9 Ways to Write for Featured Snippets and AI Answers

Featured snippets used to be their own game, worth chasing for the box at the top of the results page. Now the same passage that wins that box is often the exact text that ChatGPT, Perplexity, and Google's AI Overviews pull into their own answers. Learning to write for featured snippets isn't a side skill anymore. It's close to the baseline for showing up anywhere an engine is summarizing the web instead of just linking to it.

Why the same page can win snippets and AI answers

The same page wins both because both systems are solving an identical problem: find a short piece of text that answers the query without extra work. Google's own guidance says featured snippets appear when its systems can lift a passage that directly answers what someone typed, and that the same max-snippet setting a site uses to control snippet length now governs AI Overviews and AI Mode too. That's a strong signal that these aren't separate targets requiring separate writing.

Google reported in May 2024 that its AI Overviews accuracy was on par with featured snippets in testing, which suggests the systems are drawing on similar quality signals rather than running on entirely different logic. By May 2025, Google had expanded AI Overviews to more than 200 countries and territories, so this isn't a narrow feature to plan around. It's most of the search results page.

The market has shifted in a way worth naming plainly. Ahrefs found that AI Overviews now appear on over a quarter of keywords in a recent sample, while classic featured snippets dropped from 15.41% of SERPs in January to 5.53% in June. Snippets are shrinking as a category. AI answers are growing into the space they used to occupy. Writing for one without the other is a wasted effort, since the underlying discipline, clear claim, clean structure, is identical either way.

There's no setting that forces a page into either surface. Google is explicit that you can't mark a page to become a featured snippet; its systems decide whether a page is a good match, based on the content itself. You can block snippets outright with nosnippet, or narrow how much text is eligible with max-snippet, but there's no equivalent switch to turn them on. The only lever that works is writing the passage well enough that it earns the pick.

What makes a passage easy to lift

A passage is easy to lift when it makes sense with nothing above or below it. Google's help documentation says featured snippets work best when it can extract a short, self-contained piece of text that directly answers the query, and the same logic governs whether an AI system quotes you cleanly or paraphrases around you because your sentence didn't stand on its own.

Pronouns are the usual failure point. Say you've written a paragraph that argues retention matters more than acquisition. If the second sentence starts with "this is why it works better," an engine that lifts just that sentence has no idea what "this" or "it" refers to. Compare it with a version that names the subject again: "Retention works better than acquisition because a small gain compounds across every future cohort, not just the one you just acquired." That costs a few extra words. It buys a sentence that survives being quoted with nothing around it.

This means writing slightly more repetitively than feels natural. Restating the subject instead of leaning on "it," "this," or "that" reads a little flat if someone works through the whole page top to bottom. But that flatness is what makes each paragraph independently quotable, and for a page built to be extracted, that trade is worth making.

Specificity does the rest of the work. "Studies show it helps" gives an engine nothing worth quoting. A sentence naming an actual figure, a named method, or a direct comparison gives it something concrete to lift. Google's own documentation on generating snippets notes that a meta description can sometimes be used instead of on-page text, when it happens to answer the query more precisely, but that's a fallback, not a strategy. The page's actual content is still the primary source, so the fix is always the sentence itself, not the metadata around it.

How should you write the opening

The opening should give the direct answer in the first sentence, before any setup. If a heading asks how long a meta description should run, the next line should say "150 to 160 characters," not a paragraph explaining why meta descriptions matter first. Both snippet systems and AI answer engines grab the sentence that most directly answers the query sitting in front of them, and context is welcome, just not in front of the answer.

Take the same fact written two ways. A slower version: "Meta descriptions matter because they're often the first thing a searcher reads about your page, and getting them right can influence whether someone clicks through. Most experts recommend keeping them fairly short." A direct version: "A meta description should run 150 to 160 characters. Google truncates anything longer, and a truncated description usually reads as unfinished, which costs clicks rather than earning them." Nothing about the underlying fact changed between the two. Only the order did, and the second version gives a machine something to lift in sentence one while still giving a human the reasoning in sentence two.

Most writing instinct runs the other way: warm the reader up, establish why the question matters, then land the answer near the end of the paragraph. That instinct is what costs the box. A competing page that states the number first and explains it second gets lifted over a page that explains for a paragraph and then states the number last. Fixing this means changing how a draft gets built, not just editing the final paragraph. Write the answer first, every time, and treat the explanation as the second move rather than the setup.

Which format matches the query

The format that gets lifted depends on what kind of question was asked. A "what is" question wants a tight paragraph, ideally 40 to 55 words, with one clean idea and no sentence that opens with "this" or "as mentioned above," phrases that fall apart once pulled out of context. A "how to" question wants numbered steps written as commands: "Open Settings," not "you will want to consider opening the settings menu." A "best" or comparison question wants a table or a bulleted list, because prose forces the engine to reconstruct a structure it could have just been handed.

Google's technical writing guidance backs this up directly, recommending lists and tables specifically because they make information easier to scan than the equivalent content buried in paragraphs. That's especially true for definition, process, and comparison queries, which is most of what shows up in the "People Also Ask" boxes sitting next to featured snippets.

Getting the shape wrong is often the entire reason a well-written, accurate answer never gets pulled. A process explained in flowing prose might be perfectly clear to a human reader and still lose out to a worse-written competitor that happened to number its steps. The engine isn't grading quality in isolation. It's picking the passage that requires the least reconstruction, and prose describing a comparison table's worth of information is exactly the kind of thing that gets skipped for a page that built the table.

How do you make the page worth the click

You make the page worth the click by putting something in the body that the snippet or AI summary couldn't say on its own, since a clean liftable answer at the top only helps if there's a reason to keep reading past it. Winning the box doesn't guarantee losing the visit, but it does raise the bar for what the rest of the page has to deliver.

The stakes here are real. Search Engine Land reported on a 2025 Pew study showing people clicked a traditional search result in just 8% of searches where an AI summary appeared, compared with 15% when no summary was shown. Both figures come from the same sample, so they're directly comparable. That's a real drop in click-through, and it means the visits that do still happen need to justify themselves fast, with something the summary skipped.

That something is usually a worked example, a tradeoff the summary had no room for, or the number behind a claim the summary only gestured at. Google's structured-data documentation includes case studies showing what happens when a page is built for both extraction and depth: Rotten Tomatoes saw a 25% higher click-through rate on pages using richer markup, and Food Network reported a 35% increase in visits after enabling search features across most of its site. Markup didn't cause those numbers by itself. Clear content and structured labeling were both present, and neither one was covering for the other.

Schema is worth naming precisely here because of what it does and doesn't do. It doesn't write the answer. It labels an answer that's already been written clearly, so a crawler doesn't have to guess which paragraph is the definition and which is the process. Canonical tags solve an adjacent problem: if the same answer lives on two URLs, something has to decide which one gets cited, and a page that leaves that ambiguous hands the decision to chance. Freshness matters too, in a narrower way, particularly for any page carrying a number, since a stale figure is the kind of thing that quietly gets swapped out for a competitor's more recent version. None of these three replace clear writing. They just make sure clear writing gets found and trusted once it exists. Google's move to expand Preferred Sources in AI Search, with people reportedly twice as likely to click through to one, points the same direction: recognizable, well-labeled, trustworthy content keeps winning, in whatever box the engine decides to build around it.