Hey Boring Marketer,
Everyone says the hard part of a local AI marketing service is landing the client. Pick the niche, nail the offer, get the yes.
I've owned marketing agencies for over a decade, and that's not where it gets hard. The hard part is delivery: doing the work for every client, every month, without it eating your week.
You close a client once. You deliver for them every single month. The site audit, the fixes nobody gets around to, the posts that have to go out on Tuesday, the "can you just check this before it goes live." All of it comes back next month, and if every piece eats an afternoon, you top out at three clients and a calendar with no gaps in it.
So this week I'm featuring members who built free tools for the repeat work, and most of them check their own work before anything reaches a client:
Harish built a site crawler your AI can run. It finds what's broken, then re-checks the page later to prove the fix actually landed.
Alex built a skill that draws and animates whatever you describe, all in code. We made our Oct 22 workshop video with it, and you'll see a few scenes from it below.
Adam rebuilt the AI content team he's run for clients for two years. It drafts, checks its own facts, and waits for your OK before anything goes out.
Tony had Claude audit his app right before launch, got humbled, and turned the process into a prompt that tests every screen size and fixes what it finds.
All four are free to start.
And on Oct 22 I'm building an AI marketing service for a real local business from scratch, live and free. Register here.
#1. A site crawler your AI runs, that proves the fix actually landed (by Harish Ganapathi)
Every local business site has the same pile of problems. Broken links, missing page descriptions, links pointing to pages that don't exist anymore. Finding them is easy. The work starts after that: who's fixing what, did they actually fix it, and how do you show the client it's done.
Harish Ganapathi built VisibilityKit Crawler for that whole loop. It's a desktop app that crawls a site the way Google's bot would, and your AI can drive it directly. Ask Claude (or whichever AI you use) to audit a client's site, and the crawl starts and fills in on your screen while it runs.
Then it handles the part most audit tools leave to you:
Every problem becomes a task. Ranked by how bad it is and how many pages it hits, with an owner, a note and a status.
Fixed means re-checked. After the change goes live, it crawls those exact pages again and records what the site actually returned. His line for it: "Marked fixed is not the same as verified."
It won't fake a pass. If the re-check times out or the site throws an error, it tells you, instead of quietly counting it as done.
Built for client work. A workspace per client, presets for Shopify sites and site moves, and a branded action plan you can hand over with the evidence attached.
Client data stays on your computer. There's no crawl server on his end, so nothing leaves your machine.
The free tier covers the full audit, fix and re-check loop, and exports a plain action plan. Client workspaces and your own branding come with the $149 plan, paid once, not every year. He's also building a tracker that shows how many visitors ChatGPT and Perplexity send a site, with a waitlist open now.
The idea to steal: don't send a client a list of problems. Send them proof the problems are gone. That's the report that renews a retainer.
Download it free (Mac, Windows, Linux): visibilitykit.app/crawler
#2. Hand-drawn art and animation, written entirely in code (by Alex Greenshpun)
Alex Greenshpun got hooked on making illustrations and animations with Claude Code, using nothing but code, and turned it into a skill called anidoodle. You describe what you want and it draws it. In Alex's words, it's for anything from "slide decks, websites, GIFs for social" and beyond.
It launched in our community about three weeks ago with 9 styles. It's now at 32 (pencil and watercolor, risograph, woodcut, pixel art, chalkboard, toy bricks, embroidery and more), and the repo has passed 900 stars. Every style can also film itself being drawn, mark by mark, in the order a real artist would work, and it writes its own music to go with it, also in code.
Because the whole picture is code and not a generated image, two things work differently:
Edits don't wreck the rest. As Alex puts it, "say you want to change a detail in one of the frames? That change is just a few lines of code, while the rest stays the same."
You pay for it once. An image model spends money on every frame, every time. Alex's 47-second demo film is 1,410 frames, and once the code is written it re-renders for free, at any size, forever.
The idea to steal: when AI makes a visual for you, ask for the source, not just the file. The source is what lets you change one thing next month without starting over.
Get anidoodle (free, open source, and drop Alex a star if you like it): github.com/alexgreensh/anidoodle
We tried it ourselves and made the video for our Oct 22 workshop with it. Here are three scenes from it:
#3. An AI marketing team that won't post anything you haven't approved (by Adam Sandler)
The scary part of letting AI run a client's content isn't weak copy. It's a post going out that nobody signed off on, or a claim on the client's page that isn't true anymore.
Adam Sandler has run the same AI setup for client work for almost two years, built around exactly that risk. When Opus 5.5 came out, he rebuilt it inside Claude Cowork as his first test drive and packaged it as a free plugin. Here's how it runs:
You only talk to one AI. It acts as the manager and hands each job to a specialist (strategy, content, social, email, research, and one that keeps the brand info current). Each specialist only gets the files and tools its job needs.
Drafts only, you approve. Nothing gets published without a human saying yes, which matters once a client's AI is connected to their real accounts.
A second AI checks every draft before it reaches you, and sends it back if something's off.
Every draft comes with a paper trail. Which brand files it used, which sources it cited, which AI wrote it and what the reviewer flagged.
One dashboard for all of it. What's waiting on you, what's blocked, and where the brand info is thin or going stale. Approve, ask for changes or reject right there.
It gets better as you use it. What you approve and reject feeds back into the brand file, so the next draft starts closer.
In the video he tests it on a real brand (Owala, the water bottle) using nothing but the name. It found a "Best Inventions 2023" award badge on their site and, instead of dropping it into a caption, asked him what to do with it. He told it to block the claim until someone verifies it, because it's from 2023 and nobody knows if the brand still uses it. Then it wrote three Instagram captions, ran them past the reviewer, sent one back for a fix, and queued all three for his approval. It even flagged a question about em-dashes, because Owala's voice file didn't say whether they were allowed.
