Title tags and meta descriptions

When an AI engine surfaces your page in an answer, it usually shows a small card: a title, a link, and often a short snippet. Your <title> and <meta name="description"> are the two lines most likely to end up on that card, but what each engine actually does with them differs, and not all of them document it. This guide covers those two describe-the-page tags, plus a myth-bust on the long-dead meta keywords tag.

Other tags in the <head> get their own guides. Open Graph and Twitter cards drive social link previews, charset and viewport are technical plumbing, and the robots and AI directives govern what engines may do with the page. The question here is narrower and more practical: what ChatGPT, Claude, Bing Copilot, Perplexity and Google each actually do with the two lines of text that describe your page, and how to write them so the card they show is the one you meant.

What a meta tag actually does for AI

Think of the title and meta description as your page's citation card. The <title> is the candidate label and the meta description the candidate snippet. Both sit in plain text in the document's <head>, so an engine can lift them without reading the rest of the page. That makes them direct candidates for labels and snippets on other surfaces. Whether a given engine lifts them, rewrites what it finds, or ignores them entirely is engine-specific.

One caveat, said straight: the <title> is technically its own HTML element, not a <meta> tag. People group it under "meta tags" because it lives in the <head> and describes the page rather than appearing in it, and this guide follows that habit, but the distinction is real. The meta description and meta keywords are actual <meta> tags.

The location matters for crawlers. Because both tags sit in the raw HTML, a crawler that never executes JavaScript still reads them on the first fetch. The flip side: a title injected client-side, by a React app or a tag manager, is invisible to any crawler that does not render JavaScript. If your titles only exist after scripts run, everything below applies to a tag those crawlers never see. Which engines do and do not render JavaScript is its own topic.

The whole apparatus is two lines:

<head>
  <title>Alpine Touring Bindings: Weight vs Release Safety</title>
  <meta name="description" content="How touring binding weight trades off
    against release reliability, with lab release values for 12 models.">
</head>

What each engine does with them

The engines differ in how much they document, so it is worth going one by one. Where a vendor has said nothing, that is stated too.

• Bing / Copilot. Bing is the most explicit. Announcing its content controls for Bing Chat, Bing's Fabrice Canel wrote: "Content with the NOCACHE tag may be included in Bing Chat answers. We will only display URL/Snippet/Title in the answer" (Bing Webmaster Blog). Read that as a ceiling, not a promise: NOCACHE content may or may not be included, but if it is, Bing limits what it shows to the URL, title and snippet.
• Claude. Anthropic's documentation makes the title a required field on every source Claude cites: "Search result content blocks let Claude cite your own content the same way it cites web search results: each citation carries the source and title you provided," with the guidance to "Ensure titles accurately reflect content" (Anthropic docs). In this documented API mechanism, the developer supplies the title Claude carries into a citation; it is not evidence that Claude reads the page's <title> element.
• ChatGPT. OpenAI documents a narrow fallback scoped to ChatGPT Atlas, not to ChatGPT generally: for a page it learns of but is not allowed to read, it may surface just the link and page title in Atlas, and noindex prevents even that (OpenAI Publishers FAQ).
• Google. Google's documentation is the classic-search account, and it is thorough. The title link is "often the primary piece of information people use to decide which result to click" (Google title link docs). For the description: "Snippets are primarily created from the page content itself. However, Google sometimes uses the meta description HTML element if it might give users a more accurate description of the page than content taken directly from the page" (Google snippet docs). Note what the title-link documentation does not mention: AI Overviews. Google's AI features draw on the same classic ranking systems, so the title's documented role reaches them indirectly, but Google has not documented a separate AI-specific treatment of the tag.
• Perplexity. Perplexity documents its crawler's purpose: "PerplexityBot is designed to surface and link websites in search results on Perplexity" (Perplexity bot docs). What it does not document is any specific consumption of your title or meta description. Perplexity answers do display source cards with page names, but how those are derived is not published, so for Perplexity specifically, treat the title-as-label mechanism as unproven rather than assumed.

Now the honest ceiling. The evidence above comes from four different contexts: Bing's restricted answer format, Claude's developer-supplied citation data, ChatGPT Atlas's fallback behavior, and Google's classic-search title links. None of them, and none of the research reviewed for this article, ties a <title> or meta-description change to a measured AI-citation lift, and the GEO research that measures citation-rate changes tests only body-text methods such as adding statistics, quotes and sources (GEO paper, arXiv). So write these tags to control what the card says, not because a number promises more citations.

