Hey!

Welcome to Social Files—your no-BS guide to generating demand for your B2B product using social & content.

Hope you had a great weekend. I spent the weekend in London with my mom. Back in Barcelona now for a few days until heading back to the US.

Got a good one for you today. For the longest time, AI never really “stuck” in my writing workflow. I’d see all these posts on social about how people were using it - and I’d always feel like I was missing something.

That’s changed over the past 3 months. And today I want to walk you through all the real ways I’m using AI to write better content, faster. Examples and prompts included, of course.

Shall we?

🔎 DEEP DIVE

How I’m (actually) using AI in my content workflow

Select tools & prompts I use to save 5-10hrs per week on content creation

If we would have had this conversation back in 2023, I would have told you AI is useless for content creation. It just makes slop. It doesn’t have a meaningful impact on the best creators.

Now, I haven’t done a full 180 on you here. There is a ton of noise in the ‘AI for content creation’ market right now. A lot of current products are just vaporware thinly veiled behind a flashy demo.

That said, my view on AI has evolved. I’m surprised by just how much AI has found its way into my content creation workflow—outside of just Bluecast, the tool I built.

Today, I want to walk you through every way I’m actually using it. I’ll share real examples and a few prompt templates for you to steal.

My current AI content stack

If you're in a hurry, and don’t have time to read the full piece. No offense taken.

• ChatGPT—my ‘daily driver’ AI; also used for Deep Research.

• Claude—my primary tool for content editing and writing feedback.

• Bluecast—a cracked AI content writer for $29 a month.

• Wispr Flow—by far, the best AI dictation tool.

• Fireflies—my meeting recorder of choice.

• Perplexity—useful for finding specific stats and examples.

Ok, now let me walk you through how I’m using each of these. With prompts and examples for you to reference.

Wispr Flow for dictation.

Wispr Flow is a dictation tool, powered by AI. In short: you talk, text appears.

This is going to sound kind of funny, since dictation is such a ‘simple’ use case—but Wispr gave me my “aha” moment with AI.

I use it for everything now. I’ll draft up Slack messages with voice. I’ll “talk” with ChatGPT using Wispr. And of course, I’ll use it to draft segments of my content, as well.

Part of this newsletter was probably ‘written’ by me yapping into my laptop. What a time to be alive.

Wispr gets the formatting mostly right, most of the time—which is far better than other dictation tools I’ve tried.

I have a hotkey set up on my Mac (the ‘control’ button) that I can press to turn on my mic and use Wispr. It’s seamless.

To be clear: I have zero financial relationship with Wispr. It’s just so good I’ve been telling just about anyone who will listen about it. That said…if someone from Wispr is reading this…hit my line.

Now, let’s chat through a few AI tools I’m leaning on a lot in my actual writing workflow.

Claude as a Content Editor.

Writing Style Extraction

I don’t use Claude a ton for drafting—I use Bluecast for that. But I do find Claude very helpful for giving feedback on my writing, whether written by me or by Bluecast.

I have a Claude project set up with sample newsletters, sample LinkedIn posts, YouTube transcripts, and more loaded in as “project knowledge.”

When you first set this up, you’ll want to prompt Claude to extract your writing style and create a style guide for itself to use. You’ll need ~10 pieces of sample content in PDF format to run this. Once you have your sample content, use this prompt:

“I have a file with 10 excerpts from my writing. I want to extract common phrases, tone, sentence structures, or rhetorical devices from past writing. Once extracted, this would be static. I would love to implement multiple summary structures (e.g., listicle, narrative, key takeaways, provocative question-based, etc)? I wonder if we can extract this from our 10 excerpts. The length is pretty consistent (we can extract).”

I adapted this from Khe Hy’s piece, “How I got AI to write like me.”

Once you run this prompt, follow it up with this:

“What other information do you need from me to better understand and replicate my writing style?”

You want to fill in as many gaps as possible for Claude. After this, Claude will likely ask you more follow-up questions to get a better sense of how you write. For example, Claude asked me how I approach drafting my content.

