How the BMD Marketing Library works

How it started: a swipe archive before it was a product

Before there was a public library, there was a local archive.

We collected swipe files the way practitioners always have: public-domain scans, free releases from rights holders, museum and university collections, and galleries that publish ads for study. The laptop mirror alone sat at 8,342 files across about 200 folders, roughly 44.9 GB. The rights basis was logged, not hoped for: thousands of US public-domain pieces, thousands more released by museums and universities, free author and brand releases, open licences, and a smaller study-only set that stays off the Saudi public site.

That archive was useful and still incomplete as a product. A folder does not teach. A PDF named after a dead agency does not tell a junior which job the ad was hired to do. A beautiful still with no pattern label trains the eye to copy surfaces.

So we turned the archive into layers.

First, cards and manifests: what the piece is, where it came from, what rights basis lets us keep it.

Second, Stage 1: the pattern layer. Cluster, name, gate. Seeds we already trusted (demonstration, story lead, reason-why, scarcity close, guarantee stack, and the rest). English survivors that earned a name after review. Twelve Arabic-first patterns reserved from Gulf playbooks: Ramadan emotional arc, WhatsApp close, cash-on-delivery trust, dialect versus fusha, family gathering close, and others that Western swipe books rarely name.

Third, the public site. Teaching chrome in our voice. Pattern hubs in English and Arabic. Galleries for stills and videos. Detail pages that can carry Why it works and Steal this when when we have authored text and a cleared rights path.

The swipe archive is still the warehouse. The library is the front room we will let a stranger into.

What the library is for

Two audiences, one rule.

Marketing teams building a department use it to stop arguing about taste and start arguing about jobs. If the brief needs risk removed, we look at guarantee stack and risk reversal. If the brief needs proof by performance, we look at demonstration. If the brief is a Ramadan window in Saudi, we look at the Arabic-reserved patterns even when their card counts are still zero. The library is a shared vocabulary for people who have to ship.

Practitioners use it to steal patterns, not ads. That sentence is the whole product. Copying a headline from a 1910 patent medicine ad into a 2026 fintech campaign is cosplay. Taking the mechanism (borrowed authority costume, or scarcity that is a real limit rather than a louder adjective) and rewriting it for your offer is craft.

Who it is not for: people hunting a free CDN of brand assets to republish as their own. People who want Hormozi screenshots on a .sa domain. People who want tobacco and alcohol "inspiration" slipped through because the image is old. We fail closed on those lanes. The library is smaller because of that. That is the point.

Define the product before we judge the screenshots

People say a swipe file is a collection of ads. Change the collection tomorrow. Did the team's judgment change? Usually no. So a swipe file, by itself, is not the product.

For non-marketers: the BMD Marketing Library is a public place to see how persuasion jobs show up in real ads, with enough structure that you can reuse the job on your own brief.

For marketers: it is a pattern taxonomy (64 named jobs today, capped well under 120 so names stay distinguishable), plus templated teaching on a first wave of those patterns, plus rights-cleared galleries of stills and official video embeds, plus per-item analysis where we have written Why it works and Steal this when.

The format is how the job arrived. The job is what we reach for.

What you will find on the hub

Open bmds.sa/library. The page does not open with a wall of thumbnails. It opens with the question, then the count: sixty-four jobs.

Navigation splits the work the way we work:

• Pattern index: seeds first (fifteen mechanism anchors we protect by name), then the English survivors that passed the first gate, then twelve Arabic patterns reserved for Arabic-language work with Saudi practice first.
• Ads gallery at /library/ads/: stills, rights-cleared for the public Saudi site.
• Videos gallery at /library/videos/: official YouTube embeds, not re-hosted files.
• Lane and country filters on the Arabic Ads path: English library and Arabic Ads as separate lanes, country chips with Saudi Arabia first, no fake "GCC group" chip that pretends six markets are one brief.
• Language twin at /ar/library/ with the same structure, Arabic chrome, Western digits.

As of this draft (20 Sep 2026, Asia/Riyadh), live sitemap counts sit near 9,557 unique ad detail IDs and about 475 unique video IDs on the English tree, each with an Arabic twin page. That is larger than the first public wave (3,474 stills and 174 videos). The catalog grew when the Arabic Ads lane landed university-archive stills and more official embeds. Numbers move. The hub copy should stay colder than the catalog.

Pattern pages carry study-pool card counts (demonstration alone sits near 196 cards in the study pool; celebrity endorsement near 366). Those pool numbers are teaching density, not a promise that every pool card is a public still.

Patterns: the unit of reuse

A pattern page is the spine.

On a live templated page such as guarantee-stack or demonstration-ad you will see a stable set of fields. The Stage 1c template pack names them in teaching language: what it is, why it works, use when, do not use when, the fill-in skeleton, worked examples, and a Gulf adaptation note. On the public chrome the same ideas show up as Define, Mechanism, Verdict, Don't, Copy skeleton, and Gulf note.

