We sell schema fixes. Our own site scored zero.

We sell schema fixes. Every audit we run checks whether a site emits structured data for AI search, and our reports tell owners what to add first. So there is no gentle way to say this: the first time we ran our own audit on whocanfindme.com, the structured data category came back 0 out of 100. Not a low score. Zero. The site emitted no JSON-LD at all, and the tool told its own makers what to fix first: FAQ schema, the single highest-weighted structured data signal we check. This is the honest write-up, including the caveat that stops our before-and-after comparison from lying.

The zero, in full

We run every new check against our own site before we trust it on anyone else's. When the structured data checks went live, we pointed the audit at whocanfindme.com and got the one result you never want in your own field: 0 out of 100.

The reason was simple. The site emitted no structured data anywhere. The pages read fine to a human. They rendered fine in a browser. But at the layer where answer engines read labels rather than prose, the site said nothing at all.

We could have fixed it quietly and never mentioned it. Honesty is supposed to be the product, though, so the failure became a test case, and the test case became this post.

Structured data for AI search, in plain English

Structured data is a label maker for your pages. Your website is full of facts a machine has to guess at: what your business is, what it costs, which questions it answers. Structured data attaches an explicit label to each fact so nothing is left to guesswork. JSON-LD is the standard format for those labels: a small block of machine-readable text in the page's code that says, in effect, "this page belongs to this business, here is what it offers, here are the questions it answers".

Engines read labels more reliably than prose. A human can infer that "Monday to Friday, nine until five" means your opening hours. A machine parsing millions of pages is far more dependable when the page carries a label saying these are the opening hours, followed by the value. The four engines we measure (ChatGPT, Perplexity, Gemini and Claude) all depend on what their crawlers and indexes can extract from your pages, and labelled facts survive extraction better than well-written paragraphs. We went through what those systems actually retrieve in an earlier post on what AI search engines read.

One thing structured data is not: a ranking trick. It does not persuade an engine to like you. It removes ambiguity about what your pages already say. That is the whole job.

Why FAQ schema is the highest-weighted signal we check

FAQ schema is a specific set of labels that marks a question on your page as a question and the text beneath it as the answer. In our structured data category it is worth +30, more than any other single schema type we score.

The reasoning is not mysterious. Answer engines are looking for answers. When a buyer asks ChatGPT "does this tool need a signup", the engine's job is to find a source that answers that exact question cleanly. A page carrying FAQ schema hands over question-and-answer pairs already cut to shape. No paraphrasing needed, no guessing where the answer starts and ends. Of everything a small site can label, question-and-answer pairs map most directly onto what an answer engine is trying to produce.

Two honest qualifications. First, the +30 reflects readiness, not a promise. Our recurring shorthand: readiness asks whether the door is unlocked, visibility asks whether anyone walks through it. FAQ schema unlocks a door. It does not march buyers through it, and nobody can guarantee a citation. Second, the schema must label answers that genuinely exist on the page. Marking up questions your page never answers is lying to a machine that reads carefully, and engines are getting better at noticing.

What we changed and what moved

The fix was not clever. We wrote FAQ schema for the pages that already answer real buyer questions: what the scan checks, what it costs, whether it needs a signup. Every marked-up answer was text already visible on the page. We added the JSON-LD, deployed it to the live site, and re-ran the audit.

What moved: the FAQ schema check flipped from missing to found, which is +30 on its own, and the structured data category finally came off zero. The tool that had embarrassed us confirmed the fix the same way it caught the gap.

What did not move: our AI answers, at least not that week. This is the part most vendors skip. Engines re-crawl and re-index on their own schedule, so a readiness fix today shows up in answers weeks later, if it shows up at all. Anyone who tells you schema changed their AI visibility overnight is describing a coincidence.

The caveat that keeps the comparison honest

Before and after must run on identical origins, or the comparison lies. An origin is just the exact address being tested: your live domain, not a staging copy, not a local build. Our own methodology is blunt about why: running the audit against a local test environment costs around 35 points of technical health on its own, because a test setup fails checks (security headers, crawlability, real certificates) that the live site passes.

So the rule we hold ourselves to, and the one to hold any tool to: compare live against live, same domain, same pages. If a vendor shows you a dramatic before-and-after where the "before" was a staging site, the improvement is partly theatre. Our zero was measured on the live site, and so was the re-run. That is the only reason the two numbers are comparable.

Where to start on your own site

You can check for structured data in about a minute. Open any key page, view the page source, and search for "ld+json". If nothing comes up, you are where we were: a site that reads well to humans and says nothing at the label layer. Start with FAQ schema on the one page that answers your buyers' most common question, mark up only answers that are already there, and put the fix on your live site before you measure anything.

Or let the audit do the looking. Run a free scan at whocanfindme.com. No signup. Takes about ten seconds. It will score your structured data alongside everything else the engines check, and it will not spare your feelings. It did not spare ours.