I used Wispr to talk through my approach to drafting—way easier than typing it all out—and gave this response:

Then, Claude followed up with another question about my editing style.

And again, as if in a regular conversation, I gave it my answer:

This back and forth went on until I felt I gave Claude enough context to understand my writing style. I think I spent 30 minutes or so on this “conversation.” Once we were all set, I asked Claude to create a style guide based on my writing preferences.

Then, I uploaded that Claude-created style guide back into Claude as project knowledge.

When I first started playing around with AI 1-2 years ago, I could never get into it for content-related tasks. Explaining my writing style to AI felt like more work then doing the writing myself.

This changed when I realized I could just have Claude do that work for me. All I’d need to do is fill in context gaps. That’s also infinitely easier now with a tool like Wispr.

Identifying Logical Gaps

You can use Claude as a sparring partner who will poke holes in content. This use case is most helpful when writing about a subject you're less familiar in.

Here’s a simple prompt that will help you avoid sounding like an idiot on the social timeline

“Assume the role of a [category] expert. Based on what you know about [category], are there any logical gaps in this post?”

For example, I was writing a post for a client building software in the RevOps category. I have some surface-level knowledge of the industry, but I wouldn’t consider myself an expert.

Before AI, I’d use Google to fill in gaps as best as I could, and then cross my fingers that the post would pass client review. Now, I just ask Claude to poke holes in the content for me.

One of two things will happen:

(1) You’ll realize there are errors you need to correct.

(2) Claude will call something out, but you made that choice intentionally. That’s fine.

By the time a piece goes live, you can be confident that every choice in it was intentional. There are no unforced errors.

Content Ideation

You can also ask Claude for content ideas based on your ICP. In the spirit of keep this authentic, I will say I don’t use it a ton for this as of now. I find that the “come up with content ideas” use case is a lot more helpful for beginners than it is for seasoned pros.

It reminds me of how people use social media analytics.

Beginners and people who have no idea that they’re doing will spend all this time building out social analytics spreadsheets or signing up for expensive-ass software. In reality, most of the killers I know in content just run off of vibes and pattern recognition. In-platform analytics usually get the job done.

The fancy shit makes you feel like you're getting work done.

Same dynamic occurs when you asking Claude “come up with 25 content ideas.” Wow, so many topics! Yeah, but are they any good?

This is my current take. It might evolve, as I mentioned at the top of this piece. I will say, for less experienced founders getting into content, the ‘ideas’ can be helpful. We also built a content idea generation feature into Bluecast.

Speaking of which, let’s talk about how I used my own SaaS in my day-to-day.

Bluecast as a Content Writer.

I know I’m biased. But using Bluecast is like having a cracked Content Writer on your team, who only costs $29 per month.

I find myself using Bluecast to draft content when I’m in a rush between meetings. I still need to get content out. But I don’t have time to sit down and write for 1-2 hours.

I can take a newsletter I already wrote, paste the link into Bluecast, and then my cracked little content writer will churn out a ready-to-post LinkedIn piece.

I’m also sitting on a ton of YouTube content. So I can use that same workflow (there is a YouTube → LinkedIn template in Bluecast) to create LinkedIn posts from that material.

One other cool repurposing workflow: if you recorded a short-form video, you can use Bluecast to draft a caption for you. Easy. Just take the audio file, and upload it to Bluecast. It’ll do the rest ofr you.

Bluecast isn’t limited to repurposing, by the way.

When writing content from scratch in Bluecast, I like to approach it like I’m being interviewed.

I’ll pick one of the 50+ content prompts in the platform (more coming soon), and then use the audio recording feature to just yap through my answer. Bluecast will then take that audio recording, and transform it into a usable LinkedIn post.

I might make a couple of tweaks before publishing. But those tweaks now take me 30 seconds, not 30 minutes. The Writing Style feature we built into the tool makes it so the output is pretty damn close to your tone.

