How to write title tags and meta descriptions

Titles and meta descriptions get written in five minutes and then left alone for years. Yet they are what a person sees in the results before deciding whether to open the page. This article covers how Yandex and Google use the two elements, why length should be measured in pixels rather than characters, why a tool can report a 56-character Russian title as 92 characters long, when each engine replaces what you wrote, and which templates work for category pages, articles and tool pages.

What the title and description do

The <title> element names the page, and the <meta name="description"> tag summarises it. Both sit in the <head> and are not visible on the page itself. Search results build the snippet from them: the clickable title and the text beneath it.

Neither engine promises to show them word for word. Yandex's help page on titles and descriptions (checked 15 October 2026) lists the sources it builds a snippet from: the title element, the description meta tag, the page text including Schema.org markup, plus Yandex Market and the text of links to the page from other sites. For the description, the same page says it uses whichever text is most informative and relevant to the query.

Google's documentation on title links (last updated 10 December 2025, checked 15 October 2026) calls title generation completely automated and lists sources beyond the title element: the main visual title on the page, <h1> and other heading elements, the og:title meta tag, other large and prominent text, anchor text on the page, the text of links pointing to it, and WebSite structured data. Its snippet documentation (last updated 20 April 2026) says snippets are primarily created from the page content, and that Google sometimes uses the meta description if it gives a more accurate description of the page.

Hence the first rule: the title and description should match what is on the page. When they drift apart, the engine is more likely to take text from the page instead, and the snippet ends up being one you did not plan.

Length: count pixels, not characters

Everyone knows "60 characters for the title" and "160 for the description", but neither figure comes from a search engine. Google's documentation says there is no limit on how long a title element can be, and that the title link is truncated as needed, typically to fit the device width; it says the same about meta descriptions. Yandex's help page on writing a title puts it similarly: title length is limited by the width of the user's screen, and titles that are too long may be shortened in the results. Neither engine publishes a character limit.

Truncation goes by width, and width depends on the letters. A narrow i or l takes less than half the width of a Cyrillic ш or ж. We measured this ourselves in Liberation Sans, a font metrically compatible with Arial, at 20 pixels. The lowercase Russian alphabet averages 11.6 pixels a letter, the English one 9.8. The Russian title of our homepage, "SEO, ИИ и дизайн-инструменты в одной подписке — gbseo.ru", is 56 characters and 576 pixels wide. The English version of the same page, at 55 characters, is 522 pixels.

The exact font and size of the results page are not documented anywhere and change with the interface, so treat the absolute numbers as a guide. The relative ones hold: Russian text of the same character count runs about a tenth wider than English. If your tool applies an English rule of thumb to titles, a Russian title it passes may still be cut off, and one it flags may fit comfortably.

The practical upshot: put the substance of the title in the first 40 to 45 characters so truncation cannot eat the meaning, put the brand at the end where losing it hurts least, and check the width in a snippet preview rather than with a character counter.

Cyrillic and bytes: when a tool is off by almost double

One measurement error only hits pages in non-Latin scripts. An OG markup checker told us the og:title on gbseo.ru's homepage was 92 characters, over the limit. It is 56. The 92 is the length of the same string in UTF-8 bytes: it contains 34 Cyrillic letters, and each Cyrillic letter takes two bytes in UTF-8.

The giveaway is easy to spot if the page has an English twin. The English title, built on the same pattern and almost the same length (55 characters, 57 bytes), passed the same check without a warning. When a tool flags one language version and stays silent about the other, check the tool first. Recounting takes one line: in Python, len(s) gives characters and len(s.encode('utf-8')) gives bytes. If the "length" in the report matches the second number, the report is counting bytes.

Russian descriptions suffer from this even more. The 115-character description on gbseo.ru's homepage weighs 209 bytes. A tool that counts bytes will show it as 209 characters and advise cutting in half a text that is already shorter than most snippets.

Even a correctly counted 60 characters applies to the title element in search results. og:title appears in social and messenger link cards, each with its own width and its own truncation rules.

When Google rewrites a title

Google uses your title in most cases, but not all. In September 2021 the Search Central blog said that after changes to its system, title elements were used around 87% of the time, up from around 80% before, and that Google had used text beyond title elements since 2012 (checked 15 October 2026). The figure is five years old, and we found nothing more recent in Google's documentation.

The documentation names specific reasons a title gets replaced:

• Half-empty titles, where part of the text is missing, for example a template that filled in the site name and lost the page name.
• Obsolete titles, where a page is reused year after year for recurring information but the year in the title was never updated.
• Inaccurate titles that do not reflect what the page is about.
• Micro-boilerplate: repeated template text in the titles of a group of pages that differ by one word.
• No clear main title, when the page has more than one large heading and it is not obvious which is the main one.
• A title whose language or writing system does not match the main text of the page.

The same documentation asks you to avoid keyword stuffing, brand titles concisely, and use the same language and writing system as the page's main content. For Russian pages the last point applies directly: an English title on a Russian page, or a transliterated one instead of Cyrillic, gives Google a reason to replace it.

If Google consistently shows a title other than yours, start with that list. When the title is generic or templated and the <h1> is specific, Google has an alternative to hand: headings come right after the main visual title in its list of sources.

How Yandex builds the snippet title and description

Yandex's help says outright that the snippet title can be formed from the page content, so users may see different titles depending on their query. The description follows the same rule: the snippet gets whichever text the engine judges most informative and relevant to the query. The same page can look different in the results for different queries, and that is expected behaviour.

