Our weekly newsletter, the 'Trading View Digest' didn't exist before we built this AI newsletter workflow. Not because we didn't want to create it. We had the expertise. We had the insights. We had the data. What we didn't have was the time.
Our Monday morning meetings is a gold mine of market intelligence. Industry-leading analysis from experts deep in the numbers. But turning that into a polished, fact-checked newsletter that goes out the same day? That would have required resources we simply didn't have. The gap between insight and publication would have been measured in days, not hours. By which point, the market had moved on.
Then we started experimenting with newsletter automation with AI. Not to replace our team. But to compress the time between insight and publication. To handle the structural scaffolding so our team could focus on the thinking that actually matters.
What we built not only works, it produces our most popular newsletter. A Monday morning meeting becomes an afternoon newsletter in our clients' inboxes. Same day turnaround. Defensible facts. Editorial polish. Without the resource headcount, that would have made it impossible.
I want to walk you through exactly how we do this. Because if you're in marketing or content creation, this workflow might just show you what's suddenly possible when you stop thinking of AI as automation, and start thinking of it as leverage.
Every Monday morning, our data and market analysis team gathers in Microsoft Teams. They're deep in the numbers. We're talking about insurance market pricing movements, competitive repositioning, quotability shifts, and the subtle tactical maneuvers that tell you what's actually happening.
These meetings are gold. They're where context lives. They're where someone says "actually, what's really interesting here is…" and suddenly a data point becomes a narrative.
The problem? All that insight is locked in a one-hour Teams recording and a PowerPoint deck that sit in a folder somewhere soon to be forgotten.
That's where the AI workflow begins.
Here's the unglamorous but essential bit. Immediately after the meeting wraps, I download two things. The Microsoft Teams call transcript (a complete text record of everything said) and the accompanying PowerPoint deck from our Data Services team shared during the meeting.
Both get converted to PDF format. This matters more than you might think. PDFs are the lingua franca of document AI right now. They preserve formatting, they're reliable, and they're universally supported by the tools we use next.
Why convert at this stage? Because the next tool in our stack is NotebookLM, and PDFs import cleanly without the formatting headaches you get with native Microsoft formats.
NotebookLM is where the heavy lifting happens. Google's research notebook tool has quietly become indispensable for our workflow.
I create a new Notebook project and upload both PDFs as sources. This is crucial. NotebookLM indexes these documents, creating what Google calls a "source notebook." Every insight it generates is rooted in your source material. Every statistic it uses can be traced back to where it came from.
I then generate two specific report types that match our editorial needs.
The Market Analysis report gives me structured breakdowns of the week's movements. Pricing trends. Competitive dynamics. Strategic shifts. It's clean, logical, and facts-grounded because NotebookLM has literally pulled it from the sources you provided.
The Executive Briefing then distils that into the kind of forward-looking narrative insight that makes our newsletter distinctive. What does this mean? Why should our readers care? What's changing beneath the surface?
Both reports get exported as PDFs.
The insight here: This isn't a black box. You can see exactly where NotebookLM sourced each point. You can challenge it, using the chat function. You can refine it. It's a foundation, not a final product.
Now the story really unfolds. I take those NotebookLM exports and upload them to a Claude project I've set up at Claude.ai.
This project includes three critical reference files that shape everything that comes next.
First, our Editorial Standards. This is our stylistic north star. It specifies that we write with a conversational but sophisticated tone. We use metaphorical language. We frame insights in broader context. We maintain narrative flow while incorporating specific metrics. Every newsletter tells a story, not just reports facts.
Second, our em-dash rule (yes, this genuinely matters). We replace em-dashes with periods or commas. It's a small formatting convention that's part of our house style, and it's non-negotiable for consistency.
Third, examples. Four previous editions of the Trading View Digest. Real examples of what we actually produce. These are more powerful than any style guide because they show, not tell.
I upload the latest NotebookLM-generated reports alongside these reference materials, then I use this Claude prompt.
My Actual Claude Prompt:
"You are an expert financial intelligence editor specializing in UK insurance market analysis. Your task is to create this week's Trading View Digest newsletter using the attached NotebookLM market analysis and executive briefing reports.
