A search engine ranks your page. A model lifts a passage out of it. That one difference decides how a page should be built, and it is why a page that ranks well can still never be quoted: the information is all there, just never in one liftable piece.
The skeleton
Every page on this blog is built the same way, and the order is the point:
- The question as the title. Retrieval matches question to question. A title that is the phrase someone types beats a clever one.
- The answer in the first paragraph. Forty to sixty words, complete on its own, no setup. If someone read only this paragraph they should have the answer.
- Question-shaped headings. Each section is one sub-question, answered immediately under the heading rather than at the end of it.
- One table. Tables survive extraction intact; the same comparison in prose does not.
- Specifics in the body. Numbers, dates, prices, named limits — one per section if you can.
- Three real questions at the end. Not keyword padding: the questions people actually ask next.
- Schema markup. Tells the crawler what the page is instead of making it infer.
The paragraph that does most of the work
If you change one thing, change the opening. Compare:
Before: "In today's fast-moving digital landscape, visibility has never been more important. Businesses across every sector are discovering that the rules have changed…"
After: "GEO is the practice of making a page likely to be quoted inside an AI answer rather than ranked in a list. About three quarters of it is ordinary SEO; the quarter that differs is that the unit of competition shrinks from the page to the paragraph."
The first says nothing that can be lifted. The second is the whole article in sixty words, and it is the part that gets quoted.
What to strip out
- The warm-up. Two paragraphs of context before the point. Delete and start at the point.
- Adjectives standing in for facts. "Affordable", "powerful", "leading". Replace each with the number it is hiding.
- Answers split across sections. If understanding a point needs three scrolls, no passage contains it.
- Text locked inside images. A price in a graphic is a price nothing can read.
- Infinite scroll and tabbed content. If a crawler must click to see it, treat it as absent.
The markup worth adding
| Type | Where | What it buys |
|---|---|---|
| Organization | Once, site-wide | Who you are, stated rather than guessed |
| Article with dates | Every post | Freshness, which decides time-sensitive questions |
| FAQPage | Pages with real questions | Question-to-answer pairs a retriever can take whole |
| Product with price | Anything you sell | A quotable number instead of "contact us" |
| Breadcrumb | Every post | The shape of the site, so the cluster reads as a cluster |
Structure does not rescue an empty page
All of this makes a page that has something to say easier to quote. None of it gives a page something to say. A perfectly structured page with no number, no limit and no first-hand experience in it will be passed over for a messy page written by someone who actually did the thing — because the messy page contains a fact, and structure only decides how easily a fact is found.
So the order is: have something specific to say, then make it liftable. Doing it the other way round produces the kind of page nobody — human or machine — finishes reading.
