Spoke
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.
Follows directly from how passage extraction works and is long-established practice for featured snippets. No platform publishes a specification for it.
Almost every page on the internet buries its answer. Not out of carelessness — out of training. We are taught to establish context before making a claim, and that instinct produces an opening paragraph that explains why the question matters and a third paragraph that finally answers it.
How do you write an answer-first opening sentence?
State the conclusion in the first sentence, including its own subject, in a form that remains true and comprehensible with no heading above it and no paragraph beneath it. Then use the rest of the section for reasoning, caveats and detail.
The practical rule is that the opening sentence should be removable. Cut it out, show it to someone cold, and if they understand what question it answers, it works.
The pronoun trap
This is the most common failure and the easiest to fix.
Heading: Does Google-Extended affect my rankings? Opening: No, it doesn't.
Extracted alone, "No, it doesn't" is useless. The answer depends entirely on a heading that may not travel with it.
Opening: Google-Extended does not affect Google Search rankings. Google states it controls training and grounding in other systems, and that it is not used as a ranking signal in Search.
That survives extraction because the subject is inside the sentence.
Before and after
| Buried version | Answer-first version |
|---|---|
| Many firms wonder about review generation, and the rules here are complex and vary by jurisdiction, so it's worth understanding the landscape before deciding on an approach. | Ask every client for a review, not only the happy ones. Filtering requests by expected sentiment is review gating, and the FTC's rule on consumer reviews reaches practices of that kind. |
| Schema markup has become an increasingly important topic in recent years as search engines have grown more sophisticated in how they interpret page content. | Schema markup states what a page is in a form a machine can read. It does not rank the page and does not make a claim true — it removes guesswork a system would otherwise have to do. |
| There are several factors to consider when evaluating whether your site's loading performance is adequate for modern search requirements. | Your site's loading performance is adequate when Largest Contentful Paint is good at the 75th percentile of real visits. Lab tests do not settle it; field data does. |
Each left-hand version is competent prose. Each one is also unusable as an answer, because none contains a claim.
"It depends" is allowed
The instruction to lead with a verdict is not an instruction to be falsely decisive. Plenty of real answers are conditional.
The way to do it is to put the condition in the first sentence rather than deferring it:
Whether to block AI training crawlers depends on whether your content is itself the product. Publishers who sell access have a reason to block; businesses who want to be found and named almost never do.
That is extractable, honest, and immediately useful. Compare it with "there are arguments on both sides", which is extractable and worthless.
Where this site does it
Every question heading on this site is an answer block, and the first sentence under each one is written to this rule. The on-site search ranks those blocks and shows the best match in full — so the search feature is a working test of whether the writing rule was followed. A sentence that reads oddly when the search returns it alone is a sentence that failed.
More in AEO
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.
Snippets & voice
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.
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.