Do meta descriptions affect rankings?

No, and Google has said so directly since 2009. They affect click-through rate, which is a different mechanism: the description is the sales copy under your title, and a result nobody clicks earns nothing regardless of where it ranks. Treat it as advertising, not as optimisation.

Why is Google showing different text than my meta description?

Google replaces the description whenever it thinks a passage from the page answers the query better, which is the majority of impressions for most sites. That is usually a good outcome, since a query-matched snippet converts better than a generic one. It becomes a problem only when the text it picks is boilerplate or navigation, which is a content structure problem rather than a meta tag one.

Is it better to leave a meta description blank than write a bad one?

Yes, on pages where you have nothing specific to say. A blank tag lets the engine assemble a snippet from the query and the body, which beats a generic sentence repeated across 400 product pages. Write descriptions where the pitch is genuinely different per page, and skip them where it is not.

Meta descriptions: what they do, what they do not, and when to skip one

Google confirmed years ago that the meta description is not a ranking factor, and a lot of people stopped writing them. That was the wrong conclusion from a true premise: it never decided where you rank, it decides whether anyone clicks the position you already earned.

The short answer

Meta descriptions have not been a ranking signal for years, and they still decide whether a result gets clicked. Google uses yours when it matches the query well and writes its own otherwise, which is most of the time. Write one for pages where you have a specific pitch, and leave it blank where you do not.

ImportantAudit check · Meta descriptions

Not a ranking factor, and still worth writing

Two things are true at once here and most guidance picks one. The description does not influence position: no amount of keyword placement in it moves you up a results page. And it is frequently the only sentence a searcher reads about you before deciding between you and the result below.

Treat it as advertising copy with a fixed budget, judged on one metric, whether the click happens. That reframing changes what you write. Keywords stop being the objective and become incidental, and the question becomes what a person needs to know to choose this result.

What counts as present and in range

One <meta name="description"> per page, non-empty, in a length band wide enough to be useful and short enough to survive. Missing is a failure. Present but outside the band is reported as worth a look, not as an error, because a short description on a thin utility page is a reasonable choice.

Duplication across pages is a separate check, and it fires more often than absence. A single description written once and inherited by an entire section passes this check on every page and fails that one on all of them.

Check one page for its description

Paste a URL. This reports the description we read, its character count, and whether the tail is likely to be truncated.

Check one page for its meta description

No signup required. Each free search audits one page, paste any URL to see it in action.

When Google uses yours, and when it writes its own

Your description is a candidate, not an instruction. Google generates its own snippet whenever it judges that a passage from the page answers the query better than your submitted text, which happens constantly on pages that rank for a range of different queries.

That behaviour is more sensible than it first sounds. One static sentence cannot be the best answer to twenty different searches, so a page ranking for many long-tail queries will see its description swapped often. A page ranking for one navigational query, typically a homepage or a product page, tends to keep what you wrote.

So invest the writing time where the query is predictable, and stop trying to force a snippet on pages whose traffic is a long tail.

Writing one worth keeping

Lead with the specific claim, not with a restatement of the title. The description sits directly underneath the title on the results page, so repeating it wastes the second line entirely.

descriptions.html

<!-- Restates the title, then pads. Nothing here
     that the line above did not already say. -->
<meta name="description"
      content="Stainless steel widget 40mm. We offer the
               best widgets for all your widget needs.">

<!-- Adds what the title could not fit: the spec that
     decides the purchase, and the shipping fact. -->
<meta name="description"
      content="316-grade stainless, 40mm bore, rated to
               12 bar. In stock, ships same day from the
               UK, with a 10-year corrosion warranty.">

In a framework, the description belongs next to the page it describes rather than in a shared layout. A layout-level default is a good safety net and a bad description:

app/widgets/[slug]/page.tsx

export async function generateMetadata({ params }) {
  const widget = await getWidget(params.slug)

  // Built from the item's own fields, so no two pages
  // can inherit the same sentence. A hand-written
  // summary wins when there is one.
  return {
    description:
      widget.summary ??
      `${widget.material}, ${widget.bore} bore, rated to ` +
      `${widget.pressure}. In stock, ships same day.`,
  }
}

The pages where leaving it blank is correct

This is the part most checklists refuse to say. On some pages a hand-written description is worse than none, and writing one anyway is busywork that makes the page slightly worse.

Large catalogues and archives. Ten thousand products cannot have ten thousand hand-written descriptions, and the generated ones will all follow one pattern, which fails the duplication check. Either template from genuinely distinct fields or leave it to the engine, which will pull the passage matching the query.
Long reference pages. A documentation page answering forty related questions has no single best sentence. Google picking the passage that matches the search is better than you picking one for all of them.
Pages you have not read recently. A stale description that describes what the page used to say is actively misleading. Absent is better than wrong.

Blank is a decision. Duplicated is an accident.

If you cannot write a distinct description, leaving the tag off is defensible. Copying one sentence across two hundred pages is the failure this check and the duplication check are both pointed at, and it is strictly worse than omitting it.

Why losing one snippet is not Critical

Because the failure degrades gracefully. When the tag is missing, something reasonable still appears in the result, assembled from your own page. Nothing disappears from the index, no signals split, no crawler gets confused. You have lost authorship of one sentence.

Compare that with a missing canonical, where the consequence is structural and silent, or an accidental noindex, where the page leaves the index entirely. Important means it costs you real clicks. Critical is reserved for defects that cost you the page.

Telling whether yours is being used

Search for a distinctive phrase from the page, then compare the snippet on the results page against what you wrote. If they differ, your text was not judged the best answer for that query, which is information rather than a bug: it tells you which query the page is actually being found for.

A re-crawl confirms the tag is present, unique and in range. It cannot confirm the snippet, because the snippet is chosen per query at serve time and no crawler sees that decision.

Does this affect AI search?

Almost not at all, and this is the check where saying so plainly matters most, because meta tags are where people reach first when they want to be "optimised for AI".

An assistant fetching your page reads the body. It has no snippet to choose and no click to win, so a sentence written to earn a click in a ranked list has no job to do. Some retrieval systems use the description as a short document summary when ranking candidates, which is a marginal effect nobody outside those systems can measure.

The useful takeaway is the inverse one. If you are writing meta descriptions to improve AI visibility, you are working on the wrong check: what determines whether a model can use your page is whether it can reach and parse the content itself.

Questions this check raises

Do meta descriptions affect rankings?: No, and Google has said so directly since 2009. They affect click-through rate, which is a different mechanism: the description is the sales copy under your title, and a result nobody clicks earns nothing regardless of where it ranks. Treat it as advertising, not as optimisation.
Why is Google showing different text than my meta description?: Google replaces the description whenever it thinks a passage from the page answers the query better, which is the majority of impressions for most sites. That is usually a good outcome, since a query-matched snippet converts better than a generic one. It becomes a problem only when the text it picks is boilerplate or navigation, which is a content structure problem rather than a meta tag one.
Is it better to leave a meta description blank than write a bad one?: Yes, on pages where you have nothing specific to say. A blank tag lets the engine assemble a snippet from the query and the body, which beats a generic sentence repeated across 400 product pages. Write descriptions where the pitch is genuinely different per page, and skip them where it is not.