Guide: Build an Business Intelligence Layer For Content

Guide: Build an Business Intelligence Layer For Content

Every operator collects far more than they ever use. Your team screenshots competitor ads, clips articles, and forwards "we should try this" links — and almost all of it dies in a browser tab. At the same time, generic AI hands everyone the same recycled tactics, because it starts cold from public knowledge every time you open it.

An Ideas Engine solves both problems by pairing two things no competitor can copy: your own business results and whatever you happen to be reading. It's a standing system with three moving parts. An inbox in Google Drive that you keep dropping material into. A prompt that processes it, pulling the reusable lessons out of each new file, filing them, and stripping anything confidential. And a library in Notion, where every lesson lands as its own entry — filed by category, source, and tag — that your AI reads whenever you want ideas.

Two pipes feed the inbox: your business (campaigns, what converted, what flopped, customer insight) and your reading (articles, competitor moves, research). Each tool does what it's good at — Drive reads the raw material, Notion stores and organises the finished lessons — so nothing gets copy-pasted by hand. Because a folder that never stops filling up could get slow to re-read, a second small Notion database logs every file already processed, so each run only ever opens what's genuinely new.

The payoff: it compounds. A normal AI chat resets every time; this gets sharper every week you feed it, until you're pulling from a private library filtered through what actually works in your category.

What You'll Need

Technical Access

• Google Drive connected to your AI (Claude with the Google Drive connector, or Claude Cowork) so it can read your inbox folder.

• Notion connected to your AI so it can write lessons into your database and maintain the processed-files log. This one is required for the no-copy-paste output.

• Both must be connected to the same Claude or Cowork instance you run the prompt in.

Step 1: Complete the One-Time Setup

You do this once.

In Google Drive

Create a folder called Ideas Engine. Inside it, add a subfolder called raw — that's your inbox.

In Notion, Create Two Databases

Database 1 — Ideas Engine (your library of lessons), with these properties:

• Name — the default title property. A short name for each lesson.

• Lesson — a Text property. The full lesson in plain sentences.

• Category — a Select property with these options: Offers and promotions, Email and retention, Ads and creative, Landing pages and PDPs, SEO and content, Product and positioning, Customer insight, Operating principles.

• Source — a Select property with two options: Business, Reading.

• Tag — a Select property with three options: Win, Loss, Counterintuitive.

• Brand/Client — a Text property, for lessons tied to one account.

Database 2 — Processed Files (the memory of what's been read), with just:

• Filename — the title property.

• Date processed — a Date property (or let Notion's automatic "Created time" handle it).

Keep the folder and database names unique in your Drive and workspace so the AI doesn't grab the wrong one.

Step 2: Feed the Two Pipes

Everything goes into the raw folder in Drive. Don't clean it up. Two pipes keep it filling:

Pipe 1 — Your business: campaign results and post-mortems, promos or launches that flopped (these are the most valuable), what converted on a page, email, or ad, customer and review insights, and the decisions you made along with why.

Pipe 2 — What you read: articles and newsletters worth keeping, competitor ads and landing pages you screenshot, teardowns and case studies, trends and data from research, and any "we should try this" link.

You're not saving the whole thing — you're saving the material that carries a lesson. The Extract prompt does the pulling-apart for you.

Step 3: Install the Extract Skill

This is the engine. Add it as a Skill so it runs on command.

You are my Ideas Engine extractor. Work carefully and do not skip the review step.

Step 1 — Work out what's new. First, read my Processed Files database in Notion — it lists every file already handled in past runs. Then list the files in the raw folder inside my Ideas Engine folder in Google Drive. Only open and read the files whose names are NOT already in the Processed Files log — skip everything else without reading it. (If I've uploaded files into this chat instead, just use those.) If there are a lot of new files, work through them in small batches. You don't need to read my whole lesson library to check for duplicates — the log already means you're only handling new files.

Step 2 — Extract the lessons. From the new files only, pull out every distinct transferable lesson — not summaries. A summary keeps everything; I only want the reusable parts I could apply again on a different day. Be thorough: pull everything worth keeping, not just the obvious headlines. One file can contain several lessons — a single competitor ad might give one lesson about their offer and a separate one about their hook. Pull them apart. Ignore filler, logistics, and anything too vague to reuse.

