Meta description was assembled by a machine, not written

The check reads the content of <meta name="description"> and looks for two traces that nobody wrote the description: a tool scraped it together. It says nothing about length (that is SEO-02) and nothing about a missing description (STR-03).

First trace: words run together. The description contains a place where a lowercase letter is immediately followed by an uppercase one inside a single word, for example “third-party contentTechniservis”. On its own that proves nothing: brand names look exactly like that (“PrintChart”, “uniFLOW”, “imageFORCE”, “eShop”, “iPhone”). The finding is raised only when the page text confirms it, and both conditions must hold at once:

• the run-together form (stranTechniservis) does not appear in the page text; if it did, it is a word this site uses, not an accident,
• both halves do appear in the page text separated by a space (“…third parties Techniservis s.r.o.…”); that is the fingerprint of the two blocks the tool glued the description from.

The second condition only works because our crawler builds the page text with separators derived from block elements (apps/worker/src/crawler/text.ts): the end of a paragraph and the start of a list item are separated, while inside a word (Tor<b>umata</b>) no space is inserted. The threshold is min_part_letters: 2 from config/meta-description.json: a shorter half is never considered, which filters out one-letter brand prefixes.

Second trace: an administrative tail. The description is split into segments on separators (|, –, —, ·, •), and starting from the end we collect an unbroken run of segments shorter than max_segment_chars: 60 that match a pattern from config/meta-description.json. The strong patterns are document version, effective date, copyright and last-updated date; the weak one is a company name with a legal form. The finding is raised only if the tail contains at least one strong pattern. The first segment (the description proper) is never judged.

What the check does not do:

• It does not judge whether the description is accurate. A description that describes the page badly but reads like a sentence passes.
• It does not flag “| Novák Joinery Ltd” at the end. A company name after a separator is common and fine; on its own it never raises the finding.
• It cannot spot a glued word the page text does not confirm. If the run-together word appears nowhere in the page content (the description came from elsewhere), the check stays silent. That is deliberate: a false alarm on a brand name is worse than silence.
• It does not look for run-together words in Chinese or Japanese: those scripts have no letter case, so there is nothing to find.
• The tail patterns are written for Czech and English. In other languages they simply do not match and the check stays silent; nothing is guessed.

The finding is page-level, severity warning, category structure. It supports “Verify the fix”.

We have no evidence that the meta description affects visibility in AI assistant answers, so we do not claim it does. We put it in the weakest class and this page does not argue otherwise.

What is documented: Google uses the description as one of the sources for the snippet text in a search result, and in the same documentation states that it often rewrites it to match the query. So the meta description is not guaranteed text even for classic search; it is an offer the search engine may or may not take.

Why does the finding exist, then? Not for a measurable effect, but for the one sentence a page writes about itself. When the search engine does use it (and it does), the visitor sees exactly what is in it: “third-party contentTechniservis s.r.o. | Version 1.0 | effective from 14 September 2026”. Nobody wrote that text and nobody would; it is a tool's mistake that is visible to the outside. It is also evidence of a broken template: the same tool builds the descriptions of every other page too.

For completeness: claiming that fixing the description will raise your traffic would be a step too far. All we claim is that the sentence you publish about yourself should make sense.

The quickest test: read the description out loud. What does not work as a sentence will not work in a search result either.

The description should be one clear sentence about what is on the page. These do not belong in it:

• the document version (“Version 1.0”) and the effective date (“effective from 14 September 2026”): they belong in the page text, where the reader sees them in context,
• copyright and legal boilerplate: in a search result they take up room and say nothing,
• a repeated company name when it is already in the <title>.

<!-- Wrong: scraped together by a tool from page blocks -->
<meta name="description" content="Cookies, measurement consent and third-party contentTechniservis s.r.o. | Version 1.0 | effective from 14 September 2026">

<!-- Right: one sentence about what is on the page -->
<meta name="description" content="Which cookies this site uses, what they are for, and how to change or withdraw your measurement consent at any time.">

When an SEO plugin or a template assembles the description (typically on WordPress), fixing one description is only a patch; the tool will scrape it again next time. Fix it where it is produced:

• Yoast SEO: open the “Google preview” block on the page and fill in the meta description field by hand instead of relying on the %%excerpt%% template.
• Rank Math: open “General” in the page sidebar and fill in “Description”; site-wide templates live under “Titles & Meta”.
• Whichever you use, check the template for the whole content type, not just this one page; a run-together word is a trace of how that template assembles text.

Last step: open the source of the finished page and look at the <meta name="description"> tag in the HTML itself. That is the only place where you see exactly what the search engine sees; a plugin preview can show something other than what finally gets generated.