That mapping matters less than the discipline:

1. Define the job in plain words.
2. Mechanism names how persuasion moves (risk subtraction, proof by performance, loss of option).
3. Verdict / use when states the brief conditions that earn the pattern.
4. Don't states the conditions that make the pattern a tell or a lie.
5. Copy skeleton gives bracketed slots, not a finished ad.
6. Gulf note says what changes in Arabic and Gulf delivery: register, proof types, close channel, seasonality, refund culture.

Twenty-five of the sixty-four patterns already carry that full template pack. Thirty-nine still need it. We would rather ship 25 honest templates than 64 thin ones. The remaining templates are on the completeness track, not on a hype roadmap.

Arabic-reserved patterns can sit at zero cards on purpose. The name is reserved so clustering does not erase Gulf jobs that Western archives under-represent. Empty is honest. Pretend examples would not be.

Template fields, named

When we say a pattern is templated, we mean these fields exist and are filled, not that a PDF generator ran.

On the Stage 1c pack the labels are:

• What it is · one tight definition
• Why it works · the mechanism paragraph
• Use it when · brief conditions that earn the pattern
• Do not use it when · brief conditions that make it dishonest or wasteful
• The template · skeleton with [BRACKETED SLOTS]
• Adapting for Arabic and Gulf markets · or Gulf delivery notes on MENA-native patterns
• Worked examples · links into the library where rights allow

On the live pattern chrome the same ideas appear as Define, Mechanism, Verdict, Don't, Copy skeleton, and Gulf note. Verdict is the use-when judgment. Don't is the refuse list. If a field is missing on a pattern page, that pattern is still in the taxonomy but not yet in the 25-template wave.

Steal this when on an ad detail page is not the same object as Use it when on a pattern page. The pattern field is about the job. The ad field is about when this instance's mechanism is worth stealing again. Read both. Do not collapse them into one vague "inspiration" note.

Why it works and Steal this when

Pattern pages teach the job. Ad detail pages teach the instance.

Where we have BMD-authored analysis and a rights join that passes the Saudi public gate, the detail page shows two blocks:

• Why it works: the mechanism in this specific piece, not a generic compliment.
• Steal this when: the brief conditions under which you should reach for that mechanism again.

Example shape from a public detail page: a historical still filed under borrowed medical authority explains that visual authority transfers trust, then tells you to steal it when you are selling a home-use version of a professional service and need the persona of the professional without faking a licence you do not have. The steal is conditional. That is the point.

Coverage is uneven on purpose.

On the original English analysis set, English Why it works sits on 3,474 ad detail pages. After PR #89, Arabic Why it works and Steal this when sit on 2,637 of those pages (up from 422). We do not machine-translate the missing Arabic blocks. If the Arabic field is empty, the Arabic page omits the block. Fail closed.

Newer Arabic Ads stills that entered through university archives often ship without analysis until a human writes it. Video detail pages gained analysis later than ads; some videos still omit cleanly where no sidecar exists. An empty block is a receipt that we refuse to invent.

Saudi Instagram commercial embeds are a separate, careful lane: official embeds, not scraped files. Live brand-style embed IDs sit near 79 after the first waves. That number will grow only when rights and commercial-frame checks pass. Embeds can break. When oEmbed fails, we fail closed rather than hotlink a pirate mirror.

Templates in practice: a worked path

Here is how we want a team to use a templated pattern on a real brief. Not theory. A Monday morning path.

Step 1. Name the job, not the channel. The brief says "we need a LinkedIn ad" or "we need a Ramadan film." Translate that into a job: remove risk, prove performance, create a real limit, open a story, close on WhatsApp. If you cannot name the job, you are not ready for the gallery.

Step 2. Open the pattern page. From the hub, pick the pattern. Read Define and Mechanism before you look at examples. If you skip to thumbnails, you will copy wardrobe.

Step 3. Read use / don't (Verdict / Don't). Check your brief against both lists. Guarantee stack is for unfamiliar or expensive offers where buyers expect disappointment. It is the wrong tool when you cannot honor every clause, or when the product is already a trusted commodity and a theatrical stack becomes a tell.

Step 4. Copy the skeleton into your doc. For guarantee stack the public skeleton looks like: offer, three cover clauses (time, scope, claim process), what the buyer keeps even if they refund, then the ask. Leave the brackets. Fill them with your facts.

Step 5. Read the Gulf note before you brief Arabic. The guarantee-stack Gulf note is concrete: state the warranty in plain Arabic, name the company entity that honors it, and pair with cash-on-delivery or partial-pay language when refund culture is weak. That is not decoration. That is the difference between a translated US refund paragraph and a Saudi-usable close.

Step 6. Open three examples, not thirty. From the pattern page or the ads gallery filtered to that pattern, open a few detail pages. Read Why it works. Read Steal this when. Steal the mechanism sentence into your brief as a constraint ("the close is the disappearing chance, not a louder adjective"). Do not steal the layout.

Step 7. Write your draft from the skeleton. If your draft only works when it still looks like the example, you copied. If your draft works when you change category, offer, and face, you stole the pattern.

