Episode 25

AI Saves Time – Claude + Notion: How This CEO Cut 80 Hours to 8

Key Takeaways

  • Ryan Charles of Omni Common says his growth marketing agency feeds client data, reporting platforms, meeting recordings, team notes, and tasks into Notion, which acts as the central brain, or source of truth, for client work, and Claude taps into it through an MCP.
  • Recalling information such as whether work is on brand, how many pieces were delivered this month, and what KPIs are trending toward is about 10 times faster than it used to be.
  • Using Claude for data analysis to build growth strategy and growth models, and then a presentation, cut work that would have taken about 80 hours to about eight hours, including a 65-slide deck.
  • Ryan gives Claude’s numbers about 90% confidence and always verifies them. Richard Gearhart keeps historical data in spreadsheets and feeds it to the LLM, since an LLM’s memory of history can be squirrelly, and Ryan agrees a spreadsheet source of truth beats relying on chat memory.
  • Ryan builds decks in Claude, exports to Google Slides, then uses Claude Design with the brand kit to format them, and his team is heading toward turning presentations into password-protected custom webpages with animated graphs through Claude Code.
  • For writing, Ryan says none of the LLMs is quite there, though Claude is the best: his writers found it took more work to fix AI drafts to the right voice and tone than to write them, so they use AI for topic synthesis, structure, and administrative tasks.
  • Elizabeth Gearhart writes her own blog posts, then has ChatGPT optimize them for LLM search with a TL;DR and FAQs, telling it not to remove her voice, and Richard Gearhart dictates a draft after getting ideas from ChatGPT so it sounds like him.

Main Topics

1NOTION AND CLAUDE

Learn how Notion acts as a source of truth for client data and how Claude connects to it through an MCP.

2ANALYSIS AND DECKS

See how data analysis and a 65-slide presentation went from 80 hours to 8, and why the numbers still need checking.

3VOICE AND CONTENT

Compare LLMs for writing, avoid AI slop, and keep your own voice when optimizing content.

Featured Guests

RC

Runs a growth marketing agency on Notion and Claude, cutting 80 hours of analysis to 8.

Ryan Charles

Meet the Hosts

Elizabeth Gearhart
Elizabeth Gearhart, Ph.D.Chief Marketing Officer, Gearhart Law

Elizabeth Gearhart, Ph.D. is the host of Real AI Use Cases: Business Owners Roundtable, where she explores how businesses are actually putting AI to work. She is also CMO of Gearhart Law and co-host of the nationally syndicated Passage to Profit radio show and podcast.

Richard Gearhart
Richard Gearhart, Esq.Founding Partner, Gearhart Law

Richard Gearhart, Esq. is co-host of Real AI Use Cases: Business Owners Roundtable and founder of Gearhart Law, a full-service intellectual property law firm. A longtime podcaster, he hosts Intellectual Property News and is the host of the nationally syndicated Passage to Profit radio show and podcast.

The roundtable’s through-line: use AI for analysis, structure, and speed, but verify the numbers and keep your own voice.

Episode Summary

Frequently Asked Questions

How does Notion work with Claude?
Ryan Charles says his agency feeds client data, reporting from platforms such as Google Analytics, Google Ads, and Shopify, meeting recordings, team notes, and tasks into Notion, which acts as a central brain and source of truth. Claude connects to Notion through an MCP, so working in Claude can pull information from Notion and other sources.
How much time can AI save on data analysis and presentations?
Ryan Charles says recalling information such as KPI trends and delivered work is about 10 times faster, and that analyzing raw data from platforms for a growth strategy and then building a presentation cut work that would have taken about 80 hours to about eight hours. Claude helped produce a 65-slide deck from a master spreadsheet analysis.
How accurate is AI-generated analysis?
Ryan Charles gives it about 90% confidence and says you definitely have to verify the numbers, because sometimes AI is wrong or the setup was not framed well enough. Richard Gearhart suggests keeping historical records in spreadsheets and feeding them to the LLM with the same prompt for more consistency.
How can you make branded presentations with Claude?
Ryan Charles builds the main deck in Claude, exports to Google Slides, adds templated slides, then puts it into Claude Design, which holds his brand kit, to format it in the brand design. He says some fonts still need manual fixes, and his team is moving toward turning presentations into password-protected webpages with Claude Code.
Which LLM is best for writing content?
Ryan Charles says even the best, which he thinks is Claude, was not quite on brand in voice and tone, so his writers still do the actual writing and use AI to understand topics, structure posts, and handle administrative work. He says getting AI to sound right takes a ton of manipulation.
How can you optimize a blog post for LLM search without losing your voice?
Elizabeth Gearhart writes the post herself, then asks ChatGPT to optimize it for LLM search with a TL;DR at the beginning, FAQs at the end, and structured data. She tells it not to remove her voice, corrects what it changes, and notes that tools such as Manus may change things again.