Yandex has a separate help page on writing the description meta tag (checked 15 October 2026), and it makes several direct requirements:

• add the tag to every page, and to the important ones first, such as catalogue pages;
• write a unique description for each page; template or similar descriptions will most likely not be shown in search results;
• use properly formed sentences, without overusing keywords, capital letters or advertising slogans;
• include information useful to the user, such as price, colour or delivery city;
• match the page's language, and differ from the content of the title.

The same page gives examples of poor descriptions: temporary or internal information, outdated content, important information placed at the end, and a description only a few words long. It recommends putting the most important information first. That fits the truncation rule: the start of a description is always visible, the end not always.

Separately, Yandex recommends adding the brand to the title, at the start or the end, separated by a colon, a hyphen or a vertical bar, and advises against titles like "<Product name> — description, prices, specifications, photos, reviews". If the page has no reviews or photos, those words in the title only mislead. Substitute only what is actually on the page and move the rest into the description.

How to write a title

A working order for a single page:

1. Take the main query of the cluster the page was built for. If the page was built without a keyword core, first work out which query it answers; that is decided when you build the semantic core, not when you write the title.
2. Look at the results for that query: what type of page ranks, how the titles are phrased, what they have in common. How to read a results page and what to take from it is covered in analysing competitors in search results.
3. Put the query, or a close phrasing of it, at the start of the title. It does not have to be verbatim: "How to build a semantic core for a site" reads better than "semantic core build how to".
4. Add what sets the page apart: specifics, format, a number, a region. One detail is enough.
5. Add the brand after a separator at the end and check the width in a preview.

The title does not have to match the <h1>, but it should say the same thing. A convenient split is a longer, more conversational <h1> and a shorter title closer to the query. A year belongs in the title only if the page really is updated every year; otherwise, twelve months on, it becomes the obsolete title from Google's list.

How to write a description

The description answers what a person gets by opening the page. A good one is a sentence or two that names what the page actually contains and does not repeat the title.

• The first sentence says what is on the page: an answer, a list, a comparison, a set of steps.
• The second adds a detail the title lacks: how many items, for which engine, for which case.
• Commercial pages use facts from the product record: price from, availability, delivery, number of models. Yandex's help explicitly names price and delivery city as useful information.
• No "best", "cheapest" or "call now". Yandex lists advertising slogans among the things a description should not contain.

Google's snippet documentation adds that programmatic generation of descriptions can be appropriate and is encouraged, provided the descriptions are unique and accurate. For an article, it suggests the author and the publication date. Generation pays off on large catalogues, where describing every product by hand is impossible.

Templates for categories, articles and tool pages

A template works when the values it fills in differ from page to page and are genuinely present on the page. If one word changes and the rest stays the same, the result is exactly the repeated boilerplate that makes Google replace titles and Yandex skip descriptions.

For articles a description template is close to useless: a good article description is written by hand and takes a minute. Catalogues are the opposite case, and Yandex recommends filling descriptions automatically from database fields such as size, weight and availability. The main thing is that no variable is ever empty. An empty field in a template produces the half-empty title from Google's list.

One more rule for templates: before rolling one out to a thousand pages, export ten results from different sections and read them yourself. Grammar errors, empty fields and double spaces are obvious at a glance.

How to check, and what usually breaks

Yandex Webmaster has a Titles and descriptions page that shows duplicate titles and descriptions across the site. Its diagnostics section also warns when pages have no title or description, or when the site has many duplicates. Google Search Console has no equivalent report, but the Performance report filtered by page shows impressions and click-through rate by query. If the position is stable and the click-through rate is noticeably lower than for neighbouring pages at similar positions, the snippet is worth revisiting.

Check what the server sends, not what the CMS shows. curl -s https://example.com/page/ | grep -i '<title' prints the title the crawler will see. An SEO plugin and a theme template often each output their own title, leaving two in the code. How to catch this kind of thing systematically is covered in the technical SEO audit checklist.

Changes do not show up in the results straight away: a new title reaches the snippet only after the page is crawled again. To speed that up, submit the page for reindexing in Webmaster and request indexing in Search Console; how to confirm the page has been recrawled is covered in how to check whether your site is indexed.

Common mistakes:

• Identical titles on pagination, filter and sort pages.
• A homepage title that is just the site name, or a section titled "Catalogue".
• A comma-separated keyword list instead of a phrase.
• A year in the title that nobody updated.
• A description copied from the first paragraph, which the engine could take for the snippet anyway.
• An English description on a Russian page, left over from the theme.
• Cutting length to satisfy a tool that counts bytes.

Frequently asked questions

How long should a title be?

Neither Yandex nor Google sets a character limit: titles are truncated to the screen width. Put the substance in the first 40 to 45 characters, the brand at the end, and check the width in a snippet preview.

Why does my tool say my Russian title is too long?

It may be counting bytes rather than characters. A Cyrillic letter takes two bytes in UTF-8, so 56 characters become 92. Check whether the "length" in the report matches the byte count.

Does the meta description affect rankings?

The help pages cited here describe the description only as one source of snippet text and say nothing about ranking. What the snippet does affect is whether a person clicks through to the site.

Why is a different title showing in search results?

Both engines can build the title from the page content. Google's documentation lists the reasons: a half-empty, obsolete or templated title, or one that does not match the content. Yandex may show different titles depending on the query.

Do I need to write titles and descriptions by hand for every page?

For articles and key pages, yes. For large catalogues both engines accept generation from database fields, as long as the result is unique to each page and has no empty values.

How soon will a new title appear in search?

After the page is crawled again. Reindexing in Webmaster and an indexing request in Search Console speed this up.