Anatomy of a SERP: 28 pages for SEO optimisation, dissected

We took the 28 site pages at the top of Google Romania for "SEO optimisation" — those who sell SEO and those who explain it — and x-rayed them mechanically, the way a robot reads them: the served HTML, not the impression in the browser. The result: length is not missing (the median is 3,519 words), but 21% have no schema at all, only 18% have structured FAQ, 0% publish a price — not even the page with "prices" in its title — and 11% promise "guaranteed", although Google states in writing that no one can guarantee the #1 position. It is the anatomy of the market that sells you visibility. No names: we publish patterns, not companies.

How we measured

The SERP: the first 38 positions for "optimizare seo", Google Romania, Ubersuggest snapshot from 28 July 2026. Of those, 30 are organic web-page results; we removed two that are not sites (a video and a social-network page) and were left with 28 pages, the encyclopaedia included. On 7 September 2026 we downloaded all of them with a script — each answered 200 — and extracted mechanically, from the served HTML: title and description (lengths), number of H1s, JSON-LD types, presence of structured FAQ, number of visible words in the text, any price mention (lei, RON, euro), the word "guaranteed", declared modification date, canonical, hreflang and the final protocol (HTTPS or not).

What the script sees is what an AI extractor sees: the HTML, not the JavaScript. What is born only in the browser does not appear — and that is, in itself, a finding.

The numbers

Source: our own script on the served HTML, 7 September 2026; the SERP from the Ubersuggest snapshot of 28 July 2026.

The five patterns

Length is not the problem. What can be seen from it, is.

The median of 3,519 words says everyone has learned the "long content" lesson. But 21% of pages have no structured tag at all, and only 18% have put their frequently asked questions in FAQPage format. For a classic search engine, long text is enough; for an AI assistant looking for the answer, a page without schema and without structured questions is a forest without paths. Whoever measures the reference rate knows this is where it gets decided whether you appear in the answer.

Zero prices, in a market that sells

28 out of 28 do not publish an amount. Not even the page whose title starts with "prices": zero mentions of lei or euro in the text. The market convention is "request a quote". The consequence for the customer: they cannot compare anything before talking to someone; the consequence for search engines: the queries with "price" and "how much" — among the most searched in any niche — have no honest page to cite. We publish prices on the pricing page; it is not courage, it is the simplest way to be cited on the question everyone asks.

"Guaranteed" — the word that contradicts the documentation

Three pages out of 28 promise "guaranteed" results. Google's documentation, updated in June 2026, says: "No one can guarantee a #1 ranking on Google" and warns you to beware of those who guarantee. An attentive buyer reads both and draws their own conclusion.

Basic hygiene is missing exactly where hygiene is sold

36% titles too long, 32% descriptions too long, 7% without a description, 18% with more than one H1 — including the page in position 1, with 21. Five pages out of 28 stay on HTTP if you come in on http (they do not redirect to HTTPS). 14% have the year in the title, which will be false in January (why it matters). None of these decides the position alone; all together they say how careful the one promising you care actually is.

The invisible page

One of the 28 has, in the served HTML, 11 words. The rest appears only after JavaScript runs. Google can render it, with a delay; an AI extractor cannot. It ranks today on an engine that still executes scripts and is invisible to those that do not — a position borrowed from technology, not earned by content.

What it means for you, if you want to get onto this page

The bar is not "write more" — everyone has already cleared that. The bar is:

1. Structure a machine can read: Article, FAQPage, Organization, author, bibliography — the schema markup guide shows exactly how.
2. The answer in the first sentence, with the frequently asked questions in a readable format, not only in text.
3. The price in plain sight, or at least its logic — you would be the only one of 28 doing it.
4. No guarantees. A measured process, not a promise.
5. The small hygiene: one H1, title under 60, description under 158, forced HTTPS, declared update date, no year in the address.

Five things you control entirely, in a market where 21% have not done the first one.

What this analysis does NOT say

• It is one SERP, on one day. Positions move; the snapshot is from July, the x-ray from September. The patterns are stable; the order is not.
• We read the served HTML, not the rendered page. A schema injected via JavaScript does not appear in our figures — but not in the AI extractors' either, which is still a finding.
• Prices in images or PDFs cannot be seen in the text. We would have counted them if they were text; they were not.
• We do not name the pages. We publish patterns, not companies — the house rule. Whoever wants to verify can run the same x-ray in an hour; the method is described above.