How I Made My Own Jewelry Website AI-Ready (And What Broke Along the Way)

Andrea Li Designs: A Full AI Visibility Audit on a Live Squarespace Store

The Short Version

I ran a full AI visibility audit on my own handcrafted gemstone jewelry website, andreali.com, and implemented every fix myself. As someone who builds and operates AI audit tools professionally, I was able to move fast. For most independent jewelry designers doing this alongside their studio work, a full implementation like this would realistically take closer to a year, or longer without guidance.

Along the way, I accidentally broke my site template, discovered that my audit tool was giving me false data, found 30 URLs pointing to a staging domain I didn't know existed, rewrote every policy on my site, and asked ChatGPT to recommend someone in my category. It didn't recommend me.

Here's the full story: the wins, the failures, and what it means for your store.

What I Started With

Andrea Li Designs is a one-of-a-kind handcrafted gemstone jewelry studio in Denver, CO. I've been a bench jeweler for 18+ years. I work in 14k gold, gold filled, and sterling silver. I draw on every technique available, including fabrication, wire work, casting, and assembly, to build pieces where no single element is recognizable in the final design.

I do all of my own product photography. Every piece is genuinely one-of-a-kind: when it sells, it's gone permanently. My website runs on Squarespace 7.0.

Honestly, before the audit, my site wasn't serving anyone well. Not human visitors, and certainly not AI systems. Here's what I found:

• Zero meta descriptions on any non-blog page. Google and AI systems had nothing to work with when deciding how to describe my pages.
• Broken schema pointing to a staging domain. My structured data (the machine-readable code that tells AI what my site is about) contained 30 URLs pointing to my husband's development server instead of andreali.com. I didn't know this.
• Contradictory policies. My shipping page was 10 years old. My return policy said "30-day returns" on a made-to-order collection that no longer existed. My shipping rate was losing money on every order.
• No pillar pages, no content architecture, no internal linking strategy. My strongest content (six detailed gemstone guides) wasn't connected to anything. Dozens of pages were dead ends. Visitors could find individual products but had no way to explore by style, occasion, or gemstone type.
• My biggest revenue driver had zero content. Wedding jewelry was driving a significant portion of my sales, and I had no pillar page, no cluster pages, and no content written for that buyer. Brides, mothers of the bride, wedding guests, gift-givers: they were all finding me through word of mouth, but my site gave them nothing to land on.
• 23% of my blog content was off-topic. Pandemic-era posts about Zoom backgrounds and social distancing, a blog post about my invitation to Knit Con, a Pinterest verified merchant tutorial for Squarespace users, closet organization tips. All of it was diluting my topical authority as a gemstone jewelry expert.

In plain English: my site had been growing organically for years without a strategic plan. Individual pieces of content existed, but nothing was connected, nothing was structured for AI, and some of it was actively working against me.

The Audit That Started Everything (And Got It Wrong)

I built custom audit tools using Claude Code, which are AI-powered skills that could check my sitemap coverage, verify my schema markup, audit my meta descriptions, and score my content relevance.

The first thing I ran was a sitemap audit. The result came back: "82% of your site is invisible to Google and AI crawlers. CRITICAL."

I panicked. That's an existential finding for any e-commerce site.

This is something I see constantly with small business owners. You put up a blog post, maybe add some product photos, and move on. But without schema, without indexing, without an internal linking strategy, without thinking about how each piece fits into an overall plan... it's just a one-off page floating in space. That's exactly where I was.

So I did what the tool recommended: I tried to build a supplemental sitemap and upload it to my site. Here's where it went wrong.

To host the sitemap file on Squarespace, I attempted to link my GitHub account to Squarespace's Developer Mode. The linking process failed, and Squarespace reverted my entire site template to an old version. Navigation broke. My footer disappeared. My homepage went blank.

I filed a support ticket, found an old git backup of my template files, and restored the site. But the damage was real: hours of recovery work, triggered by a tool recommendation based on a finding that turned out to be completely false.

Here's what actually happened: Squarespace embeds enormous image data blocks inside its sitemap XML. My audit tool tried to read the full sitemap, hit a size limit, truncated the file, and concluded that most of my pages were missing. They weren't. The sitemap was fine.

The lesson: I rebuilt the tool from scratch with a "never truncate" rule, retested it, and confirmed my sitemap had zero genuine missing pages. The rebuilt tool has since been tested on three sites across two platforms (Squarespace and Shopify) with accurate results every time.

I'm telling you this because most people who sell audits don't tell you about the time their own tool broke their site. I will. The breaking part isn't optional. It's where the tools get good.

Implementation Timeline

What I Actually Fixed

Schema Markup: 30 Domain Mismatches to 0
Four collection pages had structured data pointing to a staging domain. The verification rerun after fixes confirmed: zero domain mismatches, zero placeholder URLs, zero parse errors.

Meta Descriptions: 0 to Full Coverage
Every non-blog page had no meta description at all. I wrote identity-first descriptions for every page in three batches. Each one leads with who I am and what makes the work unique.

Policy Clarity: 2.1 out of 5 to Fully Structured
I rewrote everything from scratch. Then I added MerchantReturnPolicy and OfferShippingDetails schema, and a policy summary block on every single product page.

Content Architecture: Zero to 5 Pillar Pages + 40 Supporting Pages
Every page was built with intention. The wedding cluster is a pillar page with four supporting cluster pages, ten Studio Stories, a quiz, and a tool. All validated at 0 errors, 0 warnings, 8 schema items.

Content Relevance: 23% Noindexed
96 blog posts scored. 22 flagged for noindexing. Celebrity posts about famous jewelry scored well. Without the jewelry connection, they're noise.

Google Merchant Center: 51 Disapprovals to 0
All resolved. UCP interest form submitted for AI-powered checkout readiness.

Audit Scorecard

The Before and After

Then I Asked ChatGPT to Recommend Me

After all that work, I asked ChatGPT: "Recommend independent jewelry designers who make handcrafted aquamarine pieces in gold, under $500."

My store, with its solid 14k gold, genuinely one-of-a-kind aquamarine designs, and 18 years of craft, was not in the answer. ChatGPT gave me a 6.5 out of 10 and recommended five other stores instead.

The Legibility Gap

"ChatGPT chose linguistically legible stores over genuinely better ones."

The good news: ChatGPT also called my site "cite-worthy for artisanal, one-of-a-kind gemstone jewelry." The architecture work, the schema, the pillar pages: all of it registered. I'm not invisible. I'm just early.

Here's what this means for you: Your craft might be exceptional. Your site might be beautiful. But if AI can't read the words that describe what makes you different, it will recommend someone who uses those words better, even if their work isn't as good.

That's fixable. And it starts with knowing what AI actually sees when it looks at your site.

How Does Your Store Compare?

Try this quick self-assessment. Be honest. Nobody sees this but you.

Try This Right Now (5 Minutes)

Open ChatGPT and type:

"Recommend independent jewelry designers who make [your specialty] in [your primary metal], under [your average price point]."

See if you show up.

If you don't, that's not a failure. It's information. It tells you exactly what I learned: your authenticity might be real, but it's not yet machine-readable.

That's the problem I now help jewelry designers solve.

Ready to See What AI Sees When It Looks at Your Store?