Summarize with AI
This transcript has been lightly edited for clarity and legibility.
Opening Remarks
Gina Ramsey: We’re having a few technical difficulties because Coral is not able to make it today — Katie was kind enough to pinch-hit, so bear with us as we try to figure out the tech here. I see a lot of friends in the room: Allison, Ashley, Kat, Lauren, Michael, Michelle, Vicky. We have tons of people registered for this, so hopefully we’re going to give you all the information you can possibly want. We’ll give it a few minutes to start, and then Katie will kick us off. Coral’s child is sick and told us this morning, so this is a literal last-minute switch for Katie — give us all a little bit of grace on this.
[Pause while attendees join.]
G.R.: Stacy Smith just showed up — we’ve got a lot of friends in the room, this is exciting. All right, Katie, why don’t you get going, because we have a lot of material to cover.
Welcome and Introductions
Katie Meixner-Croft: Yes we do. I want to start by welcoming all of you — thank you so much for your time today. We know your time is valuable, which is the entire point of being here: we want every minute you spend with us today to pay you back in time saved later.
A little housekeeping first: this session is being recorded, so if you need to duck out for any reason, or want to go back and rewatch something, it’ll land in your inbox within the next couple of weeks. We’re setting aside time at the end for questions, but if something comes to mind while we’re talking, go ahead and drop it in the chat — we’ll answer a couple live as we go and circle back to the rest at the end.
A quick introduction: I’m Katie — I’m the Digital Marketing Operations Manager here at Pink Dog Digital, and on a practical level, a lot of you probably know me as the person who takes care of your SEO and social media. I’m here today with Gina, of course, our resident AI expert — I’m excited for her to share some of the things we’ve been using it for.
One more thing up top: I want to emphasize that this isn’t just a marketing webinar. We’ll cover some marketing use cases, because that’s what we live and breathe on a daily basis, but the bigger goal is for this to be useful to anybody who wants more time back in their day. If you’re in operations, if you’re managing a team, if you’re doing your own books — a lot of what we’re showing you today still applies.
The Problem: Not Enough Hours in the Day
K.M.: So here’s the problem we’re trying to tackle. I’m sure every single person here today — taking a little liberty in making that assumption — does not have enough hours in the day. You sit down at the start of the day wanting to do strategic planning, wanting to do the big work that’s going to grow your business, and you end up three hours deep in your inbox, or redoing a spreadsheet, or putting out a small fire that wasn’t even on the table this morning.
It’s not just a marketing problem, though — if you run a small business, it shows up everywhere. Here are some examples we thought about while prepping this webinar:
- writing the same follow-up email from scratch over and over; just checking in, asking for availability, scheduling meetings
- dealing with repetitive admin tasks like organizing receipts, tracking expenses, sending things to your bookkeeper (Gina will talk about this one)
- experimenting with different CRM tools that never have quite the right combination of features
- manually scheduling, sorting, summarizing, repeating.
AI is not going to magically solve any of these problems, unfortunately, and it’s not going to run your business for you. It’s definitely not going to replace your judgment or the relationships you’ve built — all the things that got your business to where it is now. What it can do is take a big chunk of that repetitive work, that admin work, that “I’ve done this a hundred times before” work off your plate — marketing admin, scheduling, customer service, and so on, if you set things up right. That’s at the core of what Gina and I are talking about today: not replacing you, not replacing anyone else at your organization, but giving you back some of the hours you’re spending on stuff that doesn’t actually need to be done 100 percent by you.
So on that note, I’ll turn it over to Gina to share some of the tools we’re working with.
Meet the Tools: Claude, ChatGPT, and Read.ai
Gina Ramsey: Thanks, Katie. Thank you all who are in the room — again, I don’t want to be a talking head, I’m a super nerd about this, so please put questions in the chat. Katie will keep an eye on it, and I’ll try to as well, but we want to answer as many questions as possible.
Today we’re going to talk about the primary tools we use internally: Claude, Read.ai, and ChatGPT. We’ll cover Claude and the three ways to use it — technically there are four, but the fourth isn’t great yet. Katie’s going to cover AI for marketing, social media content creation workflows, and best practices, and then we’ll give you a simple five-step framework you can use to start this week without any technical background. Does that sound fair?