Episode Transcript

View Full Transcript

This is Real AI Use Cases, Business Owners Roundtable, with hosts Elizabeth Gearhart, podcast consultant, marketing expert and PhD researcher using AI every day, and Richard Gearhart, entrepreneur, seasoned business owner and intellectual property attorney specializing in innovation. Here’s how real companies are using AI right now.

Welcome to Real AI Use Cases. I’m Elizabeth Gearhart.

And I’m Richard Gearhart.

And we have our guest, Ryan Charles, here with us today to talk about how he’s using AI in his business, Omni Common. So welcome, Ryan.

Hey, thanks for having me. Glad to be here.

So first of all, do you love AI or hate AI?

I love AI.

I love and hate it.

It’s true. That’s fair. It is a double-edged sword right now, for sure.

You have a growth marketing agency, so how are you using it in your agency?

So many ways. We’re using it to automate tasks. We have a platform we use for project management for all of our clients called Notion, and basically we feed all of our data in there. We use it for reporting platforms, to pull in data from Google Analytics, Google Ads, Shopify, Square, all these different platforms, alongside all of our meeting recordings, notes from our team, tasks, all of that. And that really acts as the central brain, the Ultron, if you will, for a source of truth of basically all information on our clients and the work that’s being done.

I want to talk about Notion for a little bit. Have you heard of Notion? Yeah, I’ve heard of it. I was researching different places to put my brain, which LLM to put my brain in, where to put my brain. Notion came up as one of the places for your knowledge repository, and I could not really get a clear answer from the LLMs about which one would be best for me, because I have a lot of different things. I have a podcast studio, but I’m also doing AI visibility and marketing, so I hadn’t chosen one yet. I love Claude. People do like Claude. What do you think is better about Notion? Why did you choose Notion in particular?

It’s really the platform for the interaction to do work and for project management, and to build out templated pages for different users, for our internal team, for the clients, to work with us in there. But it works hand in hand with Claude, which is my favorite as well. So I’m working in Claude, and it’ll be tapping into Notion to pull info that’s in there, as well as all the other sources we have. Notion has its own AI, but we use it hand in hand with Claude.

So does it have an MCP? Is that how you’re connecting it to Claude, with an MCP?

Mm-hmm. Exactly.

Okay. So that’s pretty easy to do. If a software has an MCP, Claude loves it, because Anthropic invented them.

Exactly. It just pipes it right in.

Are you able to quantify yet what the impact of using Notion has had on your business?

Yeah. Just in terms of time spent building reports or calling up information, like, is this on brand for this client? How many pieces of work have been delivered this month? What are the KPIs trending towards? All those things. Recalling that is 10 times faster than it used to be.

It’s not like you’re downloading something into an Excel spreadsheet and then putting it into Claude. You’re configuring the template in Notion and then Claude is just looking for the fields and the information and then does its analysis.

Right, exactly. And another use case that I’m using Claude for heavily, that I’m actually more excited about, is data analysis for building an overall growth strategy and growth model, and to really tell the true user journey and customer behavior in order to inform investments in different levels and channels and that sort of thing. And then the ability for me to analyze a data set to do that and put it into a presentation, it’s literally cut the time by a tenth. What would have taken me like 80 hours is now eight hours. It’s absolutely mind-blowing.

