The meta description is the most misunderstood tag in SEO. Half the industry insists it is dead because it is not a ranking factor. The other half writes 155-character descriptions for every page as an article of faith, without ever checking whether Google used them.

Both are missing what is actually going on. The meta description does not affect whether you rank. It affects whether anyone clicks when you do — and Google, for reasons worth understanding, throws most of them away and writes its own.

This piece is about what really happens between your tag and the search result, and how to write something that survives the journey.

It is not a ranking factor, and that does not mean it does not matter

Google has been explicit about this for years: the meta description is not used as a ranking signal. Putting keywords in it does not help you rank for them. This is settled, and anyone selling you meta-description optimisation as a ranking service is selling you nothing.

But ranking is not the outcome. Traffic is the outcome, and traffic is ranking multiplied by click-through rate. A page at position three with a compelling description can easily out-earn a page at position two with a dull one, and it costs nothing to find out.

The confusion arises because SEO discourse is dominated by rankings, which are easy to measure and satisfying to watch move. Click-through rate is the other half of the equation and it is largely ignored, which means it is where the cheap wins are — the competition is not really trying.

Google rewrites most descriptions, and the number is higher than you think

Here is the fact that reframes everything: Google discards your meta description and generates its own a great deal of the time. Study estimates vary, but they consistently land somewhere in the region of two-thirds of results.

Google is entirely open about this in its documentation on snippets. The stated reason is that the snippet should be relevant to the specific query, and a single static description cannot be relevant to every query a page ranks for.

That reasoning is worth taking seriously rather than treating as an inconvenience. A page about email verification might rank for "email verifier", "how to check if an email is valid", and "catch-all domain". One description cannot serve all three, and Google would rather show a sentence lifted from your page that actually answers the query asked than show your marketing copy.

Which means the goal is not "write a description Google will use". It is: write a description Google will use for your most valuable query, and make sure the rest of the page is good enough that whatever Google extracts instead is still worth reading.

When Google keeps yours, and when it does not

The pattern, from observation rather than official statement:

  • It usually keeps yours for branded and exact-match queries. Someone searching your company name gets your description, because there is no ambiguity about intent.
  • It usually rewrites for long-tail queries. The more specific the question, the more likely Google prefers a passage from your body copy that directly answers it.
  • It always rewrites if yours is missing, duplicated, or obviously boilerplate. A description repeated across two hundred pages is a signal that nobody wrote it.
  • It rewrites if yours does not contain the query terms. Google wants to bold the words the user typed. If your description never mentions them, it will find text that does.

The practical implication is a division of labour. Write the description for the query you most want to win — usually the head term, the one with the volume. Accept that for the long tail, Google will construct something from your page, and make sure your page has clear, direct sentences that would work well as a snippet if lifted.

That second part is the one nobody does, and it is where the leverage is.

Length: the rule everyone quotes and nobody checks

The advice is 155 to 160 characters. It is roughly right and precisely wrong, because Google truncates by pixel width, not character count.

A description full of capital letters and wide characters truncates earlier than one full of narrow ones. Mobile and desktop truncate at different points. And the limit has changed several times, sometimes dramatically — there was a period in 2017 when Google displayed descriptions of 300 characters, and everyone rewrote everything, and then it changed back.

So the durable rule is not a number. It is: front-load the meaning. Put the thing that earns the click in the first hundred characters, and treat everything after that as a bonus that may or may not be displayed. If your description only makes sense when read in full, it will fail on the day Google decides to show less of it.

Our free meta description generator produces descriptions in the safe range and shows the truncation point, which is more useful than counting characters by hand.

What actually earns the click

Our blog title generator and headline generator are aimed at the title, which as noted below is doing more work than the description is.

The searcher is looking at ten results. Yours has one job: to be the one that most obviously answers the question they just typed.

What works:

  • Answer the query directly. If they asked a question, the description should indicate that the page answers it — ideally by beginning to answer it.
  • Be specific. "Learn about email verification" is worthless. "Check syntax, MX records and mailbox existence — and see why catch-all domains cannot be verified at all" tells them exactly what they will get.
  • Include the query terms naturally. Not for ranking — for the bolding. Google bolds matched terms, and bold text draws the eye in a page of grey.
  • Set an accurate expectation. A description that oversells produces a click followed immediately by a back-button, and a fast bounce back to the results is a signal you do not want to send.

What does not work: your company name at the front (they can see it in the URL), "Welcome to our website", a list of keywords, or a sentence that could describe any page on the site.

The duplicate description problem

The single most common technical failure is not a bad description. It is the same description on hundreds of pages.

This happens by default in almost every CMS: a site-wide default is set, nobody overrides it per page, and now four hundred pages share one sentence. Google will not use it — a description that describes everything describes nothing — and it will construct its own for every one of those pages.

