Genius Loci is not a print shop anymore. It is a premium AI, marketing and programming house — an agentic luxury brand on the cutting edge, building new things every day and curating and selling only the best of the best. Content’s main job now is to bring to light everything built here — the system, the agents, the automations — as social proof, as something cool, and as a reason to share. Video leads, because complex ideas need motion. Statics and carousel statics break up the feed. Minimum 3–5 pieces a day. We don’t slow down.
Futuristic. Premium. Agentic. Luxury. Never-tired.
The adjectives are the brief, verbatim — and they are a production spec, not a mood. This page is the reference render: paper ground, ink type, one brass accent, mono data labels, hairline rules, no decoration that carries no information. Every post, reel, static and carousel ships in this system or it does not ship. The hard rules:
Four failure modes, and every one of them is now punished twice — once by the audience, and once by Meta itself. Since July 2025, and harder since March 2026, Meta deprioritises unoriginal content, can make a Page non-recommendable, and can withhold monetisation. Photorealistic AI imagery is auto-labelled “AI info” from C2PA/IPTC provenance signals. The old system wasn’t just underperforming — it was accumulating platform risk on a Page that is already monetised and already down 57% in views (27.8K → 11.8K per 28 days). The fix is not to hide the machine — it is to show the real work: real builds, real screens, real decisions, real results.
Same composition, same angle, same scale, same gradient void, post after post. Uniformity at volume is itself an AI/slop signal — to humans scrolling past, and to Meta’s duplicate detection. A grid that never breathes reads as generated before anyone reads a word.
Generic superlatives, interchangeable copy (“elevate your everyday carry”), no named reference, no measured spec, no price, no verdict that could be wrong. Editors call the tell the words-to-ideas ratio: four sentences to make a one-sentence point.
AI imagery of products triggers the trust collapse the data documents: 78% of consumers say an AI-generated image cannot be “authentic” by origin, ~40% would trust a brand less if AI use were undisclosed, and only 14% are very likely to buy a product shown with AI images. For goods people buy on fidelity — dial colour, leather grain, paint condition — that cost is fatal.
Meta’s March 13, 2026 originality post lists the deprioritised patterns explicitly: re-uploads with “low-value changes like adding borders, inserting captions, and changing the reel’s speed,” watch-along and facial-expression reactions, stitch compilations, narrating what’s already on screen. The old slate leaned on several of these. Your own 14-day learning data agrees: reactions posted a median of 1 view and a 0% beat rate — the worst format measured.
“Slop: digital content of low quality that is produced usually in quantity by means of artificial intelligence.” Merriam-Webster — 2025 Word of the Year. The audience now has a word for the failure mode.
The headline finding is a split verdict, and it reshapes the whole slate: on Facebook, photos win engagement per person reached; video wins discovery by non-followers. Anyone claiming one format simply “wins” is mixing the two metrics. For Genius Loci the portfolio is video-led — video is how complex builds and ideas are actually explained and discovered — with statics and carousels as the daily rhythm that holds followers, earns saves, and breaks up the feed.
Reactions + comments + shares ÷ reach. Buffer, State of Social Media Engagement 2026 (52M+ posts, 2025 data).
⚠ Do not cross-compare absolute rates between studies: Socialinsider measures engagement by followers (≈0.15% average) where Buffer measures by reach (≈5.6%). The orderings agree — images and albums strong, links weak. Re-base any target on Genius Loci’s own Page Insights before using it as a goal.
On Facebook, multi-photo albums out-impression video and single images (Socialinsider). The album is the Facebook-native form of the roundup carousel — several products, several chances to be re-served.
Reels reach ~36% further than carousels and ~125% further than a single photo; carousels earn ~109% more engagement per person reached than Reels (Buffer). Carousels with music are eligible for the Reels tab, and an un-swiped carousel is re-served starting at slide 2 — two mechanics no single photo gets (Mosseri, via Social Media Today).
Instagram’s stated top signals: watch time, likes per reach, and sends per reach — with sends weighing slightly more for non-followers. A sendable post (“send this to someone who needs this”) out-distributes a merely likeable one.
Meta reports views and watch time on original Reels roughly doubled in H2 2025 after the crackdown began. And since June 2025, all Facebook video is published as Reels — every product film enters the recommendation system and is judged on completion.
Top accounts post consistently; a week with no posts underperforms the account’s own growth baseline, while per-post reach falls at very high frequency. A sustainable franchise cadence wins (Buffer).