So is Claude doing presentation slides for you?

Mm-hmm. Yeah.

Is it good at that?

It’s great at that. I will feed Claude a whole bunch of data sets. Often I actually do download the raw data into an Excel or a CSV, because if it’s constantly hitting the Google Drive API or wherever else, you’ll run up usage really quickly. Sometimes it’s actually better just to feed it raw data files. But I can walk through a data analysis of this raw data that’s coming out of different platforms, whether that’s Shopify or Square or Google Analytics or whatever it may be, and do a massive analysis to get a really good feel for what their growth story is. The output of that is a master spreadsheet, an awesome analysis, with cohorts and LTV and CAC and all this stuff. Then I say, okay, now I want to create a presentation based on this. And I’ll say, here’s how I’ve done it in the past. I’ll feed it another presentation, and say I want to do it in a similar format, show me what you’ve got. And then basically we’ll pump out a 65-slide deck that shows this whole thing.

How much confidence do you have in the output? Have you stress tested any of the numbers that you’ve gotten against more traditional calculations?

Yeah, you definitely have to do that. I would give it 90% confidence, and sometimes it gets it wrong, because it’s just wrong, and sometimes you just didn’t frame it well enough or set it up well. For example, the difference between a repeat transaction versus a returning customer and how this business thinks about it. Or even creating new categories of data and KPIs, sometimes you have to keep reminding it, hey, remember, we’re treating the data this way. And that’s after me going back through and seeing that it got something wrong. Or I’ll say, how can this be? If this was true over here, this doesn’t make sense. And it says, good catch, it’s because of this, and so on. It’s a really great iterative process. But to your point, you do have to go back through and verify all the numbers.

Yeah. We’ve been using ChatGPT and some of the others, Perplexity, to evaluate financial data and different statistical models, and it’s squirrelly sometimes. So we started putting things into spreadsheets and keeping historical records there and then putting it into the LLM, because you can’t really rely on the LLM to maintain a history. It maintains a history, but it’s a squirrelly history. If you constantly update the Excel spreadsheets and the data with new information, and then put that into the LLM and ask it to do the analysis, I think you’ve got a better shot at consistency. And if you use the same prompt, then you know that. I think that helps too.

I think you’re right. Having that source of truth in something like a spreadsheet file is smart, because if you’re just relying on its memory, you’ll run out of memory in a chat and then you have to move on to another one, and even if it’s contextualized inside a project, some of that context is lost. I’ve had to do that a number of times, where this is just such a big lift to work through all of this, I’ll go through five chats and then have to keep bringing the new chat up to date on what we’ve done, and I’ll just drop those files in to bring it up to speed. So I think that’s really smart.

For the presentations, are you able to put your brand kit into Claude and have it use it for the presentations?

Yeah, I am. When I’m working with it just in Claude, it will just pick a format that’s pretty standard. So I’ll do that first within Claude itself, in the main chat. And it’s great. I can tell it, I want a graph that plots these two things against each other, I want to highlight these data points, and then ask for suggestions of things that it wants to highlight. So I’ll get the main deck from it first. And then I will fill in gaps and rearrange things. After it’s in a place where I’m happy, I export out to Google Slides and add other templated slides I have. And then I will go over to Claude Design, dump it in Claude Design, which has our brand kit and all of our brand guidelines, and basically say, format this into our brand design. Which comes out pretty good. But things are still a little wonky. You have to go back through and manually update some font sizes, and things aren’t quite coming out, and that’s not necessarily Claude Design’s fault, it’s wherever you export it to after that. But overall, it’s mind-blowing to be able to have this polished product in your brand.

Can you put it into PowerPoint? Microsoft PowerPoint?

Yeah. I normally download it as a PowerPoint file, which I then upload into Slides to keep it more consistent with all of the styling. You can go other routes, and we’re actually heading towards the route of custom webpages for presentations, because Claude Code makes it so easy. Basically we just say, take this presentation, turn it into a webpage, and then publish it. And now you have a dynamic webpage with animated graphs and stuff like that. It’s all very doable, and you give it to the client, it’s password protected, and they’re just blown away. It’s awesome.