[Pause.]
G.R.: I’m going to go with yes. We typically use Claude, Read.ai, and ChatGPT — I’m sure you all know what Claude and ChatGPT are. Read.ai is the tool we use to record our meetings and get transcriptions and summaries. We never take notes in meetings anymore; there’s an app, or you can use it on a phone call as well as Zoom or whatever platform you prefer. But let’s start with Claude, and we’ll explain what each tool is for before going deeper.
I’m a heavy Claude user — or in my world, “Claudette,” because I have so many women on my team we tend to stay in the feminine. Claude is used for drafting, summarizing, and automating things like emails, follow-ups, and task creation — Katie will talk about that, since she’s brilliant with it — and even prospecting. I already mentioned Read.ai; we have workflows that use Read.ai, Claude, and Teamwork, our project manager, all in conjunction with one another.
And then there’s ChatGPT. My SEO and PPC team use ChatGPT at length — a lot of research, ideation, drafting. For a lot of users, this is their go-to. They’ve just added a Cowork-type platform, which I’ll explain shortly, to ChatGPT. Because I’m not a heavy user of it, I can’t give you many details, but explaining Claude Cowork should give you a decent sense of the ChatGPT version too. Does that sound fair?
K.M.: Sounds good.
Claude’s Four Functions: Chat, Code, Cowork, and Design
G.R.: Okay, let’s get into Claude, since this is where I spend most of my day. As I said, there are three primary functions — they’ve just added a design function; which isn’t great, I’m not going to lie, but it’s there if you’re on the paid version.
The chat function is your everyday assistant. It’s your “hey, can you… I have a meeting with Vicky, can you research I-95 and give me some talking points for our conversation.” It’s “I have a sourdough loaf in the oven that hasn’t fully proofed yet — from a business perspective – when will I know it’s ready?” It’s “help me draft an email” — using the HR plugin Claude has available, because a team member needs to be reprimanded, or whatever the case may be.
Claude Code is a more technical function, mostly for developers and power users. It creates automations, manages files, and runs multi-step tasks and workflows. We’ll talk about this a bit later. I’ve built some prospecting models, not only for the company but also for a group I’m part of called ProVisors. I’m on the executive committee as recruitment chair, and we want new people in the group, but I don’t want to step on the toes of people already in it, so we built some prospecting models within Claude Code that are pretty cool.
And then Cowork is my workhorse — I live and breathe inside it.
Claude Cowork, if you connect it to tools you work in — Google Workspace, Microsoft 365, Teamwork, Read.ai — can function as you within that ecosystem. It can draft emails, put things in folders, write things in folders. Depending on how you set it up, it can do a lot for you, and we’ll give you some specific examples of what we’re using internally.
Claude Design is supposed to compete with ChatGPT and Gemini’s Nano Banana to create images, websites, social media posts. It’s not there yet. It’s okay, but it’s not quite there.
So I’ll pause here and ask if there are any questions about the four Claude functions — I already see typing.
K.M.: I’d also add, on the Cowork front, it can sound intimidating to give it access to everything all at once. When you set it up, you can give it as much or as little permission as you want. So if you’re not comfortable with it drafting emails for you and just want something you can paste your own draft into, you can set that up. There’s a lot of freedom to customize exactly how much permission you give it.
G.R.: Correct. And I also want to point out that there are some ways to — someone’s typing, but I’m not seeing anything. Can you see two questions?
K.M.: Yep, we have two questions in the chat. The first one: do you find Claude to hallucinate less than ChatGPT?
Reducing Hallucinations and Setting Up Claude the Right Way
G.R.: In both platforms there are ways to lessen hallucinations. Within Claude, and I believe within ChatGPT too, there’s a resident memory where you can type in things like, “my name is Gina Ramsey, I own Pink Dog Digital, here’s the URL, here’s a sample of my writing, here’s what we do, here are the people on my team and the roles they fill,” and so on. You can also add: “please don’t answer unless you know the answer to my questions and can cite the resources,” and that trains the model to lessen hallucinations.
