a strong comment when there is one worth putting at the top (send the video link and exact copy to the Social Media Operations Specialist if you lack permission).
• Keep a running file of video ideas from the comments — especially highly repeated or highly liked ones. A lot of our best ideas come straight from the community. Log the idea, the video it came from, and roughly how many people asked.
• Audience corrections are fact flags: if several people dispute a fact, log it against that subject so it is resolved before the subject is used again, and tell the Social Media Operations Specialist in case the caption needs fixing.
• Mispronunciation comments: log the word on the running list, then remove the comment or send it to the Social Media Operations Specialist for removal. Any mispronunciation comment with 25+ likes gets flagged to the Social Media Operations Specialist to delete immediately — a top comment mocking the pronunciation multiplies the damage on a high-view video.
• After trust is built you may get access to the CEO's personal Instagram to reply to positive comments in the CEO's voice. Responding drives engagement and makes the audience feel connected.
• Monitor direct messages for partnership opportunities; forward anything real to the Social Media Operations Specialist the same day.
• Check that captions, descriptions, and tags are accurate. Fix small things yourself if you have access; otherwise notify the Social Media Operations Specialist.

Log the pattern, not the comment

While you sweep, keep a running note per video: rough positive vs negative split, recurring themes, pronunciation or word choice that keeps getting called out, and which hooks or topics drove the best response. That note is the raw material for the weekly CEO summary — nothing else from the comment section travels upward. Own the whole process: it is not about just getting scripts made, it is about the videos performing well, having no errors, and taking accountability for the content being worth the time and effort going into it.

The general MERCH funnel

• Many videos end with a generic "comment MERCH" CTA. Those comments hit the always-on general funnel, which runs on both Facebook and Instagram — make sure it is always live on both.
• Analyze it like a product: how many sends, how many clicks, what click-through rate, how many emails captured. Put standards in place and check against them — if it underperforms, A/B test the copy and improve it.

Video-specific CTA flows

• Any video with its own CTA ("comment SLANG", "comment NYC") needs its own flow. Know every CTA that is coming — they are flagged on the content calendar and in Slack — and make sure each one is built, tagged correctly, and working before the video posts.
• Never start from scratch: duplicate an existing flow so the setup carries over, then change the keyword, copy, and link.
• Trigger options are a specific post, any post, or the next post. Coordinate timing with the Social Media Operations Specialist: if it's the next post, set it live on "next post"; if it posts at 3:00, go in at 3:01, attach the specific post, and you're done.
• Always turn on replies to comments. And make the DM match what the video actually promised — if the video said 15% off, the DM says 15% off; if it just said "we have a New York line", don't invent a discount.
• Check the goal per flow: emails, sales, or followers — for us it's usually sales (with email capture in the flow). Turn off leftover settings from the duplicated flow that don't apply (like a follow requirement from an old campaign).
• The link must go exactly where the CTA promised — the product, the collection, or the landing page, never just the homepage. New York CTA → the New York collection; Black Friday → the Black Friday page. If the right collection doesn't exist yet, set it up (or request it) before going live.
• Keep the copy on-brand: Ciao, Grazie, the Italian flavor. Run the copy through AI for best practices.
• Flows can be built a month in advance and left saved — then attach the post and Go Live on the day.

Platforms ManyChat can't reach

• CTAs currently work only on Facebook and Instagram — not YouTube or TikTok. For those platforms, either the CTA gets removed from that platform's upload, or you go comment back manually ("link in bio") on every CTA comment.
• The one unbreakable rule: nobody comments a CTA and gets silence. Anyone interested in merchandise receives a response. Check every shopping-CTA post daily while it's live; if a flow didn't fire, report it and reply manually in the meantime. Note counts and failures in your EOD.

Spend real time getting familiar with ManyChat — it has a deep interface, plenty of YouTube tutorials, and people on the team who can troubleshoot. If something is broken, raise it; don't let a CTA run unanswered.

