# Meta description
The 1–2 sentence summary in <meta name="description"> that engines fall back to when on-page content isn't extractable. Unique per page, or every page reads as boilerplate.
By AgentSite · 2 min read · Updated 2026-05-24
A meta description is a 1–2 sentence summary in the page `<head>` that engines read as a fallback signal for what the page is about. The relevant tags are `<meta name="description">` (the classic) and `<meta property="og:description">` (the Open Graph sibling). When the body is heavy or hard to extract, the meta description is what the engine quotes.
## The two tags
Both tags ride in the page head and serve adjacent consumers:
```html
<meta name="description" content="The 1–2 sentence summary of this page.">
<meta property="og:description" content="The same summary, for Open Graph consumers.">
```
[MDN defines `<meta name="description">`](https://developer.mozilla.org/en-US/docs/Web/HTML/Element/meta/name) as "a short and accurate summary of the content of the page… search engines like Google use this metadata to adjust the appearance of a webpage in search results." [The Open Graph protocol](https://ogp.me/) defines `og:description` as "a one to two sentence description of your object" (optional in OG, but treated as load-bearing by most social and AI consumers).
Most sites ship both with the same string. Skip one and you lose that consumer's preferred read.
## When the engine actually uses it
[Google's snippet documentation](https://developers.google.com/search/docs/appearance/snippet) is explicit: "Snippets are primarily created from the page content itself" — the meta description is used "if it might give users a more accurate description of the page than content taken directly from the page." Same principle applies to AI retrieval: the body wins when the body is extractable; the meta description wins when the body is hidden behind JavaScript, locked in a paywall, or buried under hero-art markup.
For a server-rendered page with a clean [direct answer](/direct-answer) in the lede, the meta description is the second signal. For a SPA with an empty shell, it is the only signal the non-rendering crawler gets.
## Anti-patterns
Three failures show up reliably:
1. **Site-level boilerplate.** Every page in the site ships the same description ("AgentSite — the AEO platform"). The engine has no per-page distinguishing signal, and the description carries zero weight against on-page extraction.
2. **Missing tag on inner pages.** The homepage has a description; the 200 inner pages don't. Engines treat each as undescribed and fall back to extracting from whatever bytes they find, which on a SPA is often nothing.
3. **Too-long descriptions.** Anything over ~160 characters gets truncated in search results and weighs against the page in extractors that compare lede length against description length. Two tight sentences are stronger than four flabby ones.
Unique per page, written from the same template the body uses, kept inside ~160 characters.
## Where this fits
Meta description is a Layer 3/4 boundary signal — declared metadata that crosses into content-quality territory because it's what the engine quotes when the body is unreachable. It pairs with [page title](/page-title) and [og:image](/og-image) as the head-tag triplet, the [direct answer](/direct-answer) lede (body equivalent), and [FAQ schema](/faq-schema) (the richer Q&A alternative). The layered place is in [the five layers of AEO](/five-layer-aeo); the catalog of failure modes is in [AEO problems](/aeo-problems).