If you're a bootstrapped founder who can’t bring on a FT content hire or a good agency (like Compound) yet, Bluecast is a great option for offloading some of that content work.

ChatGPT as a Research Assistant.

ChatGPT is my daily driver. I use it the most of our any tools in my AI stack. It’s even earned a spot as a widget on my iPhone’s home screen.

I’ll use the 4o model to answer one-off questions. For example, just this morning I asked it: “Assume you're an expert in the software industry. What would "good" churn numbers be for a 'prosumer' SaaS?”

When I need to go more in depth, I’ll use the Deep Research functionality. My team at Compound will run a Deep Research report whenever we’re onboarding a new client.

Here’s the prompt we use. Go ahead and steal it if you want.

I'm a Content Strategist at Compound onboarding a client at [company]. We are a content agency, and will be primarily creating LinkedIn content for their founder, [name]. We help companies grow an audience on LinkedIn, get more distribution, and generate more inbound.

Here is the company’s site: [URL] Here is the founder’s LinkedIn profile: [URL]

Give me a research report which contains:

• An overview of what the product does

• An overview of the company’s ICP. Please detail the role(s), company size, category, pain points, and whatever other information may be relevant

• An overview of their current LinkedIn content approach. Does the founder post often? What topics are they posting about?

• Competitor analysis. Who are the main competitors?

• Examples of customer reviews

Based on the above, please also present a list of 25 LinkedIn content ideas that would resonate with the company’s ICP.

To get the most out of this functionality, and most of the other use cases mentioned in this piece, you need to go one step further. Ask follow-up questions, just like you're talking to an exceptional Research Assistant in real life.

Here a simple example of what that might look like. I had GPT pull a report for me on a new client in RevOps. Once it finished the primary report, I asked this follow-up.

“Thanks. I'm new to RevOps, can you give me an overview of the category for a beginner? What does it mean? What are some key terms?”

You’ll notice I haven’t mentioned Perplexity yet. I have a membership, but I don’t use it a ton right now. I’ll really only use Perplexity if I need to find a specific, one-off stat. ChatGPT is okay at this, but can hallucinate. Otherwise, GPT’s search function using the o3 model does most of what I need day-to-day.

Treat the robots like team members.

This is going to sound weird if you haven’t fully leaned in yet. But the best way to use AI tools in your workflow is to treat them like team members. Have conversations with them. Go back and forth.

And say ‘please’ and ‘thank you’ of course.

When you're having Claude edit a piece of content, ask it follow-up questions. Make it defend its points. Ask it what information it needs to complete the desired task. You can riff back and forth with Claude just like you would in a content review session with an Editor on your team.

I’ve found that using tools like Wispr makes this ‘flow’ easier. I can talk to AI in natural language, versus having to meticulously craft a prompt that may or may not work. Feels more like a conversation—less like a ‘prompting session.’

One parting thought for you. I don’t think it’s wise to stick your head in the sand anymore. The best content marketers are using it in their content workflow—not to outsource creativity, but as a force multiplier for it.

If nothing else, pick one use case from this piece and use it daily. Build the habit. Start now, and the 2027 version of you will be happy you did—assuming AGI hasn’t rendered humanity obsolete by then.

Can’t wait to use Bluecast to turn this pure, 100% human-made newsletter into 4-5 LinkedIn posts. The future is here.

🗃 FILE CABINET

Here’s my favorite marketing and business content I bookmarked this week.

• The best LinkedIn marketing strategy for 2025 by Tommy Clark 🎥

• Who Is Michael Ovitz? by Michael Ovitz 📚

• How to build a brand (full course) by Caleb Ralston 🎥

Check these out.

BEFORE YOU GO…

As always, thanks for allowing me into your email inbox every week.

More from Social Files:

• Read the rest of my essays

• Work with my agency in 2025

• Try my LinkedIn content writing SaaS

• Steal my founder-led content templates

Talk soon,

Tommy Clark