Reference the Editorial Standards document for tone, structure and style. Use the previous newsletter examples as templates for structure, tone of voice, and the balance between data and insight.
The newsletter should have the following structure:
- An opening paragraph that connects this week's market movements to broader context or seasonal themes. Avoid generic openings. Make it distinctive.
- Motor Insurance section covering key market dynamics, specific movements with data, and market implications.
- Home Insurance section with the same structure.
- A brief conclusion that synthesises the week's themes and looks forward.
- Space for a featured content link (I will populate this separately).
Remember: anonymise all brand names using descriptive terms rather than company names. Replace all em-dashes with periods or commas. Maintain narrative flow while grounding every claim in the data provided. The goal is sophisticated analysis that helps readers understand what's actually happening in the market, not just what moved.
Generate the newsletter in Word document format with professional formatting."
This is where the alchemy happens. Claude doesn't just reorganise the NotebookLM reports. It synthesises them. It finds the narrative thread. It applies editorial judgment. It makes decisions about what matters and why.
Out comes a complete draft newsletter in Word format.
Here's where most newsletter creators would just hit send. We don't.
Because the newsletter is anonymised (we don't name specific insurers), fact-checking can get tricky. A colleague might ask "What's that data point about price increases in the north? Where does that come from?"
This is where the workflow shows its real power.
I go back to the original NotebookLM notebook and use its chat function. Simple prompt.
"Fact check this bullet point: [specific claim from the newsletter]"
NotebookLM returns the exact reference. The original transcript line. The slide it came from. The context.
This is your audit trail. It's not perfect, but it's thorough enough that you can defend every claim you made. In an industry as regulated and scrutinized as insurance, that matters enormously.
The draft Word document goes to OneDrive. Our team reviews it. We make refinements. Sometimes we get feedback like "can you make this section feel more forward-looking?" or "this paragraph needs more specificity."
Because Claude generated it with reasoning and structure, not just pattern matching, we can edit it meaningfully. We're not fighting the tool. We're refining it.
Once we have sign-off from the team, the newsletter is ready for distribution.
The final step is almost anticlimactic. We send the newsletter via HubSpot to our client and prospect audience.
But here's the thing. By the time it goes out, it has been through multiple layers of refinement and verification. It's grounded in source material. It reflects our editorial voice. It carries the weight of our expertise. The open rate and click-through rates we are getting is a testament to this.
I want to be direct about what's happening here, because I think it matters for anyone considering similar processes.
We're not using AI to replace our team. We're using it to amplify them.
The meeting still happens. The experts are still there. The thinking is still human. What AI does is compress the time between insight and publication. It handles the structural scaffolding. It applies editorial patterns consistently. It creates an audit trail.
The result is that our team spends less time on formatting and structure, and more time on the thinking that actually matters. More time asking "is this the real story here?" Rather than "how do we organize this data?"
That's the genuine win.
If you're thinking about adopting something similar, here are the things we learned.
1. PDFs are your friend. Convert everything to PDF before you import it into multi-document AI tools. It normalises formatting and ensures consistent processing.
2. Source material is everything. Tools like NotebookLM that let you verify where claims come from are non-negotiable. Your credibility depends on it.
3. Editorial standards matter more with AI, not less. Because your AI tool will be consistent, your house style guidelines become even more important. They're what makes AI output sound like you, not like a generic AI tool.
4. Don't skip the human review step. Fact-checking through NotebookLM's source verification takes maybe 10 minutes. It's the difference between a polished, defensible piece of content and something that might cause problems. Don't let an AI hallucination kill your content machine or damage your brand's credibility.
AI Agents are becoming ever more powerful. I experimented with ChatGPT's Agent mode, but it failed to carry out the workflow autonomously. Things move fast in AI; maybe next week or in a month, it could. It's definitely something I will come back to review.
But honestly? The core workflow we have now works. It's reliable. It's defensible. Allows us to create something that adds real value to our prospects and clients by sharing genuinely useful market intelligence.
And it lets our team do what they do best. Think. Analyse. Find the story beneath the data.
The AI handles the rest.