When Google writes the summary of your page, what is the page for?

On Thursday, Search Engine Roundtable reported a small change that's easy to miss. In some AI Overviews, the source cards (the boxes to the right of or below the answer that link to the pages Google drew on) now show a description that Google's AI wrote. Brodie Clark, who spotted it, put it plainly: "The text doesn't appear exactly like this on the page – it is being summarised based on the information that Google thinks is most relevant to the query." Barry Schwartz said he saw it for some of his test queries, not all.

It may be a test that goes nowhere. Google tests a lot. But it's worth thinking through, because it finishes a shift that has been underway for years: the search engine no longer just points at your page. It describes it, in its own words, to suit each question.

So here's the question this piece tries to answer. If Google writes the summary of your page, what is your page for?

How we got here

The blue-link era had a simple deal. You wrote a title and a meta description, Google usually showed them, and the searcher decided whether to click. The words under your link were mostly yours.

That deal has been loosening for a long time. Google's own snippet documentation says snippets are "primarily created from the page content itself," and that Google uses the meta description only sometimes, when it gives a more accurate description than the content does. In other words, the words under your link have been chosen by Google, per query, for years. What's new in this test is that they're no longer chosen. They're written.

AI Overviews pushed things further. The answer itself is Google's writing, assembled from several sources, and the links sit beside it. Pew Research Center's study of browsing data from March 2025 shows what that does to behavior.

With an AI summaryWithout an AI summary

Show the data

Action	With an AI summary	Without an AI summary
Clicked a regular search result	8%	15%
Ended their browsing session	26%	16%

Source: Pew Research Center, browsing data from 900 US adults, March 2025 (published July 22, 2025)

Pew also found that people clicked a link inside the AI summary itself on just 1% of visits where one appeared. The source cards are the part of an AI Overview that sends traffic, and even they don't send much. Now Google is testing writing those too.

Who the summary is for

The obvious reading is that this is bad for publishers and brands: another layer of Google's words between the searcher and yours. There's something to that. If the source card already tells you what the page says about your question, there's even less reason to click.

But there's another reading, and for CPG brands it may be the more useful one. A source card with a summary tuned to the query is a better answer to the shopper's question than a snippet cut from a page written for someone else. If someone asks whether a sunscreen is reef-safe and the card says, in a sentence, what the brand's page says about its ingredients, that's a decent outcome for the brand, provided the summary is accurate.

That's the heart of it. For a lot of CPG searches, the brand doesn't need the click as much as it needs the description to be right. A shopper who learns from a source card that your bar has 12 grams of protein and no added sugar, and then buys it at Target, is a win you'll never see in analytics. We made a similar argument about why AI traffic is the wrong measure for most brands.

The risk is the other case: the summary gets it wrong, or blends your page with someone else's claim, or picks the one outdated sentence on a page you forgot to update. Then Google has written something about your product that you didn't say, under your link.

What the page is for now

If Google does the describing, a page has three jobs, and only one of them is the old one.

Supply facts that survive summarizing. A summarizer keeps what's specific and checkable and drops what's vague. "Our most delicious granola yet" disappears. "45 g serving, 6 g sugar, made in a facility that handles tree nuts" survives. Put the facts a shopper would ask about in plain sentences near the top of the page, with numbers and units. Tip 6 covers answer-first writing, and it matters more when someone else is condensing you.

Leave no room for a wrong summary. If one paragraph on your page says a product is vegan and an old FAQ further down mentions honey, a summarizer can pick either. Pages that contradict themselves get summarized badly. Audit for stale claims, old formulas and retired sizes. Tip 34 has the checklist.

Earn the click for the people who want more. Some shoppers will click: the ones comparing, the ones who want detail, the ones ready to buy direct. For them, the page still needs to be the best place to compare options or see the whole story. Tables help both sides. SISTRIX's study this week found that pages with a table with a header row were cited about 1.5 times as often as pages on the same site without one. Tip 22 is about comparison tables.

Notice what isn't on the list: writing the perfect meta description. It's still worth doing, because Google sometimes uses it. It's just no longer the main way you control how you're described. The body is.

Should you limit snippets?

Google offers real controls. The documentation describes a nosnippet rule that stops a page's text from being shown as a snippet, a max-snippet setting that caps the length, and a data-nosnippet attribute that hides specific parts of a page. Google's page on AI features points to the same controls for limiting what's shown from your pages in Search, though neither page addresses this particular test, and we don't know whether a written summary counts as a snippet for these purposes.

Control	What Google says it does	Trade-off for a CPG brand
nosnippet	Don't show a text snippet for this page	Strongest control; the result shows less, which can make it less useful to the searcher
max-snippet:[number]	Limit snippet length to a number of characters	Keeps descriptions short, but also cuts the facts you want shown
data-nosnippet	Exclude a specific section of the page from snippets	Useful for text you don't want quoted, such as old promo terms or legal boilerplate

For most CPG brands, our view is that heavy restriction is the wrong instinct. Your product facts are the thing you want repeated: ingredients, sizes, certifications, how to use it. Hiding them from the system that describes you to shoppers trades accuracy for control, and the trade usually goes badly. The targeted control, data-nosnippet on sections that would mislead out of context, is the one most brands might actually use.

Publishers whose content is the product face a different calculation, and the case for limiting snippets is stronger for them. A brand selling a physical product isn't in that position.

The part we can't answer yet

Two questions are open, and anyone who tells you the answers this week is guessing.

First, how often is the summary wrong? An AI summary of a page is another piece of AI writing, and AI writing gets things wrong. We don't have data on how accurate these card summaries are, and Google hasn't published any. The only way to know for your own pages is to check the queries that matter to you, by hand or with a tool that keeps a record of what AI search shows, such as SearchPanel from XENA Intelligence, which asks assistants shopper-style questions and keeps the answers.

Second, does the summary change who gets clicked? A query-tuned description could send more clicks to the pages that best answer a question and fewer to the ones with the cleverest copy. That would reward exactly the kind of page we've argued for in our reference book piece. Or it could cut clicks across the board. Pew's data predates this test, and nobody has measured it yet.

What we'd tell a brand on Monday

Don't panic about a test that showed up for some queries. Do treat it as a preview. Google already writes the answer in AI Overviews, as our look at commercial searches found, and it's testing writing the descriptions of the sources too. In that world, your page is less an advertisement and more a citation source: a place a machine goes to find out what's true about your product.

The brands that do well there won't be the ones with the best taglines. They'll be the ones whose pages are easiest to summarize correctly.