A good example: I’m on a condo association board, and unfortunately one of the residents is suing all the other residents — a whole other thing. I wanted help drafting language to respond to their demands, so in addition to turning on the legal skill, so it’s drawing legal language and citing Ohio condominium code, I also wrote, “please don’t provide any language that isn’t able to be cited through Ohio code.” What I got back was clean copy I could submit to the attorney — it saved me time and attorney fees, because I’d already done the legwork and made sure it wasn’t citing bogus case law that isn’t out there.
The other thing I’ll say about Claude is that it tends to be a lot less fluffy. ChatGPT is very fluffy — very much “oh, you’re so smart, I can’t believe you caught that,” all while it’s wrong, it’s hallucinating. And in the Claude world, you’re burning tokens with those hallucinations and with that fluffy language. So you can always turn on what’s called “caveman” mode — no fluff, no niceties, it just gives you yes, no, this is why, and it doesn’t hallucinate or burn tokens. Does that answer the question?
K.M.: I’d piggyback off that and add that I find Claude is great at asking refining questions too. I’ll tell it, “if you have any questions, ask me before responding,” and that also cuts down on hallucinations, because it walks me through a couple of questions that make the answer stronger, instead of wasting time giving me one answer, me correcting it, and going back again.
G.R.: Yep — that’s the difference between a push and a pull chat. A push chat is “give me all the information about ADRVantage — who they are, who their primary people are.” A pull strategy is “my prospect is ADRVantage. I’d like to sell them a website, social media, and SEO. Ask me the questions you need in order to prepare a good pitch and proposal, based on what you know about that company and mine.” I’d always recommend a pull strategy over a push strategy, because, as Katie mentioned, it won’t answer you until it has all the questions it needs to give you a good answer, not a hallucination, not a fluff piece.
K.M.: We have another question, which I think you already started addressing by accident: how invasive do you find Cowork to be with your drives and your information?
G.R.: It’s only as invasive as you allow it to be. In my resident instructions, it cannot write without my permission and it can never delete. On each individual chat, though, you can change those permissions — it doesn’t have to be a global setting. For example, I have an automation that drafts answers to emails in my inbox every day at 9 a.m. and 2 p.m., but it can’t automatically send them — it has to ask, it throws them in draft.
On the other hand, I have a prospecting model that doesn’t have to ask before it searches the Howard County Chamber of Commerce for new prospects — I just always allow that. So you can limit it both globally and per chat, based on what it can and can’t do without your permission.
I think it gets dangerous when you say, “here’s everything I own, you can do all the things, you have all the permissions, you don’t need to ask me anything, just go send stuff.” So I don’t find it problematic at all — and I can tell you I was very skeptical of all of this when it first started. Then I went down the “we’re driving nine hours to Lake Lure and I need something to do in the car” road, started playing with it, and thought, “oh my god, I just saved hours every single day.”
More recently, my father had some health issues, and I was working — not on vacation, but only about ninety minutes a day for about three weeks, and I was able to keep on top of things. That’s not an everyday thing, but it needed to be while I was dealing with my dad’s health issues. I was afraid of the lack of control at first, but I’ve realized how to use the control and make it benefit me. Does that answer the question?
K.M.: I think it’s a great answer. We have one more question, but I’ll save that for the end so we keep things moving. Next slide.
Three Automations in Action: CRM, Email, and Prospecting
G.R.: Here are some examples of automations we’ve created — I’ve already alluded to many of them. I was using a CRM called Streak that was costing me sixty dollars a month, and I was using about a tenth of its capabilities. So within Claude Cowork — not even in Code — I built a sidebar CRM. We’re a Google Workspace house, so it sits right alongside Workspace on the same screen. It automatically adds new prospects, new deals, anything happening within an account. There’s a note section where I track the referral source, how much money was involved, and — god forbid they went somewhere else — why. That rolls up to a spreadsheet so I can see the whole picture.
