You can’t get away with AI copy if you want your SEO performance to be reliable. My conclusion: AI should be used to enhance human productivity – it can’t replace human creativity.
But, I do think we can use AI to build interesting content, with some thought and the right tools. My thoughts lead me to what AI tools are meant for: being able to make things yourself that you might otherwise need help to build.
When I ran Builtvisible (now sold to Brave Bison), we were very much on the cutting edge of what we called “interactive infographics”. Usually, very simple browser based apps with jQuery, HTML and CSS. It was great for our sales (cool things tend to catch the eye in presentations) and many of these assets were amazing link magnets. But that was a decade ago now!
At the time, a big problem with these builds was the overhead. You might charge a client a mid 4 figure sum for one of these simple “interactives” before it was promoted. Mostly, because of the overhead and reasonable need for operating profit. We would have at least 1 developer, working with a designer, working with the SEO team… You get my point.
But it’s 2026 and we have a lot of new toys to play with! By we I mean, smaller teams, individuals and in-house marketers who, with AI, can do the work of a team of 5 in a few days rather than weeks. It is tempting to say “a day” but your first rodeo never goes to plan.
And by “toys” I mean Claude, setup in just the right way to be able to develop and deploy and even test.
Get practical AI tools and workflows delivered weekly.
No spam. Unsubscribe anytime.
First: The Technical Bit
I’m not going to go too hard on the technical bit in this article, mostly because I don’t want to scare anyone off! However, being able to build with AI is a learning challenge and does require you to understand the rough “how does it work?” part.
The stack is React 19 with Vite 7 for the frontend, all client-side processing except the AI recommendations. When you fill in your budget, the planner calculates locally in your browser using standard JavaScript to produce income totals, expense breakdowns, savings rate, the lot. The PDF generation uses jsPDF 4 with jspdf-autotable, and Excel exports use ExcelJS. Both exports happen entirely in your browser, no server uploads. Pretty good for a free tool, I think!
The AI recommendation feature is the only bit that hits a server. When you click “Get Personalised Recommendations”, your budget numbers (just the numbers, no personal details) get sent to a Cloudflare Worker running Llama 3.3 70B via Workers AI.
The prompt compares your spending to regional benchmarks, housing costs in the UK vs US, that sort of thing, and returns a JSON object with observations and priorities from the LLM. The data is processed in real-time and immediately discarded, nothing is stored. It’s about as private as you can make it, save for Plausible Analytics, which is literally a privacy-first analytics app.
Deployment runs through GitHub Actions. Push to master, GitHub builds the Vite bundle, deploys to Cloudflare Pages, and the whole thing goes live in about 90 seconds.
The gamification (XP, achievements, progress tracking) is pure client-side state management with React hooks. Framer Motion 12 handles the animations, those transitions when you unlock achievements or watch your savings rate climb. The whole thing weighs in at about 2.6MB after minification (mostly Recharts and ExcelJS), which is hefty but acceptable for a single-page app that works offline once loaded.
Maintenance and Security
Security-wise, you need to run npm audit to check all is well. I use Snyk for security tracking which creates a pull request should a vulnerability on one of the libraries be found. It started with 4 vulnerabilities (1 critical, 2 high, 1 moderate) and spent a day in early January upgrading everything.
Claude with Desktop Commander handled most of the heavy lifting during development. I’d describe what I wanted (“make the AI treat custom income as income, not expenses”), Claude would modify the React components and Worker functions, run the builds, test locally, commit to Git, and deploy to Cloudflare. The whole workflow, from bug report to production fix is automatic. Just like those vibe coding saas apps do.
Running a Few SEO Tests
The noscript section (view with Javascript disabled) is deliberately verbose because nearly all traffic comes from ChatGPT users (I think: asking about budget planning – TBC!). I thought I’d do this (yes it feels very much like SEO strategy circa 2004 but it’s interesting). The HTML includes 23KB of schema structured markup explaining what the app does, what questions it answers, and how it compares to competitors like MoneyHelper or PocketGuard.
Features
The AI recommendations feature compares your spending to regional benchmarks and tells you where you’re overspending. Housing too high for your income, that sort of thing – then suggests specific cuts or congratulates you on a job well done. As I mentioned in the technical bit you can export to PDF for a nice-looking report or Excel if you want to mess with a few formulas or add some privacy specific data yourself.
The point of the project was to learn Cloudflare AI workers for a larger jobs/content feed-based project I’ve been building (hence the AI jobs section on this website, which I’ll write about soon).
The analysis output is of course, just a prompt running on the AI worker. Llamma was a bit reluctnant to do the job, and frankly, given the brief, it’s all a bit overpowered for the task. I think o4-mini could have done a good job. Something to test.
Content Marketing with AI
So, this is where I think we’re headed – not working with AI to copy/paste plagiarised content from a web browser, but to enable basically anyone to learn how to do bigger things for themselves.
Everyone has ideas – but execution has always been expensive. This is no longer the case, and it really depends on your appetite to learn an entirely new, but simple to pick up skillset. It takes time, patience, perseverance, and a few late nights! Or, help from someone who has been there (get in touch).
For clients, in-house teams, this means something particularly radical. It doesn’t really matter how small or large the company is, each of your departments should have AI research and development happening to suit the individual needs of a particular team. Imagine if your content team of 10 could build and deploy stuff like this? Costs down, productivity up (with the right help) and customer needs met.
The traffic was a surprise, because this was meant to be a learning project. But here we are with no promotion and the traffic has started to trickle in!
I’m a Marketing Technologist working with clients on automation, API design, content marketing workflow and data augmentation / retrieval. Here for people with interesting problems to solve.