The idea to steal: the more access your AI has to a client's accounts, the more the approval step matters. Build it into the system before you need it.
Watch Adam build it (9 min): youtube.com/watch?v=3LUFkZkJMX0
Get the free plugin (email sign-up): the-viable-edge.kit.com
#4. The audit that humbled a "finished" app, turned into a prompt anyone can run (by Tony Morinello)
Tony Morinello, one of our community managers, was rebuilding an old project of his from scratch. He liked how it looked and figured he was close to launch. Then he had Claude audit the design and usability, and in his words, "it humbled me quickly." He ran the same audit on the mobile version and the same thing happened.
So he took what worked from both rounds and turned it into one reusable prompt. You paste it into Claude Code (or Codex) inside the project, and it does the QA pass nobody has the patience for:
Every screen, every size. Five phone sizes, five tablets and six desktop widths, plus dark mode and large text turned on.
Worst-case data on purpose. Fake 50-character names, long addresses, empty screens and error screens, so it finds what breaks before a real customer does.
The stuff you'd miss by eye. Button text that's cut off, tap targets too small for a thumb, text under 10px, colors too faint to read, colors that aren't on brand.
It doesn't touch what the site does. No fields added or removed, and forms submit exactly the same. It only fixes how things are sized, placed and grouped.
You stay in charge. It stops three times to show you before and after screenshots side by side and asks before anything it can't decide alone.
It also keeps notes from its own past mistakes. One of them is about a button that passed every check while its label ran 21 pixels past its edge, with the fix written right there: "measure the text too."
A heads-up: a full run takes hours and needs a bigger Claude plan than Pro. On a client site, it's worth it.
The idea to steal: before you call a client's site done, check it on the screens their customers actually use, not the laptop you built it on.
Grab Tony's prompt (free, open source): github.com/MorinelloA/ui-ux-audit-and-fix-prompt
Tony's "it looked great" moment is one I know well, from the selling side. My pages used to look great and not convert. So I started saving funnels that do, real ones with ads still running, and taking them apart from the ad to the last email. That turned into Boring Funnels. A free account gets you this week's teardown: boringfunnels.com
These four tools cover delivery. On Oct 22, I'll show you the whole thing from scratch.
How I'd Build an AI Marketing Service for Local Businesses From Scratch
Thursday, Oct 22 · 11:00 AM ET · free
For 90 minutes I build an AI marketing service for a real local business, live. How I pick the niche, what the offer looks like, and what I'd charge for it every month. I'll build the funnel live too, with Boring Funnels. One person who registers gets the hot seat, and I'll build the session around their business.
The replay only goes to people who register, so sign up even if you can't make it live.
Wrapping Up
Four builds, one service:
Harish made the audit prove itself. Nothing counts as fixed until the page is re-checked.
Alex made visuals you can change next month without starting over.
Adam made sure nothing reaches a client's audience without a yes.
Tony retired "it looks fine on my laptop" as the test.
The job | The tool | What it costs to start |
|---|---|---|
Find what's broken on the site, prove it got fixed | Harish's crawler | Free (client workspaces and branding: $149 once) |
Anything you can describe, drawn in code | Alex's anidoodle | Free, open source |
Run the content, nothing out without approval | Adam's AI marketing team | Free plugin |
Make the site work on every screen | Tony's prompt | Free (a full run needs a Claude plan above Pro) |
Three of the four do the work and then check it before it ever reaches you. That check is what keeps you from topping out at three clients, and it's the part clients notice most, because the mistakes never get to them.
Pick the one job you still redo by hand every month and hand it to one of these this week.
Miscellaneous vibes
The best business you can start with AI agents right now isn't an app. It's building funnels for the roofer down the street, my full case in a 15-minute read
How I'd build an AI services business for Main Street, in 39 minutes, including why I wouldn't start with HVAC or plumbing
Stop pitching "we'll build you a chatbot." Pitch "every missed call gets a reply in under a minute", plus two more swaps that sell the result instead of the tool
There's no better time to own or help a boring business in the skilled trades. Generator installs are a good one for an electrician
My neighbor runs a pool business and started installing saunas: about $3,000 wholesale, around $10,000 installed. Every agency pitches HVAC and plumbing, so go one door down
Before you pitch a local business, check if they're already paying for ads. Then pitch more booked jobs from what they already spend
Greg Isenberg on the $5 trillion AI roll-up opportunity: most of the businesses changing hands over the next decade are small, and most still run their back office the way they did twenty years ago (plus a shoutout to us for being early)
Idea Browser asks can you name your customer and call them today? Most people building a business can't
Google rewrote the Business Profile rules: pricing, time-bound offers and direct calls-to-action are now banned from listings, and Posts can't carry a phone number or email anymore. Worth a pass on every client profile
Google Local Services Ads can now charge for an unanswered call if the caller stays on the line more than 20 seconds
The New York Fed surveyed 5,000+ small businesses: 46% now use AI, and 77% of those saw no change in labor costs. It's adding capacity, not cutting staff
Keep it boring
Delivery is where most AI services quietly fall apart, and it's also where the good ones pull ahead. Every tool in this issue came out of our community, built by people doing the work for real.
Wanna watch me build an AI marketing service from scratch, live? Register free for Oct 22.
Wanna see a funnel that's selling right now, ad to last email? Get this week's free funnel.
Wanna build alongside 2,500+ marketers doing exactly this? Join The Boring Marketing Community.
Got feedback or a question? Just hit reply. I read every message.
Know someone who'd love this newsletter? Share it with a friend who's building a service for local businesses.
Keep it boring,
James
P.S. The workshop replay only goes to people who register. If you can't make Oct 22 live, sign up anyway and it'll land in your inbox.