It’s got three or four other tabs: one for referral partners and revenue per referral partner, filterable by date; one for types of deals and revenue by deal type; revenue per client — a number of things we can slice and dice, and we keep iterating on it. Katie comes up with a new idea, Coral comes up with a new idea, I plug it in and say, “hey, can you do this,” and it changes — and that information is automatically shared with Katie and Coral. I’m not paying anything more than the twenty-dollar Pro plan I already pay for Claude, so I saved thirty dollars a month, which isn’t a lot on its own, but add up everything else we’re doing and we’re saving a ton of money.
K.M.: I was going to say, with the CRM, one of the biggest things is that you can make it have just what you want. I can’t tell you how many times we’ve experimented with different tools that have a bunch of stuff we don’t need and are missing one or two things we’d really like. It’s so nice to build tools that have exactly what you need, no extra stuff — it makes everything so much easier and more intuitive.
G.R.: Yep, I agree one hundred percent. And because the three of us are working in it, we collaborate to make sure what we’re building works for all three of us — it’s not just plug-and-play like most CRMs. Sorry if there are any CRM people in the room, I apologize.
The second automation is email, and I’ve alluded to this too. First thing in the morning, I have a prepared daily brief that tells me what’s in my inbox and drafts answers to those. We have our server farm, called Flywheel, that sends analytics to each individual client once a month. I don’t have the time or inclination to read those, so it automatically files each monthly report in the customer folder, clearing it out of my inbox.
It scans my newsletters and tells me what they’re about. I got one from Vicky today with a great article about David Warshawski, who I happen to know, and it flagged that I should read it. It does this again at two o’clock and again at the end of the day. It drafts the emails and throws them into my draft folder; I scan through, make sure it makes sense, and hit send.
Again, I was working ninety minutes a day for three weeks, and that’s because I live in my email — my team does the actual work, they’re truly doing the hard work; I’m doing the email triage. These email automations have saved me probably five hours a day.
And the third one is prospecting…
K.M.: I’d actually love to talk about the email automation I mentioned earlier first, if that’s okay. Like Gina was saying, we spend a lot of time in our inboxes, and I think we’ve all heard the phrase “this meeting could have been an email” — Gina and I are now at the point of saying “this email could have been an automation.” That’s one of the ways I really use Claude in my inbox.
As an operations manager, a lot of what I do is helping get things over the finish line, so one of the biggest ways I use Claude is with Gmail. I have a tagging system set up so that every time I send an email, it gets tagged “need to follow up” — that way I know I sent it and haven’t heard back yet. When the client or a co-worker replies, the tag flips to a bright red “need to respond,” telling me I need to reply because I’m no longer the most recent contact in the thread. Once I reply again, the tag flips back to “need to follow up.”
So when I start my day, I have Claude pull every “need to follow up” email that hasn’t been touched in five to seven days, depending on the deliverable, and it automatically drafts a follow-up for each one. We all kind of hate the phrases “circling back,” “following up,” “touching base” — I use them constantly, and if I’ve emailed you, you’ve probably heard them plenty — but it really is a key part of what we do. Automating that first draft saves me hours every week, which adds up over a month. I also have a skill set up for Claude to tag emails that contain work for me and build a quick bullet list at the start of the day.
As a project manager — and I’m sure Gina feels the same as a business owner — some of what we do is basically firefighting, and when you’re a firefighter, you want to go to the biggest fire first, not the closest one. This skill helps me figure out where my attention needs to be, and it automates some of the tedious work that might otherwise keep me from getting to the most urgent client need.
I think that’s a really practical, easy way to start using Claude more. Everybody here has to send follow-ups, chase an invoice, or remind a team member about a deliverable. You don’t need to be the one drafting those emails.
G.R.: I agree, those are great ideas.
One of the other email automations I use: I’ve mentioned we use a project management system called Teamwork. As emails come in with things like “I need to add this team member to my website,” those are automatically connected to Teamwork. It creates a task for the programmer, assigns it to the proper client based on the email, and pushes the task three days out, which is our turnaround time. So I’m not setting up tasks for team members myself: that’s not something a business owner should be doing. And when I have a task like “follow up with so-and-so” in Teamwork, it drafts the email, puts it in my draft folder, checks off the task, and recurs it for next week. These email automations can work in conjunction with other programs you’re already using to run your business.
K.M.: Should we move on to prospecting too?