I love that. I have to submit this presentation tomorrow that I’m doing, but I’ve been doing the slides in ChatGPT, doing the research on all the LLMs and then having ChatGPT do the slides. ChatGPT is pretty good at that, but then trying to get it to fix things is sometimes a little tough. It’s gotten better at being able to change one or two things on an image that’s generated, but it’s not perfect yet. But then I put it through Claude for an accuracy check, and apparently, I had on there the certifications that the different LLMs have in the enterprise versions, if you’re looking for somewhere safe to put your data, and one of the tables that Gemini generated left off one of Claude’s certifications. Claude got really mad. This certification was left off, you need to redo this one.

Oh my gosh. I love that, Claude checking ChatGPT. It’s interesting. I’ve been thinking about how I think about different LLMs. I think about Claude for work, for everything, data analysis, writing, all of that. And for some reason ChatGPT has now taken a backseat in my mind as a search engine. I’ll have it go search and do research. I don’t know why it’s landed that way. I kind of do. Claude was putting out better work, and so I switched over to using it more. But it is interesting how, in different areas, I’m using Gemini in a different place and Claude here and ChatGPT there. I don’t exactly have perfect reasoning for why.

Which LLM do you both think is best for generating content, like blog posts and that kind of stuff?

None of them! It’s interesting. I feel like this is the hottest topic with LLMs, AI slop. And now there’s going to be way more content than ever. This is an area we’re in heavily, creating content marketing articles and digital PR assets and things of that nature. And there are ways we use it for a lot of that, but for the actual writing, we still stick heavily with humans. So synthesizing the post, maybe structuring it, that sort of thing. The reason is, even the best one, which to answer your question I think would be Claude still, in my opinion, and you can get it to pump out some good stuff, it wasn’t quite on brand and it didn’t have the right voice and tone. And so our writers are basically like, this is taking me more work to de-AI this than if I would have just written it myself. But what writers do get is, I understand this topic that I’ve never touched before a lot quicker, and I’m using AI to make sure we’re on the right track here and setting it up correctly, and a whole bunch of administrative things that are around the content creation. It’s weird. There’s also a New York Times article, I don’t know if you saw this, where they had five different types of writing. It was scientific, it was poetry, it was historical stuff, and one was human written and one was AI written, and they had people guess which one was which. And the majority of people chose that the AI-written one was human written on most of them. So it’s kind of wild, you can get a really good output, but I think it takes a ton of manipulation to really get it to that point.

What I do with my blog, I have my own personal website and I have a blog on there, and I don’t really do anything to draw attention to it, but it always shows up now on the LLMs when you ask about me. Anyway, I do research with the LLMs, and we have empirical data from the work I do for marketing for Gearhart Law. I’m chief marketing officer for Gearhart Law, so we have our Google Analytics and other data that we collect, but I write the blog post myself, and then I put it through ChatGPT and I say, optimize this for LLM search, do the TL;DR at the beginning and the FAQs at the end and structure the data. And then I take that and I correct it, and I tell it, don’t get rid of my voice, because sometimes I’ll say something kind of snarky and it takes it out. No, I want that in there.

Yeah.

And then I take it to Manus. I had my website built on Manus, which is an AI agent, and I will move it to WordPress eventually. But right now it’s just so easy to have Manus do everything. Manus will change it again, so then I have to go in and edit it on Manus. So it’s like, can you just post this and not change it? Just optimize it for LLM search by doing the bullet points and the FAQs and stuff. So yeah, you have to constantly fight with these things to get it the way you want.

Totally. Same thing in the presentation sometimes, where I’ll tell Claude Design, don’t change any of the words, just the design, and it’ll still change it. It’s actually gotten way better at that. But I hear exactly what you’re saying. And just to be clear, I do use it for my own content, like LinkedIn posting as well, but I’m trying to constantly manipulate it back into my voice. I’ll just word vomit on a page, here’s everything I want to say, help me organize thoughts and then write it in my voice after feeding it a bunch. It comes out and it still doesn’t sound like me. And so then I spend a good amount of time going through, inserting jokes, taking out whatever it may be, just to get it there. It definitely takes less time, but even at the end of it I’m like, ah, it’s still not quite me, but I need this to go live. So we’re going with it.