Every video gets a full 30 days of data before it is analyzed — so on September 1st you analyze all of July, on October 1st all of August, and so on. The output is one deep-dive MD file per video, and those deep dives are what update the playbooks and make next month's ideas better.

Collect the data, platform by platform

• Facebook + Instagram: Meta Business Suite → Insights → Content, filtered to the month you are analyzing, segmented by video. Toggle Facebook only first, screenshot everything (zoom to ~90% to fit more per shot; grab audience retention both on and off), then switch to Instagram and repeat. Also screenshot the comments on the actual posts — the AI needs a feel for how people reacted.
• Combine each platform's screenshots into a PDF (select all → Quick Actions → Create PDF on Mac) named Facebook, Instagram, or Meta for the combined totals.
• TikTok: TikTok Studio → Posts → search the video → View Analytics. Screenshot every tab — overview, viewers, engagement (even if engagement is empty, capture it). Create a PDF named TikTok.
• YouTube: YouTube Studio → Content → find the Short → Analytics. Lots of screens here — zoom out and get every data point. PDF named YouTube Shorts. (The channel isn't monetized yet; when it is, monetization screenshots get added.)
• Upload all the PDFs into the video's "data" folder in Drive.
• Tip: batch by platform — grab all ~30 videos' Facebook data in one sitting, then all Instagram, then TikTok, then YouTube. You can narrow the date filter day by day since the calendar is organized by date.

Analyze each video with Claude

• Open the Prompt Library → Green Screens → "Analyze Performance" and copy the prompt.
• In Claude (best model available), paste the prompt and attach everything: all the PDFs, the screenshots, the transcript — and the video file itself. Claude goes frame by frame.
• It grades the idea, script, hook, retention, per-platform performance, comment sentiment, shares, saves, and likes. This takes a while — that is good. If it finished in two seconds it didn't crawl the data.
• Run these on the side while you do other work (reel finding, comments, scripts). Be patient with the analysis, be in a hurry with everything else.
• Save the deep-dive MD it outputs into the video's folder in Drive, then check the video off on the calendar.

Done means

• Every video from the target month has all platform PDFs in its data folder.
• Every video has its deep-dive MD in Drive and a check on the calendar — the month reads 100% analyzed.
• It's the 1st (or the first two workdays): September 1st = July done; October 1st = August done.

• Open the recording in the Edits app and generate captions right off the bat. The captions aren't for the final video yet — they're a visual map of when speech happens, so dead space shows up as gaps you can physically see.
• Go through the timeline without even watching: find each gap between spoken clips and drag the clip edge to close it. Usually the second (newer) clip carries the most dead space, so grab that one and shorten it.
• Don't cut words off the beginning. If you trim too tight you'll see the caption layer shrink — that means you clipped a word. Push right up to that limit and no further.
• While you're in there, watch for repeats: if the captions show the same phrase twice ("her great-grandfather… her great-grandfather"), listen to both, keep the better take, delete the other.
• Goal after this pass: the lines between spoken clips should all be very small.

• Play it back and listen for anything that isn't clean — a fifth-of-a-second stutter, a pause where a line was being read, a hesitation before a hard Italian word.
• Split the timeline at the stutter and bump the clip so it flows like the pause never happened. Hesitations before difficult words are the most common fix.
• Fumbled the middle of a sentence? Keep the good opening, cut the fumbled part, and drop in the reset take.
• Cut the tails too — the video should jump from the end of one word straight into the next. Most editors under-cut the ends.
• Come back to it with fresh eyes if you can; a second look the next day catches what the first missed.

When you've been stopping and editing, stopping and editing, you lose the feel of the full video. Do one complete run-through without touching anything. If it plays clean, the video is good and you move on to captions.

• Click the captions → Delete all captions. Then save — this is the clean, fully edited copy with no captions that we keep in Drive forever. Make sure it says it's downloading the clean copy.
• Now generate captions again. Known glitch: Edits generates three lines but the screen only fits two. Exit the project and re-enter — that fixes it. Verify it did.
• Open the script (Edit script) and proofread the whole thing. Names are the big one: "Giada" will come out as "Jada" — if the name is wrong on every caption, it's bad. Fix it.
• Hunt for anything AI won't hear cleanly: Italian words, foreign names (Le Cordon Bleu), proper nouns (Wolfgang Puck). Check the beginning and the end — errors hide at both.
• Once the script is clean, publish and save the captioned version.

