SEO metadata helps describe a page, communicate indexing preferences and supply information for link previews. Start with an accurate title and description, the right canonical URL, intentional robots rules and valid HTML. Then check the HTTP response, initial HTML and rendered DOM separately. A correct CMS field is useful only if the intended value reaches the published page.
These controls have different jobs. A title is a <title> element; canonical and hreflang use <link>; descriptions and many robots or social properties use <meta>. Calling all of them “meta tags” is convenient, but treating them as interchangeable can cause real implementation mistakes.
Which settings deserve attention first?
| Job | Setting to inspect | What a correct value does not guarantee |
|---|---|---|
| Describe the page in search | Title, meta description and visible content | Google displaying your exact wording or a higher click-through rate |
| Control indexing and snippets | Robots meta tags and X-Robots-Tag response headers | Crawling access, authentication or an immediate search-result change |
| Identify preferred and localized URLs | Canonical and, when needed, hreflang | Google choosing your preferred canonical or accepting inconsistent alternatives |
| Support reading and sharing | Charset, viewport and Open Graph properties | A responsive layout, an accessible page or an identical preview on every platform |
When a page unexpectedly disappears, investigate access, status codes and indexing rules before polishing its description. When it is indexed but attracts few clicks, examine the actual query, result appearance and page promise. The discussion of high impressions and low clicks addresses that adjacent question; a description edit alone does not identify the cause.
Write a title and description that the page can deliver
A useful title identifies the subject and the reason to choose the page. “How to Measure a Cupboard for Storage Boxes | Example” is more informative than “Article 27” for a measuring guide. The visible heading, main content and title should describe the same task. Avoid repeating a string of keyword variants or adding a year that the content does not substantiate.
Google generates title links from several sources, including the title element and prominent page text. There is no fixed title-character limit that guarantees full display; truncation depends on the available space. A preview tool can help assess wording, but it cannot predict every result.
The description should summarize the page's useful answer. For the cupboard guide, it could mention checking usable space, lid clearance and product dimensions. It should not claim a tested fit or guaranteed result unless the page supplies that evidence. Google's snippet guidance explains that snippets primarily come from page content and may use a suitable meta description. They can change with the query; a saved description is not a command to display that sentence.
Write descriptions for the individual page rather than applying one sales sentence to an entire archive. For a large catalogue, a well-designed template can use real product attributes, but inspect its output for missing values, repetition and unsupported promises. Keep paid-search ad performance separate from organic search observations.
Robots rules: indexing, crawling and privacy are different
noindex tells a supporting search engine not to index a page. The crawler must be able to fetch the page to see the rule. Blocking the URL in robots.txt can prevent it from reading the noindex instruction; it is not a reliable substitute for removing that URL from search.
The Google robots specification covers both HTML meta tags and the X-Robots-Tag HTTP response header. Check general and crawler-specific rules together. For conflicting applicable rules, Google applies the more restrictive rule. An HTML index,follow value does not cancel an applicable HTTP noindex header.
- index,follow: the default permissions do not force indexing, ranking or link processing.
- noindex: an indexing instruction, not a password or a way to prevent all downloads.
- nofollow: concerns links on the page; it does not mean “do not fetch this page.”
- none: equivalent to noindex,nofollow, not a robots.txt crawling block.
- nosnippet and max-snippet: control supported search snippets. They are separate from the description text you would like displayed.
Use authentication and appropriate server permissions for private material. For a public PDF or another non-HTML resource, the response header can carry an indexing instruction without an HTML head. For a public article, also inspect whether snippet restrictions were intentionally set before removing them. Retired Google controls such as noarchive should not be presented as current tools for controlling cached search copies.
Canonical and hreflang: choose the right relationship
A canonical link identifies your preferred URL among duplicate or very similar pages. It is a signal, not a guaranteed selection. Google's canonical guidance recommends consistent signals: internal links, redirects where appropriate and sitemap URLs should support the intended choice. Do not use noindex as a substitute for canonicalizing duplicate pages.
Pagination needs particular care. Page two of a collection normally contains different items from page one. Google's pagination guidance says to give each page its own canonical URL rather than canonicalizing the whole sequence to its first page. Keep crawlable links between pages. Loading another screenful with JavaScript does not by itself supply those URLs.
hreflang describes language or regional alternatives. Use it when genuinely corresponding versions exist, with supported language codes and optional region codes. Each participating page needs the applicable set of links, including itself and return links from its alternatives. Use fully qualified URLs. x-default is an optional fallback annotation, not a replacement for the other versions.
Follow Google's localized-version guidance and coordinate hreflang with canonical selection. Do not canonicalize all translations to the English page merely because it was published first. A translated header around an otherwise unchanged article also needs a considered language strategy; adding a language code does not translate the content.