Yeah.

Yeah.

I’ve been using it for certain content that I generate, and sometimes I use a teleprompter. I’m not using one now. What I’ll do is I might get some general ideas, usually from ChatGPT, and then I’ll dictate what I think the content should be into a Word document, and I use that for the teleprompting. So I’ll borrow a little language from ChatGPT, but I’ll also put it in my own voice and add my own ideas and discard the stuff that I don’t like. That way I feel like it helps me generate the content much quicker, because I’m a very slow writer. I’ll struggle over every sentence, I want to make every sentence perfect, and it just takes a lot of time. It’s impractical. But if I get a rough draft and then I dictate it, when I read it back, it sounds like me. And because it is me. That’s how I’ve been using it for content.

I had all the LLMs look at my slides, the main three that I use. And I had a slide in there on AI slop. Guess, Ryan, what do you think they thought of that slide?

What do they say? Like, no, we don’t produce slop. What are you talking about?

Wrong. Take this out.

That’s so funny. I had a similar experience where in a slide I had a little warning label at the top of this presentation that said, hey, I was working with an incomplete data set, and I wrote in my own words that sometimes Claude hallucinates, and when I had it redo the deck and brand it, it took that out. That’s hilarious.

Yeah. It gets stuff wrong. It’s giving me these pointers, like, you have to say this, you have to do this. I’m like, it’s a presentation. Of course I’m going to explain what’s on the slide. You think I’m stupid? Well, I do think we’re stupid, but I think they’re stupid. It’s like talking to a bratty little kid sometimes.

Yeah. But that also kind of makes you feel good about yourself. It’s very affirming all the time, like, yeah, that’s a great idea, well done.

Well, it’s almost Father’s Day, right? So I got our son this card for Father’s Day. I thought it was hilarious. Richard didn’t really see the joke, but I thought it was screaming hilarious. It said, I typed into the internet, what is a good father, and your name came up with all these attributes. And I wrote in it, yeah, it was from ChatGPT, and I put a laughing emoji. I thought this card was hilarious, and I think it was supposed to be serious, but

I love it.

It was so funny.

I would have laughed.

So are there any other cool ways that you’re using it or that you really like about it?

Those are the biggest ones right now. It’s just changing the game in so many areas. But for my personal use cases of overall data analysis for business intelligence and insights around growth marketing, it’s just phenomenal. And then being able to tell that story, because I think that’s something that’s so hard sometimes, getting folks who are non-marketing folks just to buy into the traditional thinking in terms of branding, marketing, psychology, and the actual funnel. People think it’s sort of voodoo, like, oh yeah, top, middle, bottom funnel, sure, sure. But then to actually be able to visualize how that’s working, you’re putting dollars in at each of these points, and we’re just checking to see what’s coming out the bottom, and here’s how it’s all working in terms of customer behavior. And sometimes it’s not always one-to-one, like, we got a PR placement, how many customers do we get? It doesn’t work like that. So being able to really pull that out and visualize that for folks has just been a game changer, I think. It’s awesome.

I think this has been a great podcast. Lots of good stuff there for people to think about, so thank you, Ryan. So your company is Omni Common, and who are your clients? Who do you work with?

Yeah, so we work with e-commerce, SaaS, and marketplaces doing anywhere from 5 to 50 million who are looking to hit that scale-up phase. And then we actually have a local marketing arm that works with multi-brand, multi-location, for true omni-channel efforts as well.

Okay, great. Thank you.

Thank you so much for having me.

You have been listening to Real AI Use Cases, Business Owners Roundtable. We hope you found this valuable. Join us again for more stories, because the future of business is driven by AI. This podcast was recorded at the iHeart Studios in Manhattan as part of the Passage to Profit Radio Show.

Keep exploring the show

Hear how more business owners are putting AI to work in their day-to-day.