The old slate sold objects. The new slate proves a system. Products are one lane — curated, best of the best — not the centre. The centre is everything built here: SuperOS, the agents, the workflows, the decisions they make while nobody is watching. That is the social proof. That is the shareable thing. That is the authority.
Video leads. Complex ideas — how an agent works, why a system was built a certain way, what changed in marketing or code today — need motion, voice and a screen. Statics and carousel statics are the rhythm between videos: proof, diagrams and curated picks that a follower can save in two seconds.
The fixed elements never change: franchise name in the first line, same caption skeleton — hook → proof → what it means → one CTA — same closing move. The variable elements always change: the build, the agent, the result, the lesson, the pick.
Build-in-public, at luxury standard — one new thing, shipped today.
Every day, one thing that did not exist yesterday: a workflow, an automation, a feature, a system fix. Filmed as it happens — screen, hands, voice — not rendered afterwards. This is the flagship because it does three jobs at once: it teaches something complex, it proves we build every day, and it gives people a concrete reason to share: “look what they made.”
“We built a gate that rejects our own work. Today it held 12 drafts. Here is the one that passed.”
“This did not exist at breakfast. It is running now — watch it work.”
Voice rule: statement first, number inside the first line, zero hype adjectives. If the day’s build has no working proof shot, the slot does not run.
Frame: 16:9 master capture, cut to 9:16 with the action kept in the centre column — both crops planned at record time, never improvised in edit.
Overlay: Plex Mono bug top-left — THE BUILD · [DATE]; Newsreader 800 headline on the lower third over a paper block; brass hairline under the headline; mono timestamp + job ID bottom-right. No other graphics. Nothing baked into the footage that layout can set sharper.
Day’s build session (screen recording + kernel job log) → content_slot job drafts the script skeleton → llmRouter smart tier via editorial_polish (NVIDIA first, Sonnet fallback) → truth gate: every claim checked against the session log → QA gate → content-direction.json format gate (build enabled) → Review Desk approval → publish FB + IG + Threads.
Original by definition — nobody else has this footage, this system, or this day’s build. It is the strongest possible answer to Meta’s originality enforcement, and the clearest social proof a premium AI brand can publish.
The factory floor, reimagined — not products moving, but intelligence moving.
Short videos of the system doing real work: an agent researching, drafting, checking, routing, fixing, or getting held by a quality gate — and what a human decided next. Failures stay in. Holds stay in. That honesty is the luxury signal: we are so confident we show the machine room.
“While you slept: 12 jobs ran, 7 rotations turned, $0.92 spent. Here is what shipped.”
“At 02:14 the gate held its own draft. Nobody was awake. That is the point of the machine.”
Voice rule: the overnight numbers open the piece, verbatim from the ledger. The replay is the hook — no intro, no presenter, no summary voice.
Frame: 9:16 screen-led. The real queue, log and ledger fill the frame; the counts are set over them in layout, never drawn into a mock UI.
Overlay: Plex Mono top bar — AGENTS AT WORK · OVERNIGHT RUN; the three counts (jobs / rotations / spend) as oversized Newsreader numerals, one per beat; hold badges set in mono — HELD — beside the job they stopped; locked-kernel and cap signage reproduced exactly as the system prints it.
Overnight kernel-jobs ledger + spend ledger (jobs, rotations, $, holds) → run-summary job assembles the replay → llmRouter smart tier via editorial_polish (NVIDIA first, Sonnet fallback) → truth gate: every number re-checked against the ledger → QA gate → content-direction.json format gate → Review Desk approval → publish as the morning-after video.
This turns infrastructure into content. Every episode is proof the brand is rapidly evolving — and proof no competitor can copy without building the system first.
Ecommerce, AI and marketing — explained like the authority in the room.
One complex idea, broken down in 45–90 seconds: a model release that actually matters, a marketing system that works, a programming pattern behind the agents, a store decision and the data behind it. Not news-reading. A position, with receipts from our own work wherever possible.
“Most AI content fails one test: show the receipt. Here is ours.”
“A 9B model drafted this. A gate rejected it. Both of those facts are the lesson.”
Voice rule: a verdict that could be wrong, then the evidence — never a topic announcement. “Today we’re talking about…” is a banned opening.
Frame: 9:16 explainer, presenter or screen share alternating on the cut; diagrams built in layout software at final resolution.