A small HTML starting point
This fictional English guide shows a basic head, not a complete website or an instruction to replace your CMS template. Replace the example URLs, wording and image with your own verified resources. Add robots restrictions or language alternatives only when the page requires them.
<!doctype html>
<html lang="en">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1">
<title>Cupboard Measuring Guide | Example</title>
<meta name="description" content="Learn what to check when measuring cupboard space for storage boxes.">
<link rel="canonical" href="https://example.test/guide">
<meta property="og:title" content="Cupboard Measuring Guide">
<meta property="og:type" content="article">
<meta property="og:url" content="https://example.test/guide">
<meta property="og:image" content="https://example.test/cupboard.webp">
</head>
<body>
<h1>Cupboard Measuring Guide</h1>
<p>Fictional example: replace this with accurate measuring instructions.</p>
</body>
</html>
The charset declaration must match the document's encoding and be within its first 1,024 bytes. A UTF-8 label cannot repair content served in an incompatible encoding. The viewport setting above is a common starting point. Allow zoom; avoid user-scalable=no and unnecessarily restrictive maximum-scale values. Responsive CSS, image selection and keyboard usability still need their own checks.
Keep the head valid. Google's metadata guidance warns that invalid head elements can cause later metadata to be ignored. A banner image or content div belongs in the body. Inspect the parsed document too: a browser may move elements when repairing invalid markup.
Social previews: Open Graph is the starting vocabulary
The Open Graph protocol lists four required basic properties: og:title, og:type, og:image and og:url. A description is useful additional metadata, but it is not one of those four. Supply an absolute image URL that the intended platform can fetch and decode. If you supply image dimensions or other structured properties, ensure they describe the actual image.
The protocol does not guarantee one crop, format policy, maximum file size or cache-refresh behavior across all consuming services. Check the current requirements of each platform you use, then test a real share there. A successful browser request using your session does not prove the platform's crawler can access the same resource.
You may also encounter twitter:card and other X-specific properties. The January 12, 2026 archive of X's large-image card documentation documents the historical summary_large_image value. It is a historical reference: the old live documentation URLs redirected to a general developer page during this guide's research. Do not use that archive to assume today's image limits, Validator interface or cache-refresh behavior. Those need current platform verification.
A real local test: why checking only the DOM misses a problem
For this guide, a loopback HTTP server returned a fictional page with status 200 and an X-Robots-Tag: noindex header. The initial HTML contained the title “Generic article,” robots “index,follow” and canonical “https://example.test/guide.” A script changed only the title to “Shelf measuring guide | Example.”

Chromium with JavaScript enabled showed the changed title. With JavaScript disabled, it kept the initial title. Both responses still carried the noindex header, and both DOMs still contained index,follow. This observed mismatch demonstrates why the three checks are complementary. Under Google's documented robots rules, the applicable noindex restriction would still matter; the experiment itself did not test Google indexing, a social network or a CMS installation.
To inspect your own URL, request the actual document with GET and save both headers and body. Replace the example URL; these filenames refer to local output files.
curl --location --silent --show-error \
--dump-header response-headers.txt \
--output response.html \
'https://example.test/guide'
When redirects occur, the saved header file contains multiple response blocks. Match the final document body to its final response, while retaining the earlier redirect chain. Check status, Content-Type and all applicable X-Robots-Tag values. A HEAD-only request may not reproduce a GET response. If access depends on cookies, geography or a firewall rule, record those conditions rather than treating one response as universal.
- Read the initial HTML. Locate the title, description, canonical and robots elements in the actual response. Look for duplicates, empty template values and elements outside a valid head.
- Inspect the rendered DOM. Use the browser's Elements panel after the page settles. Compare changed values and verify whether a plugin or script added another element.
- Check the response layer. In Network, select the main document and inspect response headers and redirects. An HTML screenshot will not reveal an HTTP noindex header.
- Check the consumer separately. Search Console evidence, a real search result and a platform's sharing preview answer different questions. Record which one you actually observed.
Fix the setting that produces the wrong output
If a CMS field is correct but the public response is wrong, trace the template, SEO plugin, reverse proxy and cache in that order as applicable to the site. A plugin may store the intended description while another component outputs a duplicate. A server rule may add noindex to every response. Update the responsible component, refresh affected caches and request the public URL again.
Keep adjacent controls distinct. A Referrer-Policy governs referrer information sent with requests, not whether the document is indexed. For a permanent URL move, use an appropriate server redirect rather than relying on a delayed meta refresh; Google's redirect documentation distinguishes their handling.
After a change, save the old and new output, confirm the intended canonical and indexing rules, and verify page reading and sharing separately. Correct output is a technical result. Indexing, search wording, traffic and enquiries require subsequent observations. The Google-ranking implementation checklist covers the broader work; Guangsuan's Google SEO service (Chinese) explains the company's technical, content and link-service scope when the problem extends beyond one template.