Page titles: length, duplication, and the ones Google rewrites
The title tag is the only piece of copy on your site that has to work in four places at once: a results page, a browser tab, a shared link and a bookmark. Most advice treats it as a keyword slot. It is closer to a headline with a character budget and a hostile editor.
The short answer
A title tag survives when its first few words identify the page on their own, because every surface it appears on truncates from the end. Aim for roughly 50 to 60 characters, put the specific thing first and the brand last, and never template one across a section. Google rewrites titles it judges a poor fit for the page.
Where the title actually shows up
The <title> element does four jobs, and the constraints come from the narrowest one. On a results page it is the clickable line, truncated to fit. In a browser tab it gets a couple of dozen characters at most. In a bookmark list it is the whole entry, forever, with no surrounding context. And it is the fallback when a shared link has no Open Graph title.
Which means the useful test is not length. It is whether the first few words identify the page when everything after them is cut off, because on the narrowest surface that is what happens.
What is checked here, and what is a sibling check
Presence first: a page with no <title>, or one holding only whitespace, fails outright. Then length, against a band rather than a limit, because the pixel width where truncation kicks in depends on the characters used and no character count is exact.
Two related checks sit alongside it. Unique titles compares every title across the crawl and flags collisions, which is a different defect with a different cause. This one is about the individual page.
Check one page for its title
Paste a URL. This reports the exact title we read after rendering, its length, and whether that length sits in the band where titles usually survive intact.
No signup required. Each free search audits one page, paste any URL to see it in action.
Why Google replaces titles, and what triggers it
Google has stated openly that it generates its own title text when it judges yours a poor fit, and the substitute is usually drawn from your H1, from anchor text pointing at the page, or from the site name. It is not a penalty and there is no report telling you it happened. You find out by searching.
The conditions that invite a rewrite are consistent enough to design around. Boilerplate repeated on every page, keyword stuffing, a title that is mostly the site name, a title that contradicts what the page visibly says, and emptiness. A rewrite is feedback: the system had something better available, which means your title was carrying less information than the page itself.
The way to keep your title is to make it the most accurate description of the page available, not the most optimised one.
Writing one that survives
Specific thing first, qualifier second, brand last if at all. On a page that answers a question, the question is the title.
<!-- Padded with boilerplate. The distinguishing
words are at the end, where they get cut. -->
<title>Home | Widgets, Gadgets & More | Acme Co UK</title>
<!-- Brand-first on every page: 40 identical
characters before anything page-specific. -->
<title>Acme Co - Stainless steel widget, 40mm</title>
<!-- Distinguishing words first, brand trailing. -->
<title>Stainless steel widget, 40mm | Acme Co</title>In a framework, the title comes from a template plus a per-page value, and the template is where the boilerplate creeps in. Keep the shared part to the brand and keep it at the end:
export const metadata = {
title: {
// %s is the page's own title. Everything outside it
// is repeated on every page, so it stays minimal.
template: '%s | Acme Co',
default: 'Acme Co: stainless steel widgets',
},
}The template pattern that duplicates every title
The most common way to fail the sibling uniqueness check is not laziness, it is a template reading from the wrong field. A listing template that titles each page from the category rather than the item gives every product in a category the same title. A paginated archive that titles from the archive name gives pages two through nine the title of page one.
Both look correct in the CMS, because the field being read genuinely has a value. The symptom only appears in aggregate, which is why a crawl finds it and spot-checking never does. The fix is to include the discriminator, the item name, the page number, in the template itself.
If pagination is involved, put the page number in the title.
Not for the reader, who can see which page they are on, but because it is the only thing distinguishing nine otherwise identical documents. Untitled pagination is the single largest source of duplicate titles we see, and it is invisible without a crawl.
Why titles are Critical and descriptions are not
Both are head elements, both affect a search result, and they are rated differently on purpose. A missing description costs you control of the snippet: the engine writes one from your page, and it is frequently decent. A missing title costs you the clickable line itself, and the substitute is assembled from whatever is nearest to hand.
There is also the accumulation argument. A title is reused in tabs, bookmarks, shared links and internal search, so a bad one is wrong in five places rather than one. That breadth is what the severity scale is measuring.
Confirming a rewrite has stopped
Change the title, then search for the page with a site: query and look at what is printed. That is the only place the answer exists: view-source tells you what you sent, not what was chosen. Expect a lag, since the change has to be recrawled before it can be reconsidered.
Re-crawling here confirms the mechanical part, one title, right length, distinct from its siblings. The editorial part, whether Google now prefers yours, is a separate question and the site: query is how you ask it.
Does this affect AI search?
Not directly, and the reason is worth understanding rather than taking on trust. An assistant summarising your page has the whole page. It does not need a 60-character headline to work out what the page is about, and it is not choosing between your title and a competitor's in a ranked list.
Where the title carries weight is as a label rather than a summary. When an assistant prints a citation it needs something to call the source, and the title is the obvious candidate. A page titled "Home" cited in an answer tells a reader nothing about where the claim came from, and a citation nobody clicks is worth very little.
So the honest position: write titles for the results page, and accept the citation label as a side effect you get for free. The checks that genuinely determine whether a model can use you at all are access and readability, not head-element copy.
Questions this check raises
- How long should a title tag be?
- There is no character limit, only a pixel width, and it varies by which characters you use. Roughly 50 to 60 characters usually survives intact on desktop. The more useful test is whether the first 30 characters alone tell someone which page this is, because that is what a narrow browser tab shows.
- Why did Google change my title in search results?
- Google generates a replacement when it judges your title a poor description of the page, usually because it is boilerplate repeated site-wide, stuffed with keywords, mostly your brand name, or contradicted by the visible content. There is no report telling you it happened and no penalty attached. The fix is to make your title the most accurate description available, not the most optimised.
- Should the title tag match the H1?
- They can share wording but should not be identical by default. The title is written for a results page where it competes with nine others and has no surrounding context; the H1 is read after the click, with the whole page around it. Copying one into the other wastes whichever one had the better job to do.