That is the product loop: hub → pattern → template → examples → mechanism → your brief.

Galleries, Arabic Ads, and country filters

The ads gallery is for stills we can show. The videos gallery is for official embeds we can show. Both exist so the pattern page is not an abstract lecture.

Arabic Ads is a lane, not a sticker. University-archive commercial stills, Saudi-first country chips, and official social embeds sit beside the older English public-domain spine. Other Arabic material can sit as supporting inspiration without pretending every Arabic still is a Saudi brief.

Country filters are one country each. Saudi Arabia first. No GCC mega-chip. If a filter returns few Saudi stills, that empty state is information about the catalog, not a bug to paper over with mis-tagged Egypt rows.

Videos stay embed-only. We do not host YouTube files. That keeps rights aligned with the platform terms we actually rely on.

Rights and ethics, at high level

The library publishes what we can defend.

Public domain, museum and university open releases, brand free releases, Creative Commons where the licence fits, and official embeds. Study-only galleries, unverified Arabic dumps, tobacco and alcohol holds, and named deny lists stay off bmds.sa even when they remain in the private archive for research.

We learn from examples we can show. We do not launder examples we cannot show.

Application notes on individual pieces stay sparse. Empty is better than filler. Analysis is authored or omitted. Translation of Why it works into Arabic is human and joined by id, not auto-translated at render time.

If that posture feels slow, it is. Speed without a rights spine is how swipe blogs get taken down.

How to use the library end to end

A short path for a first visit:

1. Start at the hub. Read the opening question. Skim seeds if you are new; skim Arabic reserved patterns if your brief is Gulf-native.
2. Pick one pattern that matches the job.
3. Read the template fields in order. Do not skip Don't.
4. Open the ads or videos gallery from that pattern, or from the lane/country chips if you need Arabic Ads or Saudi-first examples.
5. Open two or three detail pages. Harvest Why it works and Steal this when into your brief as constraints.
6. Write from the skeleton. Change every surface noun until only the mechanism remains.
7. If the pattern has no template yet, still use Define and Mechanism, and treat examples as study rather than a fill-in pack.

If you leave with a saved ZIP of images and no named job, the library failed you, or you used it as a CDN. We designed for the first failure mode to be rarer than the second.

What is still growing

Honesty over launch language.

• 39 patterns still need the full template pack.
• Arabic Why it works covers 2,637 joined pages, not the entire multi-thousand ad catalog. New Arabic Ads stills often omit analysis until written.
• Video analysis is partial relative to the video catalog; omit stays clean where no sidecar exists.
• Saudi embeds are tens, not thousands. More will land only through the commercial-frame and rights pipeline.
• Native Arabic swipe beyond embeds and university lanes is still thinner than the English public-domain spine.
• Detail navigation (related ads, stronger next/prev, "has analysis" filters) is still product work.
• Taxonomy headroom remains under the 120 cap if Gate review earns new Gulf names.

We might be wrong about the order of those tracks. We are not wrong about leaving gaps visible.

Season 1 of the case-study series is closed at EP24. This library post is not episode twenty-five. It is the manual for a product that should outlive any one series.

A note on counts that move

Catalog counts will drift between the day we draft and the day we publish. Treat the receipts at the bottom as a dated snapshot (20 Sep 2026, Asia/Riyadh), then re-check the live hub and sitemap before the post goes out. The teaching claim does not depend on any single thousands-digit. It depends on the posture: patterns first, examples second, gaps visible, rights fail closed.

If the Arabic Ads lane adds another university batch next week, the ad total rises and the analysis coverage percentage falls until humans write Why it works. That is expected. We do not paper the percentage with generated Arabic.

Receipts used in this draft

A dated snapshot (20 Sep 2026, Asia/Riyadh). Re-check the live hub and sitemap if you are quoting a thousands-digit.

• Live hubs: bmds.sa/library · bmds.sa/ar/library
• Patterns on hub: 64 (15 seeds + English survivors + 12 Arabic reserved)
• Templates: 25 / 64
• Live unique ad IDs (EN tree): about 9,557
• Live unique video IDs (EN tree): about 475
• Why it works EN (analysis set per PR #89): 3,474
• Why it works AR after PR #89: 2,637 (was 422)
• Saudi Instagram-style commercial embed IDs: about 79
• Local swipe archive index: 8,342 files, 44.9 GB, about 200 folders
• Season 1 closed at EP24

Frequently asked questions

About BMD

Most companies don't have a marketing problem. They have a marketing department that was built to underperform. BMD is a boutique consultancy based in Riyadh, Saudi Arabia, that builds structured, measurable marketing departments inside mid-market companies across the GCC. We don't run your campaigns, and we don't hand over a strategy deck and leave. We build the structure, the measurement and the clear responsibility that turn marketing into a function leadership can rely on. Delivered in Arabic and English, founder-led.

Redha Alayesh

Redha Alayesh, Founder of BMD, is a marketing consultant who has worked with 40+ marketing departments.

The verdict

The brief will still arrive. The only open question is whether we reach for a format we like, or a job we can name.