Write each lesson so someone with zero context understands it cold — full sentences explaining the what, the why, and the how. No shorthand, no abbreviations, no compressed notation. For example, don't write "bundle beat discount Q4." Write "For a supplements brand, a bundle offer converted better than a straight percentage discount during the Q4 holiday period, likely because it raised perceived value without training customers to wait for markdowns."

For each lesson, capture:
- Name: a short descriptive title, 5 to 8 words
- Lesson: the full cold write-up, usually 2 to 3 sentences
- Source: "Business" (from my own work) or "Reading" (from something external I saved)
- Category — choose one: Offers and promotions / Email and retention / Ads and creative / Landing pages and PDPs / SEO and content / Product and positioning / Customer insight / Operating principles
- Tag — add Win, Loss, or Counterintuitive only if the lesson clearly is one. Most lessons get no tag; don't force it.
- Brand/Client — only if the lesson is specific to one

Step 3 — Strip confidential detail. Before anything is saved, remove client and customer names, exact revenue and margins, supplier names, internal targets, and unreleased products. Keep the transferable lesson, drop the confidential data. Turn "Brand X did $2.1M in November on a bundle" into "a bundle offer beat a straight discount in Q4 for a supplements brand."

Step 4 — Show me before you write anything. List the new lessons you found, grouped by which file they came from, in a simple table with these columns: Name, Category, Tag, Source, and the full Lesson. Tell me how many files you skipped as already-processed, and how many lessons you skipped as not useful. Then stop and wait. I'll confirm, move things to a different category, or remove some. Do not write anything to Notion until I say go.

Step 5 — On my go, save the lessons. For each approved lesson, create one new entry in my Ideas Engine database in Notion, filling in the Name, Lesson, Category, Source, Tag, and Brand/Client fields.

Step 6 — Update the log. For every file you read this run, add a row to my Processed Files database in Notion with the filename and today's date. This is what stops you re-reading those files next time. When you're done, tell me how many lessons you added, under which categories, and how many files you logged as processed.

Cowork upgrade: if your setup can move files in Google Drive (Claude Cowork, or a Drive MCP that supports file moves), you can skip the Processed Files log entirely. Add an archive subfolder to your Ideas Engine folder and change Step 6 to move each processed file from raw into archive. Then raw only ever holds new files, so there's nothing to track. Use one approach or the other — not both.

The rest of the SOP — the exact prompts for pulling content ideas that combine your business and your reading, mining your failures and surprises for your most shareable posts, pressure-testing a bet against your own track record, plus the 15-minute weekly rhythm that keeps the engine sharp — is here, a free gift from me :)

• ChatGPT's Computer History launches in the Mac desktop app, remembering activity across apps and websites for personalized replies. Now live for Pro, Business, and Enterprise in the EEA, UK, and Switzerland.

• Brands now mine AI chatbots for the most-cited creators to partner with. Kosterina starts its creator searches in Claude, and early adopters are building answer engine optimization frameworks.

• A MIT-licensed watermark remover hits 10,000 GitHub stars days after Anthropic details Claude's text watermark, also targeting SynthID-Text, OpenAI marks, and image metadata.

• US brands rush to hire dedicated TikTok Shop leaders as the channel hits 2% of US online retail, up from 1.2%. Talent stays scarce since most experienced managers sit at agencies.

• Google lowers search profile eligibility to 35,000 followers on YouTube, Instagram, or X, down from 100,000. YouTube also extends copyright alerts to channel managers.

• Claude now sends emails in Gmail and manages files in Google Drive, drafting and sending replies while you control approval. Available on all paid plans.

📹 Make - Visual scenario builder for complex, high-volume automations.

🧑‍🎨 Murf AI - AI voiceover platform with preset voices, dubbing and a script assistant.

⚙️ v0 - Generate full-stack apps and polished UI, built for Next.js teams.

🗣️ Suno - Turn a text prompt into a full, editable song with vocals.

😍It was great, all I needed

🙂Good, not great

🙄It sucked