I tested my website's AI visibility. Here's what I found.

I tested my website's AI visibility. Here's what I found.

A few weeks ago in April, I started building an AI readiness audit as a new Zest service. My rule has always been the same : show before you sell. So before pitching it to any client, I needed to pressure-test the methodology on a real site. My own was the obvious candidate.

Quick context : at the time of the test, Zest was running on Lovable. That was a deliberate choice. I wanted to understand the platform firsthand, see what the fuss was about, and have a clean, simple storefront up quickly. A single-page site made sense for that. If you're reading this, the migration to Framer is already done. But the test happened on Lovable, and that context matters for everything that follows.

I opened ChatGPT, Perplexity, and Google, typed variations of "digital marketing consultant Dubai" and waited.

The results weren't wrong exactly. But they weren't right either. Perplexity read my site and returned "social media management" as a core service. ChatGPT didn't cite me at all. Google AIO only picked me up when my name was explicitly mentioned.

The AI wasn't lying. It was reading exactly what I had published, or failing to find enough signal to include me. That's a different problem entirely.

What did AI actually say about Zest?

Here's what came back across the three platforms, before any changes.

ChatGPT didn't cite Zest at all on a straight "digital marketing consultant Dubai" query. When I pushed and asked directly why, it analysed the site and gave me a verdict I wasn't expecting : positioning too vague, not enough proof, too generalist to recommend for a serious SEO or UX audit. Its exact words : "The site doesn't provide enough signals to appear in a spontaneous recommendation." Coming from an AI evaluating my own consultancy, that stings a little. But it's useful.

Perplexity fetched the site. That's the good news. The bad news : it read my old meta description and reproduced it as fact. "Social media management" appeared as a core Zest service. That hasn't been my focus for a while. The AI wasn't making things up, it was reading exactly what I had published and treating it as current truth.

Google AIO was the most generous of the three. It cited Zest, named me by name, got the positioning roughly right. But only when my name was explicitly mentioned in the query. On a cold category search, I was invisible.

Three platforms, three different failure modes. None of them were lying. That's the uncomfortable part.

What can you actually fix, and what has to wait?

Before listing what I changed, it's worth being honest about the constraints. Zest runs on Lovable, built as a single-page landing. One URL, one page, everything stacked vertically. That architecture has a hard ceiling when it comes to AI readiness : no dedicated service pages means no structured signal per service, no blog means no fresh content for crawlers to index, and no server-side rendering means some AI crawlers may not execute the JavaScript at all before moving on.

You can patch a single-page site. You can't fully optimise it. That's a redesign conversation, not a quick fix. More on that in the next article.

Within those constraints, here's what I changed.

The meta description. It still said "social media management" as a lead service. Perplexity was reading it and reproducing it as fact. I rewrote it to reflect what Zest actually does today : SEO, AI search visibility, growth systems. One change, immediate impact on how the site gets described.

The JSON-LD schema. The existing structured data was incomplete and partially incorrect. I replaced it with a full ProfessionalService schema including contact details and LinkedIn, and added a second FAQPage schema block with seven questions and answers. This is what search engines and AI crawlers read when they want structured facts about a business.

The llms.txt file. A relatively new standard that gives AI systems a clean, readable summary of what a site is and does. I created one and deployed it at /llms.txt, explicitly listing the crawlers allowed access : ClaudeBot, GPTBot, PerplexityBot, GoogleOther, Bingbot. Only 10% of sites have this file. On a no-code platform like Lovable, it takes about fifteen minutes to implement.

The canonical URL. A mismatch between www and non-www versions was creating inconsistency across schema, sitemap and meta tags. Standardised to non-www across all files.

Robots.txt. Verified that no AI crawlers were accidentally blocked. A surprisingly common issue, easy to miss, easy to fix.

Five changes. Two to three hours of work. Validated via Google Rich Results Test, which confirmed three valid structured data elements.

These fixes cover what's possible within a single-page architecture. The rest goes on the redesign list.

Moving away from the landing page format entirely. Building a proper site structure with a general services page and individual dedicated pages per service, each with its own FAQ section. Rewriting the copy to be not just SEO-optimised but structured for GEO : declarative openings, direct answers, named entities. Adding a blog for fresh indexed content. And a project portfolio, because AI systems cite third-party proof and case studies far more readily than self-promotional copy.

That's the Framer migration. But that's the next article.

Early results : what changed, and what hasn't yet.

It's been a few weeks since the changes went live. Too early for a full verdict across all three platforms. AI systems don't update in real time, and some of these signals take weeks to propagate. But Perplexity, which fetches sites directly rather than relying on a cached index, already tells a different story.

The same query that returned "social media management" in April now reads : "Digital marketing strategy for SMBs ready to grow — SEO, AI search visibility & growth systems. Dubai-based, international scope." That's the updated meta description, read and reproduced correctly. One change, visible result.

ChatGPT and Google AIO are a different conversation. I haven't retested them yet, and honestly I'm not going to until the Framer migration is live. Testing now would be like measuring the impact of repainting a house that still needs a new foundation. The structural work comes first, then the measurement.

Part 2 of this series will cover the full before/after across all three platforms, post-redesign. That's where the real data will be.

The ugly truth you want to know before losing business.

Most SMB websites have the same issues Zest had. A meta description written years ago that no longer reflects what the business actually does. Structured data that's missing, incomplete, or outdated. No llms.txt. AI crawlers either blocked by accident or simply not given enough signal to work with.

AI systems don't know your business. They know what your site tells them. And if your site is telling them the wrong story, or no story at all, that's what gets reproduced every time someone asks ChatGPT or Perplexity for a recommendation in your category.

The good news is that some of this is fixable quickly. The meta description, the schema, the robots.txt, the llms.txt: these are hours of work, not months. The harder part is the structural work: dedicated service pages, fresh content, case studies, a blog. That's where the real AI visibility gets built.

But you can't fix what you haven't measured.

Want to know what AI says about your business?

Send me your URL. I'll take a look and give you an honest read of where you stand across ChatGPT, Perplexity, and Google AIO. No pitch, no automated report. Just a real assessment from someone who's done it on her own site first.

You can reach me at hello@zest-your-business.com .

And if you want to follow the rest of this experiment, part 2 is coming once the Framer migration is live. That's where the real before/after data lands.