Google rewrites snippets. Is optimising them still worth it?

When you search on Google, the result you get consists of:

• the link,
• a title, and usually
• a short descriptive text.

Together these form the snippet – and that is exactly what can decide whether somebody clicks this result or not.

We can supply those details ourselves. That does not mean Google uses them – and in many cases that is even the better outcome, as our study shows.

Because Google often does not take the description you wrote. It builds a fitting excerpt from the page content instead. At first glance that sounds as if you could save yourself the work on the meta description.

You can't!

The more important question is a different one anyway: Is a weak click-through rate actually caused by the snippet?

A low click-through rate is not yet a reason to act

A page at position 18 with few clicks probably does not need a new snippet. It needs a better ranking position.

It gets more interesting with a page that already sits near the top and still gets noticeably fewer clicks than you would expect at that position.

Search intent belongs in that consideration too. Is somebody looking for a specific piece of information, a product or a brand? Does the page really match that? And are there other results already catching the click?

Only once those points hold does the question become worthwhile: Can a better snippet lead to more clicks?

The amount of data matters as well. With few impressions the click-through rate can swing wildly. A difference from 3.3 to 4.1 per cent means something else at 200 impressions than it does at several thousand.

So not every deviation is a problem. Sometimes it is simply statistics.

A snippet should carry the user's question

A typical company title sounds something like this:

Running Shoes Overpronation | Nordlicht Laufsport – Your Specialist Store Since 1998

The company is telling you about itself. What the user actually wanted to know was which running shoes suit overpronation.

More interesting would be:

Running Shoes for Overpronation: 6 Models Compared

And instead of a generic description:

Six models compared by weight, drop and price.
Plus a simple self-test you can do at home.

That is not some special SEO trick. The content of a snippet should simply show, as quickly as possible, what the user will find on the page.

But if Google rewrites the snippet anyway, why adjust it?

Google generates snippets dynamically and can draw on text from the page content to do so. That is why one page can end up with different descriptions depending on the query.

It does not make your own meta description useless, though.

It is not an instruction to Google. It is a proposal.

The more accurately it describes the page content and matches the search intent, the more useful that proposal becomes.

And even when Google uses a different text, the title still matters. A good page structure also makes it easier for Google to find suitable passages for the snippet.

Did the change actually achieve anything?

Checking whether a measure worked is where things get imprecise fast.

Change the meta description, wait a few days, observe a higher click-through rate – that does not prove success. In the same period the ranking, the queries, the competition and the result page itself may all have changed.

So before making the change you should at least record impressions, clicks, click-through rate and position.

After that it means waiting, gathering data and comparing.

The click-through rate is not the only figure of interest. What decides the matter is whether it genuinely improved at a comparable position and with enough data behind it.

The answer can also be:

It achieved nothing.

That is not a failure. It simply means this change had no discernible effect.

Which is annoying, and rightly so. You invested time and got no measurable result.

That is why SEO and GEO measures should always be weighed up and grounded in data. A suitable tool like Rankmio can help here – or, of course, your own experience.

What does this mean for AI search?

Here too you should not promise more than the data supports.

AI search systems work with content from the web and can take a page's title, text and other information into account. How they weigh those in detail is something no provider discloses — and the systems differ from one another and change quickly. A good snippet can therefore be part of a page that is well structured and easy to understand overall. More than that cannot honestly be claimed.

That a better meta description automatically leads to more citations by ChatGPT, Gemini or other AI systems, on the other hand, is not established.

For GEO you should therefore not try to simply write an "AI snippet".

What matters more is the content behind it: Clear statements, concrete information, traceable sources and a structure individual answers can be lifted out of without trouble.

The upshot: Not every SEO figure needs a measure

That is perhaps the most important insight.

A low click-through rate is not automatically a problem. A poor value is not automatically a reason to act. And a change is not automatically an improvement.

Good SEO therefore does not begin with the question:

"What can we optimise?"

But with:

"Is there anything to be gained here at all?"

Making exactly that distinction visible is what a good SEO analysis should do – and what good SEO consultants should take into account.

Because in the end it is not about generating as many tasks as possible.

It is about finding the right ones.

Our study: what 36 checked websites show

So much for the theory. We wanted to know how it looks in practice, so we ran 36 websites through our free snippet check. Among them retail chains, news sites, public bodies, universities, associations, a few small specialist sites and five English-language ones. Mostly homepages of well-known brands — on small websites the figures tend to be worse.

A word on the measurement first. Google does not cut off an overlong title by letters but by width. An "i" is narrow, a "W" is wide. That is why we measure in pixels rather than characters. Depending on the display, roughly 580 pixels are visible — around 55 to 60 characters, though that can vary considerably with the wording. There is no hard limit, and Google publishes none.

On 58 per cent of the pages we found a handwork error

That alone explains why Google so often writes its own text. If more than half of all stored snippets contain a formal error, your own proposal simply is not the best text available.

Almost half give space away rather than overrunning it

That was the most surprising result for us. You hear everywhere that snippet work means shortening. In fact that applies to a quarter of cases.

Sixteen of 36 titles use fewer than 400 of the 580 available pixels. The average across all pages is 464 pixels. On average, then, roughly a fifth of the display area simply stays empty.

That is not an error, and so no tool warns about it. It is unused room where a reason to click could have been.

For 42 per cent there is nothing to do

Fifteen of 36 pages had not a single handwork finding. Title within limits, description present, nothing to complain about.

That figure matches the core of this article. A tool that checks any given page and has nothing to report in four out of ten cases is working correctly — not badly.

The costliest mistake is invisible on a single page

Six of the 36 websites open every single one of their pages with the same wording. Before a word appears that distinguishes one page from the next, a large part of the width is already spent — 57 per cent on average.

Duden shows that a recurring brand reference need not be a mistake. The same idea costs 70 pixels there instead of nearly 400.

What matters is how this finding becomes visible at all: You only see it once you lay several pages of the same website side by side. Anyone checking a single address cannot find it. That is why it is so widespread and still goes almost unnoticed.

Why counting characters is not enough

The extreme case in our analysis comes from kreativland.de, our founder's own website. The title measures 1,108 pixels there against 580 visible. The reason: The name appears in it twice in a row.

Freiberufler Knut Nickol – Freiberufler Knut Nickol -
Testspezialist, Product Owner und Scrum Master, Web- und KI-Beratung

Counted in characters that is 126, roughly twice the recommendation. Counted in pixels it is nearly twice the display — and only that figure shows how much is actually lost. On the strength of this analysis the title there is now being rewritten.

A description has only two states

Six of the 36 pages have no description stored at all. Six more have one that runs longer than the display.

Between those sits a remarkably even figure: Every description that does exist averages 154 characters, against roughly 160 displayed. Anyone who writes one at all gets the length right nearly every time.

So the question about a description is rarely whether it is too long or too short. It is whether there is one.

What these figures do not say

The sample is small and one-sided. It consists mostly of homepages of well-known brands, and there somebody usually looks after title and description. It is a snapshot, not a benchmark.

And it says nothing about whether Google's own version is the better one in any individual case. That would need the click-through rates of both variants, and we do not have them. What the figures show is something else: How often your own proposal already falls short as a piece of craft, before Google even decides.

Frequently asked questions

Quick quiz: is there anything to be gained here?

Sources

• Google Search Central: How to Write Meta Descriptions
• Ahrefs: How Often Does Google Rewrite Meta Descriptions?
• Zyppy: Google Rewrites 61% of Page Title Tags
• Portent: How Often Google Rewrites Our Title Tags
• Own analysis of 36 websites, collected on 27 August 2026 with the Rankmio snippet check