Overlay: the claim set in Newsreader 800 on a paper block, lower third; a Plex Mono evidence strip pinned to the bottom of frame for the whole piece — source name + the number that backs the claim; brass hairline separates claim from evidence. One claim per piece; a second claim is a second Breakdown.
Live system example + research note → editorial_draft job writes the claim-and-evidence script → llmRouter smart tier via editorial_polish (NVIDIA first, Sonnet fallback) → truth gate: every claim needs a named source or our own measured data → QA gate → content-direction.json format gate → Review Desk approval → publish.
Authority compounds. These are the most shareable teaching pieces in the slate — the ones sent to a partner, a client, or a team with “watch this.”
One static, one undeniable fact — the feed’s daily rhythm breaker.
A single static between videos: a result, a before/after, a screenshot detail, a number from the system, a finished output next to the brief that made it. No essay. If The Build is the story, The Proof is the receipt stapled to it.
“12 held. 1 shipped. That ratio is the product.”
“$0.40 of $1.00. The cap held. So did the standard.”
Voice rule: the card is the caption’s first line, word for word. The Proof never explains itself twice — one fact, one frame, done.
Frame: 1:1 or 4:5 card on the paper ground. One number or result at Newsreader 800, oversized; Plex Mono label above it naming the metric and the window; brass hairline beneath; a single real screenshot detail or output inset below the rule — never a generated backdrop.
Overlay: mono footer strip — THE PROOF · [DATE] · GENIUS LOCI SYSTEM. No logo lockups, no badges, no second message.
Same-day ledger result (kernel counts / spend / shipped output) → content_slot job, static variant → llmRouter smart tier via editorial_polish (NVIDIA first, Sonnet fallback) for the two-line caption → truth gate: the number must match the ledger exactly → QA gate → content-direction.json format gate → Review Desk approval → publish as the midday static.
Statics earn the engagement-per-reach that video doesn’t (see §02), cost minutes once the day’s work exists, and keep the feed from becoming an unbroken wall of video.
Carousel statics that map the machine — swipeable, saveable, shareable.
A numbered carousel that lays out a system in slides: the tools in a workflow, the models in the router, the steps in an automation, the stack behind a result. Each slide is one component with one line on why it earned its place. This is the “send this to someone building” format.
Cover: “5 tools. One router. A $1-a-day ceiling it cannot cross.”
Final slide: “The full stack, one frame. Save it — this is the map we run on.”
Voice rule: every slide names its component and states what it replaced. A component with no “replaced” line hasn’t earned its slide.
Frame: 4:5 slides, identical crop and margins across the set. Cover = number + system + claim in Newsreader 800 on paper; component slides = real screenshot or clean diagram of that one component, one line of Archivo beneath.
Overlay: Plex Mono slide counter top-right — 02 / 06 — and a brass progress hairline that fills slide by slide; final slide carries the full-stack recap and the single keyword CTA. Music on the Instagram version for Reels-tab eligibility.
Stack definition in content-direction.json + live router / workflow config → carousel assembled from real screenshots and layout-set diagrams → captions via llmRouter smart tier, editorial_polish (NVIDIA first, Sonnet fallback) → truth gate: each component claim checked against the live config → QA gate → format gate → Review Desk approval → publish as FB album + IG carousel.
Albums lead Facebook for impressions and carousels earn ~109% more engagement per reach than Reels on Instagram (Buffer; Socialinsider). On Instagram, add music for Reels-tab eligibility and the un-swiped re-serve from slide 2.
Curation, kept in its lane — only the best of the best, with a reason.
Products still matter — as one lane. When Genius Loci sells something, it is because it passed the same standard as everything else here: best in class, useful, worth the money. No filler drops, no generic catalogue posts. Every pick carries the why, the evidence, and who should skip it.
“We scored the whole category. This is the one that survived.”
“Best of the best is a short list. This week it has exactly one name on it.”
Voice rule: the standard is stated before the product is named — and who should skip it is always in the caption. A pick that suits everyone convinces no one.
Frame: 4:5 or 1:1 real photography — the product in use, in place, in real light, a hand for scale where it helps. No floating renders, no generated objects, no exceptions.
Overlay: Plex Mono price line and a brass KEPT verdict mark set in layout on a paper strip; the product tag carries the purchase path — a caption link never does. Carousel version: contenders on slide 2 with their scores, the survivor last.
