Spoke

Featured snippets and voice answers

Before AI answers there were snippets and voice assistants, which also had to pick one answer instead of listing ten. Everything this section teaches was worked out against those.

Documented by the platform

That snippets are selected programmatically with no markup to request one, and that snippet eligibility gates the AI features, are both documented by Google and linked below.

Before generated answers there were featured snippets and voice assistants — surfaces that had to pick one answer rather than list ten.

Everything this section teaches was worked out against those. That's the best evidence available that it isn't a passing fashion.

You can't ask for one. Google picks a passage it judges answers the question, and there's no markup that requests selection.

What you control is whether a clean, self-contained passage exists on your page for it to pick. In practice: a heading matching the question, a direct answer right underneath, and the supporting detail in a shape that matches the question — a list for a sequence, a table for a comparison, a short paragraph for a definition.

Why the history matters

The techniques that worked for snippets are the techniques people now sell as GEO. Question-shaped headings. Answers that stand alone. Structure matched to content.

That's reassuring in one direction and a warning in the other.

Reassuring, because practices that survived a decade and transferred to a new surface are more likely to last than tactics invented last quarter. A warning, because it means a lot of what gets marketed as novel AI-era insight is ten-year-old snippet practice with new vocabulary on it.

It also means the honest framing is modest. These are techniques for being selectable. Not for being selected.

Shape the answer to the question

The pattern holds across every surface:

  • Definition question → one-sentence definition, then the elaboration.
  • How-to question → a numbered list of steps.
  • Comparison question → a table.
  • Yes or no question → the verdict, then the condition.

Mismatching the shape is the common failure. A procedure written as prose reads fine to a person and is awkward to lift as a passage.

Voice

A voice assistant reads one answer aloud. No list to fall back on, no visual formatting carrying meaning.

That constraint is clarifying. An answer that works read aloud, with no screen, works nearly everywhere.

If your opening sentence depends on the heading above it or the bullets beneath it, it fails the read-aloud test — and that same dependency is what breaks extraction generally.

The snippet trap, again

Worth repeating, because it lives in the same settings and the cost has grown.

nosnippet, data-nosnippet and a tight max-snippet limit what Google can show from your page. Google also documents that a page has to be eligible to show with a snippet to appear in its AI features.

So suppressing your snippet now costs you two surfaces instead of one. Sites do it for defensible reasons — usually stopping a competitor lifting a definition. The trade has changed.

The caveat this section carries

Google says you don't need to write a particular way for its AI features.

Nothing here contradicts that. Snippet practice is craft that makes passages selectable, it demonstrably works for the snippet surface, and it isn't a documented requirement for AI answers. The section overview says so at the top.

More in AEO

Answer-first writing

The first sentence under a question heading is the only part guaranteed to be read, quoted or extracted. Here is how to write one that works with nothing around it, and the before-and-after that shows the difference.

Question headings

A heading tells a reader what's coming and tells a machine what the passage is about. Those used to pull against each other. Mostly they don't anymore.

Extractable formatting

Six steps in a paragraph and six steps in a list carry the same information to a reader. To a machine they're nothing alike, and that difference decides whether you get quoted precisely or approximately.

Question markup

FAQ rich results were deprecated in May 2026. The FAQPage type wasn't. Knowing which of those is true saves you from the stale advice and from the overcorrection.

Back to AEO

Audit your own site

Run these checks against your site and get a report back: what AI engines can read, what they can quote, and what is stopping them.

See what the audit checks

Want someone to do it with you?

The audit tells you what is wrong. If you want help fixing it, Maestro Strategic Partners works with law firms on exactly this.

Talk to Maestro