The pipeline exists because three failures hit in one day: a video recorded for one date posted six days early from a swapped folder, a delivery came back missing captions, and a batch came back with broken audio. One root cause — nothing between "editor uploads" and "posting" verified that the right, finished file was in the right place. Posts are deliberately sequenced, so a wrong post doesn't just look sloppy, it burns a planned moment.

The delivery checklist — every video

• Correct file in the correct date folder, with the content calendar link verified.
• Captions on — the standard for ~90% of videos. Skipping captions is a deliberate choice that gets said out loud in the channel, never assumed.
• Audio checked start to finish, levels balanced across speakers.
• Final filename matches the calendar entry.
• If a video moves folders for any reason, the calendar links get fixed in the same sitting and a note goes in the social channel.

What happens after you deliver

• The Social Media Operations Specialist spot-checks deliveries the day they land — not on posting day — and flags problems back to you same day, so there is time to fix before the post date.
• Before posting each day, the video goes in the group chat for a same-day confirmation ("Posting this today, confirm"). Ten seconds of friction, zero wrong posts.
• Revision etiquette: feedback arrives as one consolidated pass per batch, bulleted per video. Apply the full pass and redeliver — no drip-feeding single fixes. If a note is unclear, ask before editing, not after.

• Best answer first. Lead every video with the strongest or funniest person. When testing someone new up front, note it — we are building a roster of who hooks viewers in the first 3 seconds.
• Cut back and forth between people. Alternating speakers makes thin footage feel full — two answers cut as four beats plays like a complete video.
• Cut the filler. Trim host acknowledgments ("appreciate that") but keep the laugh. Jump straight into the funny part of an answer; cut wind-up lines.
• Zoom on the speaker. When someone other than the CEO lands a line, the cut zooms to them.
• Cut weak segments entirely. If a clip is too short or flat to earn its screen time, it goes. No mercy inclusions.
• Balance audio across speakers — no one noticeably quieter.
• Overlays support, never cover. Pop-up imagery appears only after the person says the name, overlaid partially, never covering the full screen.
• One version per video for now — the two-version A/B test idea is parked.

• From your phone, go to Drive → Media → Green Screen Reels. Every video already has its own pre-made folder — find it.
• Upload two files: the clean no-caption edit and the captioned edit.
• Rename them clearly, with "Captions" in the captioned one: "Five Facts About Giada" and "Five Facts About Giada Captions". Open each to confirm which is which before naming. This is how the social media manager knows what he's grabbing.
• In Slack → Social Media, open the Green Screens canvas. Find the video on the planned list and paste the Drive folder link over it.
• Open the Content Calendar, find the video's row, and check it off. The row highlights — that's the signal it's 100% done and good to go.

Sourcing is done from a dedicated account, "Sunday Sauce Szn" — it exists only for this, so its entire algorithm is trained to feed us Italian content. Get the login from your manager (it lives in LastPass); if Instagram or TikTok asks to verify the profile on your first sign-in, flag it and it will be taken care of. Once in, stay logged in. Sourcing happens primarily on TikTok and Instagram. Bookmark and engage with good Italian content and hit "not interested" on everything unrelated, so the For You page does the heavy lifting for you.

Three ways to find content

• Your feed: once the algorithm is tuned, good content comes to you.
• Competitor UGC pages (we provide a list): look at their best-performing videos, then go find the original creator and the original video — never credit the repost page.
• The Discovery / Explore page: search "Italian" and related terms, engage, and let the algorithm learn.

