A meta description is a candidate, not a caption

Most writers treat the meta description as the caption under their search result: write it once, and that is what people see. Google's own documentation says otherwise. The snippet is built mainly from the page, and your description is one candidate Google may pick when it fits the search better than the page text does. That changes what a good one is for. Its job is not to restate your opening but to answer the likeliest search more plainly than any paragraph on the page can.

Google writes the snippet, and it can write a different one per search

Google's snippet documentation, last updated on 20 April 2026 when we read it, is direct about who holds the pen. Snippets "are primarily created from the page content itself", and Google uses the meta description when it might give users a more accurate description of the page than content taken from it. The same page says snippets are chosen to match the specific search, so one post can show different snippets for different queries.

Read plainly, there is no single caption to control. A post that ranks for six questions can be summarised six ways, and each time your description is competing with your own paragraphs. It wins when it describes the page better than any excerpt would. It loses, quietly, when it does not, and nothing in your CMS tells you which happened.

We cannot tell you how often Google picks the description over page text. Confident percentages circulate, none of them from Google, and the documentation gives no figure. What it gives is the rule, and the rule is enough to write by.

The opening paragraph is the worst candidate on the page

Among Google's examples is a bad description that "doesn't summarize the page": a nostalgic line about gathering eggs as a child, on a page that is really a guide to cooking them. That is what an opening paragraph usually looks like once it is pasted into the description field. Openings are written to earn the next paragraph. They set up a tension, introduce a scene, hold back the payoff. All of that works on the page and fails in a results list, where the reader has one line to decide whether this page answers their question.

There is a structural reason too. A description that repeats the opening offers Google nothing it could not already excerpt from the page. The only way a description beats page text is by saying something the page never says in one place: the whole argument, compressed. Our style guide has always forbidden copying the first paragraph into the description, and as of this morning our publishing lint refuses it as well. A copy is a candidate that cannot win.

Write for the search, not for the post

The useful question is not how to summarise the post but what the person who should click it actually typed. Answer that in plain words, name the specific thing the page gives them, and stop. Google asks for descriptions that are unique to each page and accurately describe it. It also says there is no length limit, and that the snippet is truncated to fit the device. So the 160-character ceiling we enforce is our convention, not Google's. We keep it because the cut happens anyway, and a description that lands its point before the truncation is one that survives on a phone.

This is the same argument we have made about titles. Most people who see your result never click, so the line they see has to carry the claim rather than tease it. The description is the second line of that promise.

An engine fails on sameness, not length

Google's documentation explicitly allows programmatically generated descriptions on large sites, and asks that they be "human-readable and diverse". That second word is where machine-written descriptions break. It is the same distinction behind our argument that Google doesn't penalize AI content: the problem is never that a machine wrote the text, only that it reads like every other page.

We checked our own before writing this. All 52 of our descriptions sit inside the length rule. But our recent ones have converged on a single shape. Six of the last thirteen open with "Why", "How" or "When" and go on to list what the post covers, clause after clause. That is a table of contents, not an answer, and across an archive it drifts toward the "identical or similar" descriptions Google calls unhelpful. Our lint measured what was easy to measure. It never asked whether the descriptions sounded alike, and neither did we.

Submit the line you would want chosen

If Google is choosing between your description and your paragraphs, the job is to make the description the best candidate on the page: the plainest answer to the likeliest search, specific to that page, different from its neighbours. Accept that you will not always win, and that you will rarely know when you lost. The fix for a weak description is not a tighter character count. It is reading the last ten side by side and asking whether a stranger could tell them apart.