📊 Performance Marketing Weekly Intelligence Brief

A fully automated system that scans 30 industry publications, uses AI to curate the ~12 stories that matter for a performance marketing agency, and delivers them to Slack every Monday morning.

🏗️ System Architecture

📡 30 RSS Feeds

Adweek · Digiday · SEJ · TechCrunch · Nikkei · CNA · e27 · The Diplomat · and 22 more

→

📥 blogwatcher-cli

Go binary scans all 30 feeds concurrently in ~60s. Stores articles in local SQLite. Tracks read/unread state.

→

🧠 Hermes Agent + AI

Monday 9am cron trigger → DeepSeek V4 reads 400 titles → filters noise → deep-reads ~15 → curates to ~12

→

📤 Make.com → Slack

JSON payload to webhook → Slack mrkdwn → team channel

Detailed Architecture — Performance Marketing Intel System LAYER 1 — RSS Feed Sources (30 feeds, 8 categories) Adweek · AdExchanger · AdPulp Digiday · Marketing Dive MarTech · SEJ · SmartBrief Ad Tech Daily · VideoWeek TechCrunch · The Verge Ars Technica · Wired AI Social Media Today Creator Econ · ICYMI Adweek Creator · NetInfluencer Nikkei Asia · CNA Latest CNA Business · e27 Tech in Asia · KrASIA Vulcan Post · DealStreetAsia The Diplomat Gallup Asia Pacific Eco-Business 30 feeds total across 8 categories 📥 blogwatcher-cli — RSS Feed Aggregator & Article Database Scans 30 RSS/Atom feeds concurrently (8 workers) · Stores articles in SQLite · Tracks read/unread state · ~60s per scan 🧠 Hermes Agent Cron Worker + DeepSeek V4 Pro Triggered every Monday 9am SGT · Isolated worker session · Full pipeline in ~3 minutes 1. Scan 30 feeds ~60 seconds 2. Read 400 titles Filter → ~15 3. Deep-read ~15 Full article extract 4. Curate → ~12 Relevance test 5. Write + POST Make.com → Slack AI Model DeepSeek V4 120K ctx window ~$0.25 / run Marketing/Ad Tech/AI Creator/Social APAC Policy/Consumer

⚙️ How the System Works

Step 1 — Feed Collection

Every Monday at 9am SGT, a cron trigger fires. A lightweight Go program called blogwatcher-cli hits all 30 RSS feeds simultaneously and downloads their latest article lists. This takes about 60 seconds and produces a batch of ~400 article headlines (titles, dates, source URLs). Each feed typically returns 10–50 recent articles.

Step 2 — AI Title Scan (Pass 1)

The full article list — all ~400 titles — is sent to DeepSeek V4 Pro. The AI reads every title and applies a relevance filter based on its curation prompt. It asks: "Would a performance marketing agency care about this?" Articles about consumer gadgets, gaming, space, pure politics, company PR, and generic listicles are eliminated at this stage. This pass narrows 400 articles down to ~15.

Step 3 — Deep Content Reading (Pass 2)

The AI fetches the full text of the ~15 surviving articles using web_extract — a tool that converts any web page to clean markdown. It reads each article in full and evaluates whether the substance matches the headline. Some articles that seemed promising by title get dropped here if the content is thin or irrelevant.

Step 4 — Curation & Writing

Each surviving article is evaluated against a stricter test: "What should a media buyer DO with this information?" The AI writes a one-line summary, adds context on why it matters for the agency, and categorizes it into one of 8 output categories. Empty categories are omitted entirely — no filler. The brief is written in Slack mrkdwn format.

Step 5 — Delivery

The brief is JSON-encoded as {"data": "..."} and POSTed to a Make.com webhook. Make.com formats it and delivers it to the designated Slack channel. Simultaneously, the system marks all processed articles as "read" in the database so next week's scan only sees new content.