For every video you flag

• Duplicate check first: search the Reel Finder for the creator's handle and the video link. If the video is already logged, skip it; if the creator is already in the pipeline, add the new video to their existing row notes instead of a new outreach.
• Copy the link into the Reel Finder sheet with: link, creator handle, platform, date found, content type, video views, creator follower count, Messaged = No, Permission = No, Posted = Not Yet.
• Pre-check before anything goes on the list — do not waste outreach on content we cannot legally use: no licensing-agency watermarks (ViralHog, BVIRAL, Jukin, and similar — if you see one, skip it, or flag it to the Social Media Operations Specialist if unsure); confirm the account is the original creator, not a reposter; skip heavy third-party music or licensed film/TV clips that would get the repost muted or flagged; skip accounts that appear to be under 18; and skip anything against brand guardrails — mafia glorification, corny Italian stereotypes, anything that would embarrass the brand.
• Look at the creator's page — if they make Italian content regularly, follow them from the sourcing account.
• Share the post to the Hardcore Italians page so there is a direct link for the team.

Minimum ten per morning — twenty is better. We need a video to post every single day, and only a fraction of outreach ever comes back.

You do not send the DMs — the Social Media Operations Specialist does, from the brand account. Your job is to make it effortless: for each creator, prepare a numbered message block with the creator's name, a link to their page, a link to the video, and our standard reach-out message from the outreach document. All the manager should have to do is click and send.

The daily pack

• Keep finding and logging 10–20 a day, but SEND only the best five: five reels a day, DMed to the Social Media Operations Specialist, one numbered block per creator in priority order. Weekends flex — five Saturday and five Sunday, or the weekend's batch on Friday; he posts weekends too. The daily count may grow later, but never overwhelm him.
• Mark each sent creator with the "Sent to Social Media Operations Specialist" status on the Reel Finder, so the sheet always shows what's been handed off. Confirm the pack was seen — a pack that is never opened means nobody gets DMed, and then finding 80 great creators was worthless.
• When the manager confirms a DM went out, mark Messaged = Yes with the date. If a creator was skipped, note why.

Tracking to a result

• Check replies yourself in the Meta Business Suite inbox (bookmark it) — never ping the Social Media Operations Specialist "did they reply?" over and over. You can see every conversation and its context there; tag creator relationships (positive with the brand, allows posting) as you go. Only the initial DM depends on him, because it has to come from the brand page.
• Permission is legal proof: the DM uses ownership language ("do you own this video?"), and screenshots of the permission conversations get saved to the Proof We Can Post folder in Drive — check whether the Social Media Operations Specialist is already filing these and assist.
• Creator says yes → send them the JotForm link (via the manager if the conversation is in the brand inbox) and mark Form Sent.
• No reply after 7 days → one follow-up message goes in the next day's pack, marked FOLLOW-UP. No reply after the follow-up → mark No Response and stop; do not message a third time.
• Creator declines → mark Declined and do not contact again about that video. Report approvals and denials to the Social Media Operations Specialist at end of day, every working day — even if the answer is "nothing new".
• Creator asks something you cannot answer — payment, exclusivity, usage terms, a takedown request — do not improvise an answer. Pass it to the Social Media Operations Specialist.
• Every row always has a current status. Update the sheet the same day anything changes — it must reflect reality.

Done means

• Every creator found today is in a numbered, click-ready pack, delivered and acknowledged.
• Every open outreach row has a status and a last-touched date.
• Follow-ups due today are in today's pack.
• A weekly recap goes to the Social Media Operations Specialist every Friday: what was sent, who approved, who denied, and recommendations ("these five never responded — drop them?"). He's your main point of contact on reels.

When a creator agrees, they fill out the JotForm: their info, the link, and a signature giving us permission to repost. Each submission shows up three places — an email notification, the JotForm submissions area, and the form responses tab of the Reel Finder.

Validate the submission

• Complete and signed: name, handle, the exact video link, and a signature. If anything is missing or the link does not match the video we asked about, reply (through the manager) asking them to resubmit — do not mark Permission = Yes until it is complete.
• The person submitting must be the creator of the video (handle matches the account that posted it). A third party cannot grant permission for someone else's video. Ask the exclusivity question directly and record the answer. No verbal or DM-only approvals — the signed form is the only thing that counts.
• Exceptions to escalate to the Social Media Operations Specialist before approving: the video contains other people's copyrighted material (a TV clip, a full song), the creator asks for payment or conditions, the creator appears to be a minor, or the content has changed since we found it.

