The Stars Were Fake:
A Structured Data Audit of This Site

The pitch for structured data is seductive: paste a <script type="application/ld+json"> block of schema.org JSON into your page, and Google rewards you with stars, sitelinks, accordions, even a search box under your result. Some of that menu no longer exists. Some of it was never controlled by markup at all. And the most clickable item on it — review stars — is the one people fake.

So I did the boring thing: ran this site against the current menu, type by type, keeping only what each page can honestly show. This is the audit, the fixes that shipped, and one confession from the git history.

What the menu actually offers in 2026

Three years of quiet deprecations cut the menu down. In August 2023, FAQ accordions were restricted to well-known, authoritative government and health sites — everyone else’s FAQPage markup still validates in every testing tool and renders nothing. HowTo went desktop-only that same month and disappeared entirely the next. In November 2024 the sitelinks search box — that little input under big-brand results — was removed from results altogether; Google kept the WebSite schema it rode in on, but only for reading your site’s name.

What’s left for a normal site is a shorter, steadier list.

What the audit found

The site was in better shape than I expected. Both long posts already carried full TechArticle schema — headline, dates, author, publisher, and the entities each piece is about. FAQ sections render from one array that feeds both the visible markup and the FAQPage JSON, so the two cannot drift apart. The tool index emits a CollectionPage. The sitemap lists only canonical URLs, with lastModified matching the date each page displays and the date its JSON-LD carries.

Then the gaps. Breadcrumb markup existed on exactly one route type. The notes section — summarized talks and interviews, every entry in the sitemap — carried no structured data at all. And nothing anywhere declared the site’s own identity: no Organization, no WebSite.

The stars this site once had

Here is the confession. This site used to show, at the top of every tool page, a “Community Rating” panel: a star average, a per-dimension breakdown, and a confidence line reading Wilson confidence ±0.15 (95% CI). Those numbers came from a mock fixture — the fixture’s own comment said Mock confidence interval. Below it sat a “Rate this tool” form that validated your input, thought for a second, told you your rating was submitted, and dropped it on the floor. There was never an endpoint behind it.

It’s gone now. And reading Google’s current review-snippet rules, it could never have been honest markup either: ratings must be visible on the page, sourced from actual users, and not aggregated from other sites. Since 2019 a review is “self-serving” — displayed as nothing — when the entity being reviewed controls the page the review appears on. And in July 2026 Google made the obvious rule explicit: fake or undisclosed incentivized reviews are banned from markup, full stop.

The advice still circulates anyway: put a rating widget on the page, add the schema, and let users assume Google scored you. Users probably will assume that — that is the whole pitch. It works until a spam classifier or a human reviewer notices there is nothing behind the numbers, and a structured-data manual action is a much worse souvenir than the clicks were worth. The honest version is slower: these tool pages review third-party software, SoftwareApplication is a legitimately eligible type, so if a real rating feature ever ships — real users, real storage — the markup follows the data. Not the other way around.

What shipped

Breadcrumbs, one source of truth

The highest-return fix was the most boring one. A single component now renders the trail and its schema from the same array:

// One array is the single source. The trail you see and
// the BreadcrumbList Google reads are the same data.
const items = [
  { name: 'Home', href: '/' },
  { name: 'Blog', href: '/blog' },
  { name: 'The Stars Were Fake', href: '/blog/…' },
];

const jsonLd = {
  '@context': 'https://schema.org',
  '@type': 'BreadcrumbList',
  itemListElement: items.map((item, index) => ({
    '@type': 'ListItem',
    position: index + 1,
    name: item.name,
    item: `https://saikodev.ai${item.href}`,
  })),
};

Hand-maintaining two copies — one for humans, one for crawlers — is exactly how markup drifts from the page, and markup that disagrees with what the page says is a guideline violation, not a gray area. The component replaced a bespoke implementation on the tool pages too, so there is now exactly one breadcrumb in the codebase. Every route two or more levels deep carries it: posts, notes, the starter-kit pages, tool detail pages.

Notes got Article schema, with one field doing the interesting work

These notes summarize other people’s talks and threads. So the author of the page is this site, and isBasedOn points at the source — the interview, the thread, the talk. Tempting as it is to list the interviewed founder as the author, that would claim they wrote the page. The dates are the dates the notes were logged, matching the sitemap and the visible timestamps.

The root layout now declares who the site is

An Organization and a WebSite node, in one @graph with stable @ids that per-page markup can attach to. The WebSite name is what Google’s site-name system reads for the label above results on mobile — one of the few SERP touches a site fully controls.

And one deliberate omission

SearchAction — the markup that powered the old sitelinks search box — is not there, for two reasons. The header’s search button is decorative; pointing Google at a search URL that cannot answer would be markup describing a page that does not exist. And the feature it fed was removed from results in November 2024 anyway.

Describe the page, never the wishlist

Every decision above is the same decision: markup describes what the page already shows. Not what the page could show, not what a SERP feature used to show, not what would look nice under the title. That is why the boring types won — a breadcrumb, a date, a name are things a page genuinely has. A five-star average it does not have is the one thing it cannot honestly say.

The full menu is documented in Google’s structured data docs, linked below. Read them before copying a recipe from an SEO post — the recipes age faster than the docs.