🧠 Curation Logic — What Gets Included vs Filtered

The AI follows a structured prompt that defines exactly what matters and what doesn't. Here's how it decides:

Prioritized — Makes the brief

Filtered out — Never appears

CategoryReal examples from June 11 scan (filtered out)
Consumer gadgets"Logitech's foldable mouse", "Starlink $10 hardware fee"
Gaming / entertainment"Valve kills gift card program", "Paramount vs Netflix"
Automotive"Honda Prelude review", "GM energy storage"
Space / science"NASA assigns Artemis III crew", "Commonwealth Fusion reactor"
Pure politics"Alex Vindman running for Senate", "RFK Jr. controversy"
Company PR / profiles"Five friends making comedy together", individual creator brand profiles
Healthcare"OB-GYNs release vaccine schedule", "Diabetes org apologizes"
Military"Cheap Iranian drone downs US helicopter"

📡 Sources by Output Category

Our 30 feeds are organized to directly feed the 8 sections of the final brief. Here's which sources map to which output:

📱 Media Buying & Ad Platforms
  • Adweek — adweek.com/feed
  • AdExchanger — adexchanger.com
  • AdPulp — adpulp.com
  • SmartBrief Marketing — smartbrief.com
  • Ad Tech Daily — adtechdaily.com
  • VideoWeek — videoweek.com
📈 Analytics & Martech
  • Marketing Dive — marketingdive.com
  • Digiday — digiday.com
  • MarTech.org — martech.org
  • Search Engine Journal — searchenginejournal.com
🤖 AI & Emerging Tech
  • TechCrunch — techcrunch.com
  • The Verge — theverge.com
  • Ars Technica — arstechnica.com
  • Wired (AI section) — wired.com
👥 Social Media
  • Social Media Today — socialmediatoday.com
💰 Creator Economy
  • Creator Economy (Peter Yang) — creatoreconomy.so
  • Adweek Creator Economy — adweek.com
  • ICYMI (Lia Haberman) — liahaberman.substack.com
  • Net Influencer — netinfluencer.com
🌏 SEA & APAC Intelligence
  • Nikkei Asia — asia.nikkei.com
  • CNA Latest — channelnewsasia.com
  • CNA Business — channelnewsasia.com
  • e27 — e27.co
  • Vulcan Post — vulcanpost.com
  • Tech in Asia — techinasia.com
  • KrASIA — kr-asia.com
  • DealStreetAsia — dealstreetasia.com
⚖️ Policy & Regulatory
  • The Diplomat — thediplomat.com
🛒 Consumer Behavior
  • Gallup Asia Pacific — news.gallup.com
  • Eco-Business — eco-business.com

📝 The Curation Prompt

The AI doesn't guess what matters — it follows a structured 400-word prompt that defines every aspect of its curation behavior. This prompt is what makes the system produce consistent, high-quality briefs every week.

Click to view the full system prompt the AI receives
You are curating a weekly intelligence brief for a performance marketing agency. Your audience are busy agency executives who need SIGNAL, not noise. You must REVIEW and SELECT before reporting — never dump raw article lists. CURATION PHASE: 1. Scan all 30 feeds for new articles 2. List all unread articles 3. Read every title — filter out: PR fluff, gadget reviews, gaming, space, pure politics, generic listicles, anything that fails "so what for a media buyer?" 4. Deep-read the 8-15 promising articles via full content extraction 5. Apply final relevance test — each item must have a clear action/implication 6. Skip empty categories entirely — no filler OUTPUT CATEGORIES (only include if articles exist): 🔴 Top Signals — 3-5 highest-impact stories 📱 Media Buying & Ad Platforms — platform changes, ad formats, bidding 📈 Analytics, Data & Martech — measurement, attribution, new tools 🤖 AI & Emerging Tech — AI for marketing only (filter hard) 👥 Social Media & Creator Economy — algorithm changes, monetization 🌏 SEA & APAC Market Intel — regional shifts, platform moves ⚖️ Policy & Regulatory Watch — privacy, antitrust, ad policy 🛒 Consumer Behavior Signals — purchase/discovery shifts 📌 Worth Monitoring — remaining notable items, one line each FORMAT: Slack mrkdwn. *bold* (single *), • bullets, <url|text> links. Use ─────────────── dividers. No markdown headers, no tables. Use Monday-Sunday date range in header. QUALITY: Curation before writing. 8 curated items beats 30 dumped. Every item must answer: "why should a media buyer care RIGHT NOW?" If the answer is no — skip it.