G.R.: Yeah… and this one’s kind of a trial balloon. I don’t know if it’s going to work in actually landing clients, but I wanted to play with Claude Code and do something useful, so I built an automation that scrubs chambers of commerce, rotary clubs, and other business groups for people who fit our target client profile: the same profile stored in that resident memory I mentioned earlier. Every day it surfaces five new prospects for me to email. They’re not the generic “saw you on the internet, want to schedule an appointment” — they’re specific, like: “Hey Katie, I saw you’ve been in business for twenty years, but you’ve only got ten Google reviews — did you know a Google review is a significant marker in where you appear in search and AI results? Would you like to talk about how we could increase your Google reviews?”
I’m freestyling that example, but it gives you the idea.
The beauty of it is I’ve taught it that most prospects need seven to ten touches to become a client, or at least an interested prospect, so I have a whole ten-part sequence. I send an email, then a week later another, and another, and for ten weeks that prospect is getting these emails. All the while, on the back end, it’s logging everything to a spreadsheet, so even if we’re actively emailing five, it may have sourced ten but couldn’t find emails for some, or scored some as private equity or franchise, which aren’t good targets for us. Those go into the spreadsheet with a status note. If I get a bounce, I feed it back into the system and it takes that person out of the sequence. If we schedule an appointment, I tell it, and it takes them out too.
This runs as an independent sales tool for my business. I know a lot of people hate generic emails: I don’t know if it’ll land any clients, but it takes me about three minutes to trigger it and approve the emails.
The reason I don’t have it running fully automatically, and manually trigger it instead, is that Claude runs on tokens, and tokens renew every five hours. First thing in the morning I’m a heavy token burner because I’m doing all my getting-ready-for-the-day tasks, so I might trigger this if I have tokens left in the morning, or wait and trigger it once my tokens renew five hours later.
That was a lot of information — any questions on those items?
Live Q&A: Errors, Automation Setup, and Token Usage
K.M.: We’ll pause for questions, then loop back to the question we saved from before. Someone asked: “I’m using Cowork and it’s jammed and sent error messages twice in one week — one said ‘we’re too busy’ and the other just failed on the task. Any tips?”
G.R.: Typically I find you should reload Claude — it’s usually related to an update on their end, and sometimes it gags and you have to reload or re-log in. Sometimes I’ll just ask it, “are you stuck?” and it’ll basically kick itself in the head and say, “nope, here I go.” So try the easy route of asking if it’s stuck first, and if that doesn’t work, reload or re-log into Claude.
K.M.: Another question just came in: “How did you set up your workflows? I’m manually entering things into ChatGPT daily, multiple times a day, and I’d love to automate certain processes.”
G.R.: That’s what Claude Cowork does. Claude Code does it too, and the new ChatGPT feature, which I don’t know the name of since it literally launched this week (editor’s note: this is called Codex) does the same thing. Talk to it like you’d talk to a team member: “I’d like to set up an automation that fires every morning at 9 a.m., Monday through Friday, to do this — here are the rules — create this skill.”
A skill is essentially a saved slash-command, so it runs automatically from then on. If you want to run it on a different application one time, you just type the skill’s name with a slash and it runs — you don’t have to retype everything. Tell Claude what you want as a scheduled item, and it’ll do it. It’s the same idea in ChatGPT, just under a different name than Cowork.
K.M.: Great answer. One more question: “I use Claude Pro a lot and didn’t know about tokens — where do you see your usage against capacity throughout the day?”
G.R.: In the settings — though I’ll admit it’s a bit of a pain in the desktop app, so I check it on my phone. If you have Claude on your phone, open it, go to the bottom-left corner (your first initial), tap that, and the third item down is “usage.” It shows your usage for the day and across the week, since it resets weekly, along with any purchased credits. I built a home budget with my husband, Brian, and ran out of tokens mid-project on a nine-hour drive to Lake Lure, so I bought fifty dollars in credits back in April — I still haven’t burned through them. It’s just easier to check on the phone than in the desktop version.
K.M.: Great — we have one more question, but I’ll save that for our next break so we keep moving. Here’s where I’ll take over and talk about the social side of things.