Your pipeline, step by step

• Valid submission → mark Permission = Yes (with date) and Posted = Not Yet.
• Notify the Social Media Operations Specialist: send the approved post's link and the creator's handle for credit, and agree on a posting date.
• Once a date is confirmed, mark Scheduled with the date.
• Approved content should have a post date within 48 hours of approval. If it does not — or if a video with rights never goes up — chase the Social Media Operations Specialist directly until it is resolved. Two statuses should never sit still: "Approved, no date" and "Approved, never posted". Both are yours to chase.
• Check our Instagram (and the other platforms it was planned for); when the video is actually up, confirm the tag, the credit, and the caption are all correct, then mark Posted with the date and the link to our post. A repost live with the wrong caption, wrong crop, or missing credit gets flagged immediately, before it gains traction.
• The pipeline is updated the same day anything changes, and reviewed top to bottom every Friday — it must always reflect reality.

Done means

• No row shows Posted without Permission = Yes and a form on file.
• Every Permission = Yes row is either Scheduled with a date or Posted with a link.
• Friday review complete; no stale rows.

Where to look

• In Ads Manager, skip the campaign level — go straight to the Ads tab and sort by amount spent over the last 7 days. Highest spend = most eyeballs = top priority.
• Filter to active ads. Dynamic ads (no picture preview) can be skipped — they do not collect comments.
• On each post, check all comment filters: "all comments" (Facebook auto-hides things wrongly — someone saying "I placed an order" can get flagged), "newest", and "most relevant" (those show first to shoppers).
• Do the top spenders every day in the first hour — anything around $100+ spent in the last 7 days. Lower-spend ads rarely have comments; sweep those weekly.
• You are in there to manage COMMENTS, nothing else. Never turn an ad off, change a budget, or touch any ad setting — be deliberately cautious in Ads Manager; disrupting a live ad costs real money.

The moderation guide — one action per comment

• RESPOND — positive about the product ("I love this", "I need this shirt"): like it and reply to the best ones. Likes push those comments to the top of "most relevant", so a shopper sees love first. Someone says they ordered: always reply with positive encouragement — it shows people are buying.
• LEAVE — funny, tagging friends, sharing a story, harmless off-topic chatter: leave it. That engagement is exactly what we want. Also leave honest neutral questions ("does this run small?") and answer them — that is a sale waiting to happen.
• HIDE — product suggestions (kids sizes, long sleeve, crew neck) and sizing/price complaints ("$32 for a shirt?"): hide (the commenter and their friends still see it; nobody else does). Keep a tally of suggestions — one or two mean nothing; a hundred is worth telling the CEO. Also hide: order or delivery problems (then ESCALATE to customer service), and purist put-downs of the product ("real Italians don't say gabagool") — see voice note below.
• DELETE — spam, scams, links to other stores, repeated copy-paste comments, hate speech, slurs, harassment of other commenters, sexual content, and anything clearly posted to damage the brand. Deleting is for abuse and junk, not for disagreement.
• ESCALATE — order/delivery/refund issues go to the customer service Slack channel with the order detail so CS reaches out privately (hide the comment first). Threats, legal claims, press or influencer inquiries, accusations of stolen designs, or anything you are unsure about go to the Social Media Operations Specialist the same day. When in doubt, hide and ask — never argue in the thread.

Voice and culture

Replies are lighthearted: Italian flags, emojis, "Ciao", "Grazie". And know the brand: we are Italian-American. Gabagool, gravy, Sunday dinner — that is our culture. A purist commenting "real Italians don't say that" is talking down the product, so it gets hidden; a genuinely hateful version gets deleted. If you are unsure whether a cultural comment is a joke or a dig, ask in the Slack customer service channel.