📋 Example Output

Actual output from June 11, 2026 test run. 16 curated items from ~400 articles across 30 feeds. This is what arrives in Slack:

📊 Performance Marketing Weekly Intel — June 4–11, 2026

🔴 Top Signals

• ChatGPT Ads Gets Measurement via LiveRamp — Server-to-server CAPI closes "can target but can't measure" gap. 1B users + IPO looming = real third option to Meta/Google duopoly. Read →

• Accenture Acquires Whalar ($600M+ creator campaigns) — US creator spend at $44B. Publicis bought Captiv8, now Accenture bought Whalar. Creator marketing has graduated to core media channel. Read →

• Claude Is Fastest-Growing AI Traffic Source (+386%) — Still tiny at 1.4% share (ChatGPT = 78%), but the growth curve is the signal. US 10 months ahead of EU/UK. AI referral traffic now 0.33% of all web traffic. Read →

───────────────

📱 Media Buying & Ad Platforms
• Instagram Reels post-view ads go global — All advertisers, Campaign Manager. Reels = 50% of IG time.
• World Cup in-game ads debut — First time in FIFA history: hydration breaks in 104 matches carry ad inventory. Fox didn't sell it at upfronts — value could rival Super Bowl over time.

📈 Analytics, Data & Martech
• Brand is the new backlink for AI SEO — WordCamp panel: AI search rewards citations over raw traffic. Practical: structured data + consistent brand positioning across platforms.
• AI referral traffic hits 0.33% of web — Up from 0.19% year ago. Prepare for AI as a real traffic channel.

🤖 AI & Emerging Tech
• AI "girlfriend ads" fuel MFA fraud wave — DoubleVerify: 243K AI-slop domains (2× YoY). Your brand ads may be appearing next to this. Recommendation: quarantine Google Exchange.
• Gemini adds Business Profile management — Direct connection to GBP. Likely a new ad surface coming.
• "Marketing needs AI outcomes, not pilots" — Gartner: most orgs have more AI activity than AI value.

👥 Social Media & Creator Economy
• Instagram "Your Algorithm" hits main feed — Users can customize topics. Mosseri: most won't bother, but AI drives all Meta engagement.
• Pinterest + Amazon Storefront — Social commerce infrastructure converging. 631M MAUs, 50% shopping.

🌏 SEA & APAC Intel
• TCS partners with Anthropic — India's largest IT firm goes all-in on enterprise AI. Signal: APAC enterprise AI spend accelerating.
• India-Oman CEPA signed — New trade corridor. India's economic diplomacy expanding.

⚖️ Policy Watch
• Meta deletes face-recognition from smart glasses after WIRED report. Privacy backlash still forces product changes.
• Japan visa rules threaten immigrant businesses — ripple effects on consumer economy.

📌 Worth Monitoring
• OpenAI files for IPO · Snap AR for World Cup · Google: 23% queries to open web · LadBible profit warning from AI traffic drain

💰 Costs

ItemProviderMonthly
VPS hostingTencent Cloud Lighthouse ($3/mo) or Zeabur ($3/mo)$3.00
AI API (4 runs × $0.25)DeepSeek V4 Pro$1.00
Webhook deliveryMake.com (free tier)$0
Feed readerblogwatcher-cli (open source, MIT)$0
OrchestratorHermes Agent (open source)$0
Total~$4.00 / month

🔧 Technical Setup (Summary)

# Install feed reader
curl -sL https://github.com/JulienTant/blogwatcher-cli/releases/latest/download/\
blogwatcher-cli_linux_amd64.tar.gz | tar xz -C /usr/local/bin

# Add feeds (30 total — grouped by output category)
blogwatcher-cli add '[Ad] Adweek' 'https://www.adweek.com/feed/'
# ... (29 more feeds)

# Create weekly cron
hermes cron create \
  --name "Weekly Intel Brief" \
  --schedule "0 9 * * 1" \
  --skill blogwatcher

Built with Hermes Agent + blogwatcher-cli + DeepSeek V4 Pro · Delivered via Make.com → Slack