Using AI for Social Media Ideas and Content Repurposing
Katie Meixner-Croft: AI can be a fantastic tool for coming up with ideas, especially for social media. Idea generation is where a lot of people get stuck: staring at a blank page is always harder than starting with a handful of ideas.
So one thing we do in HeyOrca, our social media planning and scheduling tool, is start with a topic, a content pillar, or an idea; feed that to AI for post ideas, select the best ones, refine the copy, schedule it, and publish it. That’s one of the ways we make sure we’re taking full advantage of AI internally.
I’ll have Gina pull up the next slide to show one of my favorite things to do with AI for social: making the most of what you already have. For a lot of our social clients, they get regular blogs and content from us, and unfortunately it often gets published once on the website, shared once on social, and that’s where it stays. That’s not necessarily bad: it’s important for SEO and GEO, and your blog lives on after it’s published. But you can do a lot more with a single post. Here’s a look at the multiple ways you can take one blog post and break it into several pieces of content quickly, using ChatGPT, Claude, or even Gemini, which I find works well for this too.
We’d feed in the blog post or idea and ask for social media captions: ways to post the blog on social and drive traffic back to the website. We can get a newsletter idea or blurb for an email draft. We can get FAQs… a big thing we’re encouraging clients to do is have FAQs on their site and update them regularly, and blog posts are a great source for the questions clients are actually asking. Future blog topics: one blog can spin off into several more once you spot the good ideas buried inside it. And a big one: video. I know not everyone’s comfortable with video, and there’s a bit of a learning curve, but you’ll get great results with Reels and TikTok. The proof is in the pudding, we see it every day. You can turn that blog post into an easy video script for a Reel or TikTok and start getting video off the ground.
I’ll have Gina pull up the next slide to show what this looks like in practice, since it’s easier for me to picture than it might be to describe.
G.R.: On the FAQs: this is one of the single best ways for your website to get found in AI. AI wants unique answers to the questions people are actually asking. If your site doesn’t have an FAQ page (which I know sounds like something out of the 1990s), it’s one of the major drivers of AI visibility. So if you’re missing that piece, it’s worth adding.
K.M.: Yes, and on our own site we have a large number of FAQs that we pull from, and they’re a key driver of traffic, we see it in the reports every month.
So here’s an example I mocked up with an HR client. The blog post was “Five Ways to Improve Workplace Culture.” You post it to the website, share it on social once… now what?
You could turn it into future social captions. You could turn it into email copy: what if you sent a newsletter titled “Five Workplace Culture Shifts Your Team Will Actually Notice”? Now you’re repurposing the content, touching base with your subscribers, and again driving traffic back to the site.
For FAQs: “How can we improve culture without a big budget?” Pull that answer straight from the blog and turn it into FAQ content.
Future blog topics could be a kind of sequel to the original: “How to Give Feedback Employees Actually Want to Hear,” “Signs Your Team Is Quietly Disengaged.”
And a video script: “Want a Better Workplace Culture? Start Here” — talk through three or all five points and link back to the blog, again driving traffic.
That’s one example to show how an abstract blog post can turn into concrete content across the rest of your business.
Why the Human Touch Still Matters
G.R.: One thing I’ll add: we’re not writing blogs with AI. We get ideas from AI, but we have human beings writing the blogs. We actually have a post on our own blog about the difference in SEO, AIO, and GEO performance between human-written and AI-written content — the human-written posts got picked up 68 percent more than the AI-written ones. So if you think you can fire your marketing team and go strictly to AI, you could, but your results won’t be where they’d otherwise be.
At no point do I think AI should be used to eliminate human beings. It should be used to give human beings more time for the important work, instead of getting stuck in the weeds on follow-ups, emails, and sending receipts to the accountant. When it comes to marketing, when it comes to SEO, AIO, GEO… you have to have a human touch. If you don’t, the AI filters will find you and you’ll get sandboxed.
K.M.: Absolutely, that’s a really important point.
We’ve got a couple of chat questions — one for you, Gina: do FAQs need to be visible, or can they be stored behind the scenes on the website?
G.R.: Yes and no. They should be visible. If they’re hidden text, Google will flag it as black-hat optimization.