Writing the title

Google publishes the most complete guidance, so it is the sensible editorial baseline even though other engines may derive their labels differently. The starting rule is coverage: "Make sure every page on your site has a title specified in the <title> element." Google says that when a <title> is missing or unhelpful it may build the title link from headings, og:title, prominent text or anchor text; other engines do not document their fallbacks.

Then make it specific: "Write descriptive and concise text for your <title> elements. Avoid vague descriptors like "Home" for your home page, or "Profile" for a specific person's profile." A title's job is to say what this page is about to someone who cannot see the page, which is exactly the position an AI engine's user is in when the card appears in an answer.

Do not stuff it. Google's example is blunt: title text like "Foobar, foo bar, foobars, foo bars" "doesn't help the user", and repeating variants of a keyword reads as spammy to people and engines alike. Related: keep titles distinct per page. Boilerplate titles that repeat across a site make individual pages indistinguishable on any results surface, classic or AI.

On length, the honest position is that there is no rule. Google states no character limit; truncation happens by device width, so the same title can survive on desktop and get cut on a phone. The widely quoted 30 to 60 characters is a soft convention that tends to survive truncation, not a documented threshold, and a clear 65-character title beats a cryptic 55-character one.

One pitfall undoes all of the above: if your framework injects the <title> client-side, make sure the real title is also present in the server-delivered HTML.

Writing the meta description

The first thing to internalize is that the meta description is a fallback, not a guarantee. In Google's words: "Snippets are primarily created from the page content itself. However, Google sometimes uses the meta description HTML element if it might give users a more accurate description of the page than content taken directly from the page." Content first, your description second, and only when it wins on accuracy. How often that happens has been measured outside Google: Portent put the rewrite-or-ignore rate for first-page results at 71 percent on mobile and 68 percent on desktop (Portent study). It is not a ranking factor; its value is entirely in what gets displayed.

That makes accuracy and specificity the whole game. Google again: "Identical or similar descriptions on every page of a site aren't helpful when individual pages appear in search results. Wherever possible, create descriptions that accurately describe the specific page." A site-wide boilerplate description forfeits the one thing the tag can do, which is describe this page.

And it is a sentence, not a keyword list: "Keep in mind that meta descriptions comprised of long strings of keywords don't give users a clear idea of the page's content, and are less likely to be displayed as a snippet." A keyword-stuffed description fails on its own terms; it makes the engine less likely to show it.

One honest nuance: for thin or near-duplicate pages, letting Google auto-generate the snippet from strong body content can beat a weak hand-written description. The tag is worth writing where you have something specific to say about the page; it is not worth generating mechanically for pages where you do not. Length, as with the title, is convention: around 120 to 160 characters tends to display without truncation, but there is no documented limit and clarity outranks the character count.

The meta keywords tag is dead

The third member of the family deserves one tight paragraph, because it still ships on a surprising number of sites. The <meta name="keywords"> tag is a legacy artifact: Google announced in 2009 that it does not use it in web ranking, and Bing said in 2014 that it was dead for SEO value. None of the AI vendors reviewed here, OpenAI, Anthropic or Perplexity, documents using it for retrieval or citation either. It contributes nothing to how any engine represents your page. What it does contribute is bytes on every request and, to anyone reading your source, a signal that the page is maintained by a stale plugin or an old checklist. Remove it.

FAQ

Do title tags and meta descriptions get me cited by AI?

They are the closest thing to a citation card, but each engine differs: Bing limits restricted Bing Chat answers to URL, title and snippet, Claude carries a title supplied with each source, ChatGPT may show a link and title as an Atlas fallback, and Google builds its title link and sometimes uses your description. But no study shows a title or description lift in AI citations, so write them for an accurate card, not a promised number.

How long should my title and meta description be?

Google sets no character limit and truncates title links and snippets to fit the device width; other engines do not publish limits. As a rough convention, keep titles around 30 to 60 characters and descriptions around 120 to 160, but clarity matters more than hitting a number.

Does Google use my meta description?

Only sometimes. Google builds the snippet from page content first and uses your meta description when it gives a more accurate description; Portent measured a rewrite-or-ignore rate of 71 percent on mobile and 68 percent on desktop. Write an accurate, page-specific one and expect it to be a fallback, not a guarantee.