Catalog index + release decision (no decision, no post) → pick drafted with scores and skip-case → llmRouter smart tier via editorial_polish (NVIDIA first, Sonnet fallback) → truth gate: price, spec and stock checked against the catalog → QA gate → content-direction.json format gate → Review Desk approval → publish with product tags from the Shopify-fed catalog.
Scarcity of endorsement is the luxury. Because The Select appears 2–3 times a week — not ten times a day — a pick actually means something, and product tags turn that trust into a tappable path to the store without burning a link slot.
3–5 pieces a day only works if the day’s real work is the shoot. We do not stage a separate content day. We capture the build session once — screen recording on, camera on the desk — and cut it five ways. That is how 21–35 pieces a week (3–5/day × 7 days) is a capture habit, not a second job.
Screen record + desk camera + real audio for every build / agent session. The work is the footage. No re-enactment, no scripted “content time.”
The Build (full story) and Agents at Work (tight proof cut) come from the same recording. Video leads the day because it carries the complex idea and the discovery.
From the same session: The Proof static (best frame / result) and The Stack carousel (components, in order). Minutes of layout work — the substance already exists.
Three times a week, the day’s lesson becomes The Breakdown — 60 seconds, verdict first, evidence from the session you just watched being built.
Morning: The Build video. Midday: The Proof static. Afternoon: Agents at Work or The Breakdown video. Evening: The Stack carousel or The Select. That is 3–4 pieces on a normal day, 5 on a Drop / Select day.
7 Builds + 5 Agents + 3 Breakdowns + 7 Proofs + 4 Stacks + 3 Selects = 29 pieces/week, ≈4.1/day — inside the 3–5/day floor, from two real work sessions a day, not a content factory.
The fixed spine of every day, labelled the way the Loci System labels everything: frame, caption hook, overlay. The first two slots are shown as a worked replay — the overnight run and the day’s gate results, with the real shape of the numbers the system reports. Authority and Select pieces rotate into the afternoon and evening slots around this spine (see §07).
“The gate did its job today. Here is what it stopped — and what shipped.”
holds 12tech_reel heldlocked kernel$0.40 / $1.00 caps“While you slept: 12 jobs, 7 rotations, $0.92. Here is what shipped.”
12 jobs7 rotations$0.92 spent“12 held. 1 shipped. That ratio is the product.”
Cover: “5 tools. One router. A $1-a-day ceiling it cannot cross.”
“Looks pro” is not a vibe — it’s a checklist, and most of it is subtraction. This table is the gate every post passes before it ships; it replaces taste arguments with observable facts.
Attention here converts twice. First, into authority: the daily proof that Genius Loci builds, ships and knows — that trust is what sells services, partnerships and everything after. Second, into the store: The Select’s curated picks, sold through product tags and DMs. Link posts serve neither job well — with 98.9% of Feed views link-free and link posts the lowest-engagement format, that road is closed. The working funnel in 2025–26 is native-first, and where it ends on the store, Meta ended native Shop checkout in September 2025: the Shop is now a discovery surface that hands the shopper to the merchant’s site to pay.
Hero photo set or album carries the full verdict in the caption — the post is valuable with zero clicks. No caption link, ever: Meta’s own dashboard has warned since 2025 that caption links harm distribution.
The post ends with a single keyword CTA — “Comment STACK and the full breakdown lands in your DMs.” A Meta-approved automation (ManyChat is the category standard) sends the breakdown, links and — for a Select pick — prices, and can collect an email opt-in first. The comment feeds the ranking signals; the DM does the converting.
A single Shopify-fed catalog powers the Facebook Shop, Instagram product tags and ads. A tagged product in a post is a tappable path to the product page that does not consume a link slot — which makes catalog hygiene (accurate titles, in-stock status, real high-resolution photos) a distribution input, not back-office admin.
Since December 2025 Meta has been testing a cap of 2 organic link posts per month on some Pages, with September–October 2026 reports saying links in Page comments now count toward it too. Spend both on the month’s highest-intent pieces — a flagship Build with a real offer behind it and a Select reveal — and put every related product link in that post’s comments. Ads are exempt from the cap: a small always-on catalog ad set carries systematic store traffic.
Email/SMS opt-in inside the DM flow converts rented reach into an owned list — the only asset the next algorithm change cannot take away. Stories link stickers and the profile link round out the path for high-consideration pieces, where the sale closes in conversation.