Instagram ad comments don't offer the same management tools, since you're not logged into the page. A genuinely bad comment on an Instagram ad: DM the post straight to the Hardcore Italians page with a note ("bad comment on this ad — please delete") so the Social Media Operations Specialist handles it. If he says he can't see it, Instagram may have auto-hidden it — have him double-check. Regular Instagram post comments work the same way: flag once a day; monitoring them is his role, flagging is yours.

When the CEO is away: the audit sheet

Any stretch where your reviewer is unavailable, keep a one-sheet running log of every action: comments deleted and why, likes and replies given, invites sent, ManyChat checks, anything escalated. When he's back, he reviews your judgment on one document instead of digging through every ad — and either says "good to go, keep it up" or corrects course. Deliver it unprompted.

Done means

• Every active ad around $100+ spend reviewed today on all three comment filters.
• Every comment touched has one of the five actions; order issues are in the CS channel.
• Suggestion tally updated; anything escalated is noted in the EOD.

People like our posts every day without ever following the page. Logged in as the Hardcore Italians page, open the recent posts, click the likes, and hit Invite on as many people as it allows (Facebook caps it daily).

Do it once a day on that day's posts: someone who liked a post this morning and gets an invite the same afternoon still has us top of mind, and they join. Do the same on high-spend ad posts while you are moderating comments — it all grows the page. Volume guide: a few hundred invites a day at least, more on slow days — but never at the expense of the green-screen images; invites are the flex task, not the priority. Note the number invited in your EOD.

The CEO's attention is the most limited resource in the company. Every tag fix, unanswered DM, and one-off insult that reaches him is time pulled away from work only he can do. The Social Media Operations Specialist exists to absorb all of that; your job is to keep the pipeline running and bring the CEO the pattern, never the incident.

To the Social Media Operations Specialist — same day, the default

• A comment needs removing and you lack permission → send the link plus a screenshot and ask.
• A comment to post or pin → send the video link and the exact copy.
• A wrong tag or credit on a repost → send the post link, the wrong tag, and the correct handle.
• Creators to be DMed → send the pack; you do not do first-touch outreach.
• Daily approvals and denials → an end-of-day note every working day, even if it is "nothing new".
• Rights secured but no post date, or a video with rights that never went up → ask for a date and follow up until you get one.
• A creator question you cannot answer (payment, exclusivity, usage terms, takedowns) → pass it up; do not improvise.
• Anything you would describe as annoying, stuck, or waiting on someone → the Social Media Operations Specialist, not the CEO.

Who to go to, by topic

• Reels and creator questions → the Social Media Operations Specialist, always.
• Facebook backend, ad comments, page management → the Executive Assistant (the Social Media Operations Specialist doesn't work the Facebook backend).
• Green screens → the team — the Executive Assistant, the Editor, and the Social Media Operations Specialist are all involved; the Executive Assistant is the best first stop.
• ManyChat technical problems → the Executive Assistant immediately; she engages the contractor who maintains the flows.
• Scripts and video ideas → queue for the CEO with links; he reviews in batch when available.

If the Social Media Operations Specialist is not responding

• Message in the normal working channel.
• No response in 24 hours → direct message with a clear ask and a deadline.
• No response in 48 hours → tag them in the shared channel so it is visible.
• Still blocked → log the item in the tracker as Blocked, with the date and every follow-up attempt.
• A non-response only becomes a CEO topic when it is a repeated, multi-week pattern measurably costing content — and then it goes in the weekly summary as a process issue with numbers attached ("six approved videos have sat without a date for two weeks; suggest a standing 48-hour rule"), not as a complaint about a person.

Red line — both, immediately

• Legal threats, cease-and-desist language, or any message from an attorney or claims agency.
• Copyright or trademark claims, strikes on the account, or takedown notices.
• Impersonation accounts, fake stores, or anyone selling our designs under our name.
• Account security issues: lockouts, suspicious logins, disabled ads or profiles.
• Credible threats, doxxing, or targeted harassment aimed at the CEO, his family, or any team member.
• A coordinated pile-on — a sudden unusual volume of hostile comments from accounts that look connected.
• Anything with press, partner, or public-figure exposure, or that could plausibly end up screenshotted outside the platform.
• For threats and harassment: screenshot first, preserve the handle and timestamp, do not reply, do not delete before it is documented. Hide it from public view if you can, but keep the record.