Worse, it is a signal about the site. A page whose description was clearly not written for it is a page nobody looked at, and Google's whole business is working out which pages someone cared about.

Check for duplicates first, before you optimise anything. It is a five-minute audit in any crawler, and it usually finds more than you expect. Our SEO slug generator and internal linking suggester cover the two neighbouring problems that usually turn up in the same audit. Fixing it is more valuable than rewriting your ten best descriptions, because it addresses the pages that currently have nothing.

Pages that should not have a description at all

An underused move: sometimes the right description is none.

If a page ranks for many varied long-tail queries, and Google is going to rewrite the description for most of them anyway, a hand-written description can only serve one of those queries and will be discarded for the rest. Leaving it blank lets Google construct the most query-relevant snippet every time, which is frequently better than anything you would have written.

This is genuinely counter-intuitive and it is what Google's own documentation implies. The tag is not mandatory, and its absence is not a penalty.

Where to write one deliberately: your commercially important pages, where you know the query you want to win and you want to control the pitch. Where to consider leaving it: deep, long-tail, informational pages that pull traffic from a hundred different phrasings you could never anticipate.

How to know whether yours is working

The measurement is available and almost nobody uses it. Google Search Console shows impressions, clicks and click-through rate per page and per query.

What to look for:

  • High impressions, low CTR, good position. The classic signature of a bad snippet. You are ranking and nobody is clicking. This is the highest-value page on your site to fix, and the fix costs one sentence.
  • CTR far below the expected curve for your position. Position three typically earns a CTR in a fairly predictable band. Substantially below it means the snippet is failing.
  • CTR that changed without a position change. Usually means Google started rewriting your description, or stopped.

Change one description, wait a fortnight, look again. This is one of the very few SEO interventions with a fast, clean, attributable feedback loop, which is precisely why it is worth doing and precisely why it is strange that so few teams do it.

Writing for the snippet Google will build

Given that Google constructs the snippet most of the time, the highest-leverage work is not in the tag at all. It is in the page.

Google lifts passages that directly answer the query. So the page should contain sentences that are direct answers — self-contained, complete, and understandable without the paragraph around them.

Practically: after each heading that poses a question, answer it in the first sentence, in full, without preamble. Not "There are several factors to consider when determining whether an email is valid." That sentence answers nothing and cannot be lifted. Instead: "An email is valid if the syntax parses, the domain has an MX record, and the mail server accepts the mailbox — though a catch-all domain will accept anything, so it cannot be confirmed."

The second sentence is a snippet. It stands alone, it answers the question, and Google can lift it verbatim. Writing this way is a discipline that improves the page for human readers as well, which is the usual sign that an SEO tactic is actually a writing tactic in disguise.

The title tag is doing more work

Worth saying plainly, because attention spent on descriptions is often attention taken from something more important.

The title is a ranking factor, it is the largest and most prominent text in the result, and it is what people actually read first. A great description under a weak title is a well-decorated door nobody walks through.

Google rewrites titles too — less often than descriptions, but it happens, usually when the title is stuffed with keywords, is boilerplate, or is much too long. The same discipline applies: write for the primary query, be specific, front-load the meaning.

If you are triaging, fix titles first, then duplicate descriptions, then descriptions on high-impression low-CTR pages. In that order. The last of those is where most people start, and it is the third-best use of the time.

Structured data, and the snippet you cannot write

Beyond title and description, the result itself can be augmented — star ratings, FAQs, breadcrumbs, prices. These come from structured data, and they change the shape of your result rather than merely its wording.

A result with a review-star rating occupies more vertical space and draws the eye. An FAQ block can double the height of your listing. These are, in click-through terms, worth far more than any rewording of a description, because they change how much of the page you own.

Our FAQ schema generator produces valid markup for the question-and-answer blocks, and it is a better first move than most description work — you are competing for attention on a crowded page, and taking up more of it beats writing a better sentence in the same small box.

Automating it without producing sludge

If you have four hundred pages with no descriptions, writing them by hand is not going to happen, and a template will produce exactly the boilerplate Google ignores.

The middle path that works: generate a first draft from the page's own content — its H1, its first paragraph, its key entity — and then hand-edit the pages that matter. This gives every page something specific rather than something generic, and it concentrates human effort where the traffic is.

What to avoid is the fully automated version that produces four hundred grammatically-correct, semantically-empty sentences. That is not better than no description; it is worse, because it will be ignored by Google and it will consume the effort you might have spent on the twenty pages that actually earn.

Govarova's AI SEO Writer generates the drafts and the schema together, which is the point — the description, the title and the structured data are one problem, not three, and treating them separately is how sites end up with beautiful descriptions and no rich results.