Do they need to be in the top navigation? Absolutely not. Ours are buried in the footer, you have to go looking for them.
But they should absolutely be a visible page, or pages, that Google can crawl and navigate to.
K.M.: And one more: “A trend I’m seeing is pushback from consumers and partners on AI-written social content and AI-generated visuals — are you seeing that too?”
I can start on this one. I think there’s real pushback, especially around AI-generated visuals. I’ve seen a lot of it in the restaurant space, where flyers all start to look the same, or the AI imagery all looks the same. I think that’s part of why it’s so important to use AI appropriately. If you’re using it for everything and putting it out there for everything, your audience will notice, which is one of the reasons why, like Gina said, we’re not just going to AI-generate a blog and publish it, or AI-generate a graphic or caption and throw it out there. People notice, and it can reflect poorly on the business behind it, so you want to be careful, especially with graphics and imagery.
G.R.: I think you covered it well.
One example: we had a client selling a product into the vegan market, and yet their website was covered in pictures of fish and meat, clearly AI-generated. They really needed to redo their site. Everybody’s seen the 135-pound, rock-solid fitness influencer who supposedly did “tai chi walking” for seven minutes a day and now looks like she just finished military training; people recognize that as fake. It’s AI slop, and nobody wants to see it. The words should be your words; they shouldn’t read like AI slop.
We see a lot of pushback. And again, I’m not suggesting you automate everything, especially forward-facing material like marketing. But the most repetitive time sucks in your day? Automate away.
A Simple Five-Step Action Plan
K.M.: That leads well into our action plan.
G.R.: I do want to mention we put together a presentation on AI that we’ll be adding to the website — a white paper comparing ChatGPT, Gemini, Perplexity, Claude, and the other major tools. It’s not on the site yet, but if you’re interested, email me or Katie and we’ll send it over.
K.M.: We’ve got a couple more chat questions, which I’ll save for the end so we can get through the last few things. This is a quick action plan you can use right after this session, since I know we’ve thrown a lot of information at you and it can feel intimidating to just get started.
First, think about your most repetitive, time-draining task. Not time-consuming, since some time-consuming things genuinely do need you: just time-draining. What’s the one thing that, if you could take it off your plate, would free up real time in your day?
Next, think about which AI tool can help. We’ve covered Claude a lot today, ChatGPT a little, and Read.ai is another we use daily to record meetings and give us a transcript and action items so we remember what was actually discussed. I don’t know if anyone else gets into a fugue state during meetings where you walk away with no memory of the decisions made, but it’s incredibly valuable to be able to go back and check!
Next, start building. Create that prompt or workflow template. Like Gina said, you can just ask the AI, “here’s what I want to do, what do you need from me to get there,” and it’ll walk you through building the workflow. Any other workflow tips you’d add, Gina?
G.R.: You can change them even after they’ve been saved as a skill. For example, I had a workflow today for the Teamwork follow-ups. It was scheduled to run at 5 p.m. last night, but it ran at 9 this morning instead and gave me feedback on a 2 p.m. meeting today that clearly hadn’t happened yet. So I went back in, redid the skill, and had it re-memorize that if my computer isn’t on at 5, it should run the next morning but sweep for prior meetings, not future ones.
That’s a complicated example, but the short version is: iterate on these things until they’re right, and you can do that as many times as you need.
K.M.: Definitely… and that runs right into “test.” Run it, refine it as much as needed. And five is the most important step: once you find something you love, share it with your team. That’s actually how Coral and I started using Claude more heavily: Gina came to us and said, “I made this super cool thing in Claude, everybody should use it.”
You might need buy-in from other team members, but if you show up with a solution to a problem they’re also dealing with, they’ll get on board once they see how much easier it makes their life. Documenting and sharing it is really important.
G.R.: And I think the takeaway is: you don’t have to automate everything. Start with one thing, get comfortable with it, and build from there. Don’t go all in and automate everything at once, because you won’t know what’s actually working. Pick one pain point, fix it, then move forward.
We’ve touched on a lot of these already, and I want to get to the questions, but quickly:
- Don’t publish without editing, don’t publish without reviewing.