The comment-funnel numbers are vendor numbers. ManyChat’s own blog reports Reels using its automation averaged 326 comments vs 22 without (Base44’s analysis of 350 sponsored Instagram Reels) — engagement counts from an interested vendor, not audited sales. No independent, curation-brand-specific revenue case study was found in the research. Treat the funnel as well-evidenced mechanics, unproven revenue — and measure it ourselves from week one.
The link cap needs on-Page verification. The 2-per-month limit rests on multiple consistent secondary reports of Meta’s Help Center, not a newsroom announcement, and “limits may not apply to all Pages.” A counter appears in the Page composer when adding a link — check ours before restructuring around it (it’s a Phase 0 task).
Stay off the engagement-bait line. Meta demotes classic bait (“tag a friend”, comment-to-enter). A keyword that delivers a genuinely requested list is the compliant form of the same behaviour; a keyword with no payoff, or single-word comment farming, is the risky form. One keyword, real value in the DM, real comments answered in a human voice.
Four pieces on a normal day, three on a light day, five when The Select or a launch lands: 29 pieces a week, ≈4.1 a day. Video occupies the morning and afternoon anchors every day; a static or carousel sits between them so the feed breathes. Tap a slot to light up its franchise card above.
| Mon | Tue | Wed | Thu | Fri | Sat | Sun |
|---|---|---|---|---|---|---|
| The BuildAM video · flagship | The BuildAM video · flagship | The BuildAM video · flagship | The BuildAM video · flagship | The BuildAM video · flagship | The BuildAM video · flagship | The BuildAM video · week’s best build |
| The ProofMidday static | The ProofMidday static | The ProofMidday static | The ProofMidday static | The ProofMidday static | The ProofMidday static | The ProofMidday static · week in numbers |
| Agents at WorkPM video · system proof | The BreakdownPM video · authority | Agents at WorkPM video · system proof | The BreakdownPM video · authority | Agents at WorkPM video · system proof | Agents at WorkPM video · system proof | The BreakdownPM video · week’s lesson |
| The StackEvening carousel | The SelectEvening · curated pick | The StackEvening carousel | The StackEvening carousel | The SelectEvening · curated pick | The StackEvening carousel | Agents at WorkEvening video · weekend run |
COUNT: 7 BUILDS + 7 PROOFS + 5 AGENTS AT WORK + 3 BREAKDOWNS + 4 STACKS + 2 SELECTS IN THIS GRID = 28 SLOTS; THE THIRD WEEKLY SELECT ROTATES INTO A 5-PIECE DAY, FOR ≈29/WEEK. LIGHT DAYS DROP THE EVENING SLOT AND STILL CLEAR THE 3/DAY FLOOR.
The reset archived everything; this is the list that does not come back, and why. Everything else from the old slate is parked in the archive — its useful DNA has already been folded into the franchises above.
The brand has transformed. Print_of_day, the print-shop story and every format built around printing are dead — Dominic doesn’t print anything; made-to-order printed objects are an occasional sale, not an identity. No franchise in the new slate is anchored to a press, a print floor, or a daily product render.
Dead on three counts: Meta’s March 2026 post names watch-along and facial-expression reactions as unoriginal; your own data gave them a median of 1 view and a 0% beat rate; and the pipeline was broken at the input anyway — the YouTube downloader has been bot-check-flagged since Oct 8. Escalating reaction writing would have polished a pipeline that fails before it starts.
A manufacturer’s press photo, a logo border and a caption is Meta’s literal worked example of unoriginal content. Repeat offences risk a non-recommendable, demonetised Page. Supplied assets may still appear — credited, licensed, and carrying our own photography and analysis on top. Never alone.
Auto-labelled “AI info” when provenance signals are detected, and commercially toxic even when it isn’t: 14% buy intent (Clutch), 98% calling authentic imagery pivotal (Getty). AI keeps the stagehand jobs — cleanup, backgrounds, layout assistance — per the rule in §05.
Link posts: lowest engagement format, ~1% of Feed views, and now rationed — they survive only as the two monthly flagship spends. Talking-head takes written from summaries (the old industry_take pattern; the lane has been off since Oct 6) are “confidence without receipts” in video form. If we didn’t handle it, measure it or stand in front of it, we don’t opine on it.
Research compiled 20 Oct 2026. “Verified live” = the page was opened and read during research; “index” = search-result text. Full hedges travel with each claim in the sections above.
GENIUS LOCI
Occasional new arrivals, unusual finds, and useful field notes. No daily noise.