Rule of thumb: if you are asking yourself "is this a legal or safety thing", treat it as one and send it up.

Bring the CEO

• What is landing and what is not — which videos, formats, hooks, and topics people respond better to, and which fall flat.
• Language and delivery patterns — words, phrases, and pronunciations that keep coming up in the comments, and what you are noting for future takes.
• Sentiment direction — whether response to a format is trending up or down, and your read on why.
• A recommendation — always end with "do more of this, less of that". He wants a decision to make, not a report to interpret.
• Throughput numbers — reels sourced, DMs sent, rights granted, videos posted. One line, not a spreadsheet dump.
• Decisions only he can make — brand direction, a new content format, anything that changes how he shows up on camera.

Do not bring the CEO

• Individual nasty comments — by quote, screenshot, or paraphrase. Who said what about him: not ever.
• Tracker minutiae, single-creator status, one-off scheduling questions.
• Frustration with the Social Media Operations Specialist or anyone on the team.
• Anything phrased so it reads as "you are being attacked", or anything the Social Media Operations Specialist can resolve in under an hour.
• If a specific comment genuinely matters strategically, describe the theme without the quote: "there is a recurring complaint about pronunciation on regional dish names" is useful; repeating the insult is not.

The format he wants: "After looking across the last several videos, comments are noticeably more positive when you do X. They get more negative when you do Y. Recommend more X, less Y." Keep the tone casual and human — write like a person on the team, not a corporate policy document.

Credentials

• All brand logins live in LastPass and only in LastPass. Never paste a password into Slack, email, a text, a Google Doc, or a screenshot. If someone asks you for one — even someone who sounds like a teammate — point them to LastPass or to your manager.
• Use the LastPass shared folder you were given; do not copy credentials into your personal vault or browser password manager.
• Turn on two-factor authentication (2FA) on your own LastPass, your work Google account, and any account you are the primary holder of. Never approve a 2FA prompt or share a 2FA code you did not trigger yourself — that is how accounts get taken over.

Which account for which job

• The sourcing account ("Sunday Sauce Szn") is for finding content only. Allowed: watching, searching, liking, bookmarking, sharing posts to the brand page, following Italian creators, and hitting "not interested" — everything that tunes the feed. Not allowed: posting, commenting, sending DMs, following non-Italian accounts, or personal browsing — every action trains its algorithm.
• Brand pages (Hardcore Italians on Facebook/Instagram/TikTok): act only within your role — moderation, invites, ManyChat, flagging. Posting and creator DMs belong to the Social Media Operations Specialist.
• The CEO's personal Instagram, if you are granted it: reply to comments in the CEO's voice only; never post, change settings, follow/unfollow, or DM from it without explicit instruction.
• Your personal social accounts are never used for brand actions — no moderating from your own profile, no DMing creators personally, no reposting brand content claiming to speak for the brand.

Devices and sessions

• Work from your own device with a passcode/lock screen. Do not log into brand accounts on shared or public computers; if you must, log out fully and clear the session afterward.
• Keep the sourcing account logged in on one device only. If it gets logged out or asks for verification, tell your manager before retrying repeatedly.
• Keep the apps and browser updated.

Red flags — report immediately

• A login alert from a place you were not, a password-reset email you did not request, or a two-factor prompt you did not trigger — report these immediately, before anything else you are doing.
• A DM or email asking you to "verify" an account, claiming a copyright strike, or offering partnership money with a link to click — these are almost always phishing. Do not click; screenshot and send to your manager.
• Any post, story, follow, or message from a brand account that you did not make and cannot account for.

When you leave the role, your access is removed and shared passwords are rotated — that is normal and not personal. Return anything brand-owned and delete local copies of brand files.