- Don’t rely on AI for strategy: nothing beats the human mind in strategic planning.
- Give it something to execute; don’t let it tell you what you should be doing, or you’ll end up relying on it too heavily.
- Don’t over-automate too fast; pick one thing, get it working the way you want, then move to the next.
- Don’t ignore your brand voice: if it sounds like AI to you, it’s going to sound like AI to the people reading it.
Sometimes that’s fine — I once responded to a legal matter without trying to make it sound like me or any board members, since we just needed to answer the questions accurately and let people take it from there. Michael Shaw probably deals with this constantly — people trying to save money on legal fees run things through AI, send him a brief, and he’s like, “this is crap.” Use your brand voice, use your education, and let AI write for you… but with proper checks and balances.
Live Q&A: Security, Accessibility, ADA Compliance, and Claude Pricing
K.M.: If anybody has more questions, drop them in the chat — I’ll circle back to a couple we didn’t get to.
One: “Do you worry about the security of the information you’re giving AI access to?”
I know this is a big one for a lot of people.
G.R.: Yep. Each platform is different, but there are ways to make sure the AI isn’t training on your data. You can turn off that setting, and I highly encourage you to, particularly around HIPAA and legal information.
The one platform I think is really rough on this front is Gemini, because if you turn off memory, you can’t build automations that persist: you have to leave it on. So make sure you have a closed system, like you can with ChatGPT and Claude, and I believe Perplexity as well.
K.M.: Definitely. “How are you using AI to assist with accessibility, like alt text and captions?”
I can start on this one. Read.ai is one way. It’s a powerful tool for giving you a transcript you can reference later, especially useful if someone needs live captions or has trouble following along in a meeting. It’s built into Canva too, so when you’re creating a video or Reel in Canva, it’ll automatically generate captions, which is a big help, and it’ll generate alt text for images going on a website or social platform.
It’s a strong tool overall, though I’ll say it’s imperfect — we have a client called Duct Doctor, and it consistently transcribes it as “Duck Doctor,” so we’ll see “you discussed Duck Doctor” in our notes instead of “Duct Doctor.”
Still, overall, I find it really valuable for accessibility.
Gina, anything to add?
G.R.: This is almost a setup question — I’m not sure who asked it, but thank you!
One of our big pushes this year is ADA compliance, since we’re seeing more firms get sued for non-compliant websites. We use a number of proprietary tools we’ve purchased: a design tool called Figma, among a couple of others, to help make sure what we’re putting out there is ADA-compliant. You can ask AI to check your site for ADA compliance, but it’ll only be about 50 percent accurate, so don’t rely on it to catch everything. It needs manual review, but it can get you halfway there.
K.M.: Absolutely.
And I think the last question I see for now is: “What’s the difference between Claude co-work and Claude Pro?”
G.R.: Claude Pro is the paid tier you use to access the platform — it’s twenty dollars a month, and it gives you access to Chat, Cowork, Code, and Design.
K.M.: Perfect.
We’ve got a couple more comments — we’ll take one more question: “How does accessibility relate to ADA compliance?” That follows right along.
G.R.: Accessibility work is never going to guarantee you won’t get a demand letter or a lawsuit. It used to feel like the magic bullet, it’s not anymore.
If you want to be ADA-compliant, you have to actually build your site that way. Doing accessibility work shows you’ve made an effort, but it won’t guarantee you’re safe from being sued.
Closing Remarks
K.M.: If anyone else has questions, feel free to email Gina, myself, or Coral, who was actually supposed to be here today. I’m sure all of us would be happy to answer them. Thank you, and lots of love in the chat for this one.
G.R.: Thank you so much. I also want to mention that if you’re looking for help with any of these automations, we’re actually building them for clients now. So, shameless plug, we can help with that.
K.M.: Absolutely. And with that, I think we’re done!
Everyone have a wonderful day, thank you for attending. We’ll be sending out the video; if you have any specific questions, shameless plug, feel free to reach out.
G.R.: Thank you so much for your time today, everybody, take care.
Thanks for jumping in, Katie, and thanks to Coral for helping get us squared away.
Take care, everyone. Thank you.
