Blog · By Rea Hailley, CEO and Co-Founder
How We Rebuilt Our Own Website for AI Search
I run New Idea Machine, a Calgary custom software and AI agency serving non-technical business owners across Canada and the US. We build custom software, automations, and AI agents for business owners every week. This month I finally pointed the same AI tools at our own website. In three days I rebuilt it with Claude and Gemini and optimized it for GEO, which means showing up as the answer when customers ask AI tools like ChatGPT who to hire. Here is exactly what I did, what the AI caught that a professional marketer missed, and the playbook you can copy.
Here's what makes this worth your time: I'm not the technical one. Al Del Degan, my co-founder, is the CTO. I run the business side, I don't write code, and until this month I knew very little about SEO and even less about GEO. If I can do this, so can you.
I expected a faster typist. What I got was closer to a junior partner who never sleeps, argues with me when I'm wrong, and occasionally needs to be told no.
What I did in three days
- Rebuilt the homepage around the three things competitors can't copy: clients own every line of code, we help Canadian companies get government funding for the build, and we'll talk you out of hiring us if a $100 a month tool solves your problem.
- Built two new service pages around search terms with real volume, wrote seven blog posts grounded in real client stories, and filled 22 testimonial pull quotes by pulling the transcripts from 19 of our own YouTube videos. That whole job took one afternoon.
- Fixed bugs and design problems along the way: videos that stretched across the screen when you hit play, walls of text nobody would read, a pricing claim that contradicted itself across three pages.
The GEO playbook
GEO is where buying decisions are moving: when someone asks ChatGPT or Gemini "who builds custom AI agents for small businesses," you want to be the answer it gives. Most websites are invisible to it. Claude taught me the playbook by applying it to my own site:
- Write declarative sentences AIs can lift word for word. "We deliver digital transformation solutions" gets ignored. "We are a Calgary custom software and AI agency serving non-technical business owners across Canada and the US" gets cited. Every page now opens with a direct answer paragraph a machine can quote.
- Add an llms.txt file. A plain text file at yourdomain.com/llms.txt that tells AI crawlers exactly what you do, who you serve, and where your key pages are. Takes an hour. Almost nobody has one.
- Structured data everywhere. Schema markup that spells out your services, your city, your service area, and your FAQs in a format machines read natively. We validated every block.
- One page per search intent. Nobody searches "operational efficiencies." They search "AI agents for business," a term that grew triple digits this year, and "AI automation agency Calgary." So those became pages. The boardroom phrase became nothing.
- Your claims get cross-checked. This is the part most people miss. AI engines verify your site against LinkedIn, directories, and reviews before citing you. Which leads to the biggest lesson.
The AI pushed back on me. Here's what it caught.
- An unprovable claim. We do serve Europe, but we had no case studies or directory listings anywhere on the web to back the claim. Claude argued that an unverifiable claim doesn't just fail to help you, it lowers the AI's trust in everything else you say. I cut it until we can corroborate it. Claim what you can prove. Expand when the proof exists.
- The temptation to invent stories. I gave Claude seven blog titles. For two of them it had no underlying facts, and instead of inventing a client story it wrote pattern pieces and asked me for the real details. A made-up story that collapses under one question costs you more than ten true posts earn. AI engines and buyers both check.
- Inconsistent claims. Our cost calculator was "free" on one page, "no charge" on another, and "$197, credited back" on a third. Our structured data pointed at a LinkedIn URL that didn't match our real profile. Small stuff. Except an AI reading three different prices doesn't know which is true, so it trusts none of them. Inconsistency is invisibility. One truth, enforced everywhere.
- Funding claims that were too broad. Our site implied every client could get government funding. US clients don't qualify for the Canadian programs. We scoped every claim to Canadian companies, including a plain disclaimer in our llms.txt so no AI ever tells an American prospect otherwise. Clarity beats reach.
If your AI never pushes back, be worried.
Here's the uncomfortable part: I had hired a marketing person before this. The contradicting prices, the mismatched LinkedIn URL, the unscoped claims, the keyword nobody searches. All of it shipped past a professional. Not because they were bad at their job, but because no human proofreads an entire website against itself. An AI does it in minutes and doesn't get bored.
Second opinions now cost a prompt
When Claude drafted the new homepage headline, it wrote three options and recommended one. I ran all three past Gemini. Gemini picked a different one, and its reasoning fit my audience better, so that's what shipped. Claude implemented the option it had argued against without complaint. Second opinions used to cost a consulting fee. Now they cost a prompt.
I also fired WordPress
Our old site ran on WordPress: plugin updates, page builders, and fees for all of it. The new site is plain, fast HTML. No plugins to update, no page builder to fight, and a fraction of the cost.
The workflow now: I tell Claude what to change, it makes the change, and the production site updates automatically. No admin panel, no waiting on anyone. A copy tweak goes from "I want this" to live in minutes.
And before you ask about SEO: every original URL was kept exactly as it was, and the two pages we did rename got proper 301 redirects. Years of Google equity, zero broken links, zero 404s.
The division of labor that made it work
Every word on the site was approved by me, word by word. Claude drafted, hunted, fixed, and pushed back. Gemini arbitrated. I decided. The tools don't replace your judgment. They sharpen it.
Total cost: three days of my attention and a subscription.
Want to do this for your own business? We turned this exact playbook into a live four-hour bootcamp: AI for Business: Build Your Website in 1 Day & Get Found By AI (No-Code). You build your site yourself, with our CTO guiding every step.
Our students' words, not mine: "Everyone who has ever thought 'there has to be a better way to do this' should attend the Bootcamp," says Heather Friesen. Crystal Chen: "The value of this workshop lies not only in how to start building, but also in when not to start building to avoid costly mistakes." And David Hunt, one of our students, is 87 years old. "I'm much better prepared now than I ever was yesterday." If he can do it, anyone can.