What AI search does to all of this

The ground is moving. AI Overviews, ChatGPT, Perplexity and their successors increasingly answer the question directly, and the user never sees a list of ten blue links at all.

In that world the meta description has no role whatsoever — there is no snippet, because there is no result. What matters instead is whether your page is the source the model draws from and cites.

The uncomfortable and slightly amusing thing is that the discipline is the same. Our People Also Ask extractor is a cheap way to find the questions worth answering in the first place. A model synthesising an answer lifts clear, self-contained, directly-answering sentences from pages it judges authoritative — which is precisely what Google's snippet extraction already does, and precisely what makes a good meta description in the first place.

So the advice does not change; only the stakes do. Write pages that answer questions in complete, liftable sentences. Structure them so the answer is adjacent to the question. Be specific enough that a summary of your page is worth citing rather than being one of six sources merged into a paragraph.

The teams that spent a decade writing keyword-stuffed introductions that say nothing before paragraph four are the ones with the most to lose here. The page that gets cited is the page that answered the question, quickly, without preamble — and that page also, as it happens, wins the click when there is still a click to win.

The four mistakes that recur

  1. Writing for the algorithm rather than the person. The description is not read by a ranking system. It is read by a human deciding between ten options. Keyword density in it accomplishes nothing except making it worse to read.
  2. Describing the page rather than the value. "This page discusses email verification methods" tells the searcher what the page is about. They already knew that from the title. Tell them what they will be able to do afterwards.
  3. Writing the same description for a page and its category. Two pages competing for the same query with the same description is a signal that you have not thought about either.
  4. Never checking whether Google used it. The simplest possible verification — search for the page and look — is skipped almost universally. A great many people are carefully maintaining descriptions that Google has silently ignored for two years.

That last one is worth doing today, on your ten most important pages. It takes ten minutes and it frequently changes where people decide to spend their effort.

A worked example

Take a page about catch-all email domains.

The bad description: "Govarova offers industry-leading email verification solutions. Learn more about catch-all domains and how our AI-powered platform can help improve your email deliverability today."

It leads with the company, makes an unfalsifiable claim, uses "AI-powered" and "industry-leading", and tells the reader nothing they did not already know. Google will ignore it, and if it does not, nobody will click it.

The good description: "A catch-all domain accepts mail for every address, so verification cannot confirm whether a specific mailbox exists. Here is what that means for your bounce rate, and what to do with an address you cannot check."

It answers the question in the first sentence. It contains the query terms naturally, so Google can bold them. It sets an accurate expectation of what the page contains. And it is useful even if the reader never clicks — which sounds like a loss and is in fact the strongest possible signal that the page is worth reading.

Notice that the good version is not more persuasive in any marketing sense. It makes no claims at all. It is simply more informative, and in a list of ten results competing for one click, informative wins.

What to do on Monday morning

In order, because the order matters more than the tactics:

  1. Crawl the site and find duplicate descriptions. Fix those first. It is the largest single win available and it requires no creativity.
  2. Open Search Console. Sort by impressions, descending. Look at CTR. The pages with many impressions and poor CTR are your list.
  3. Rewrite the top ten. Answer the query, be specific, front-load, include the terms.
  4. Add FAQ schema where the page genuinely answers questions.
  5. Wait two weeks and measure. Not two days. Search Console data lags, and a fortnight is the shortest honest read.
  6. Leave the long tail alone. Let Google build those snippets. It is better at it than you are, because it can see the query and you cannot.

None of this makes you rank higher. All of it makes the rankings you already have worth more, which is a cheaper and more reliable way to grow traffic than trying to climb — and it is available today, on pages you have already written.

One last thing worth internalising

The meta description is a small tag with an outsized capacity to reveal how a team thinks. A site whose descriptions are all variations on “leading provider of innovative solutions” is telling you that nobody has sat down and asked what a person searching for this would actually want to know. A site whose descriptions answer questions is telling you the opposite.

Google is, in the end, an extremely well-funded attempt to work out which of those two kinds of site you are. Every mechanism described in this article — the rewriting, the snippet extraction, the preference for direct answers — is a machine reading your pages and deciding whether anyone cared when they wrote them.

So the durable strategy is not to write better descriptions. It is to write pages worth describing, and then to describe them accurately. That has been the right answer for twenty years, it survives every algorithm update, and it is the only one that will still be right when the ten blue links are gone entirely.

Descriptions are the easy part; the pages behind them are the work. Govarova’s AI SEO Writer drafts the page and its schema together, rich results like star ratings from Review Management lift the click, and a lead-capture chatbot catches the visitor once they arrive. Start free.