The 40–60 word answer block: anatomy of a paragraph AI engines actually quote

1. Why the block matters more than the page

A ChatGPT answer is typically 150–400 words long. A Perplexity answer sits around 200. A Google AI Overview runs 60–160. None of these read like a web page — they read like a distilled answer, with 4–8 citations linked underneath. When the AI engine decides which brands to quote, it is making the decision at the passage level, not the page level.

What the engine is actually doing, simplified: it takes the buyer's question, retrieves the top ~20 most relevant passages from its training and live-web sources, re-ranks them, pulls 4–6 passages into the context window of the generator model, and writes a fluent answer that stitches those passages together. Your job is to make sure one of those 4–6 passages comes from your site.

SEO optimised the page. GEO optimises the quotable passage inside the page.

This reframes everything about how content is written. A 2,500-word SEO article with a strong H1 may still rank — but if it contains zero discrete, extractable answer blocks, it will be summarised at best and skipped at worst. Every page now needs at least one purpose-built answer block, and ideally one per buyer question the page targets.

2. The 5-part anatomy of a cited block

Across the 1,400 cited passages we studied, the same five components appeared repeatedly. They are not all mandatory, but cited blocks nearly always include four of the five.

Component 1 — The question restated

The first sentence of the block restates the buyer's question as a declarative statement. AI engines use this as a semantic anchor. Example: "Generative Engine Optimisation is the practice of..." instead of burying the term in paragraph three. This pattern alone lifts citation probability significantly.

Component 2 — A precise, self-contained answer

The second and third sentences deliver the answer without dependencies on surrounding paragraphs. A cited block should be extractable — if you cut it out of the article and pasted it into a Slack thread, it should still make sense. No "see above" or "as mentioned below".

Component 3 — A quantified claim

Numbers are the single strongest extraction signal. A sentence like "58.5% of all searches now end without a click" is roughly 3× more likely to be cited than a qualitative equivalent. Cited blocks nearly always contain at least one specific number — a percentage, a multiple, a time frame, a revenue figure.

Component 4 — A named source or expert

AI engines weight passages that reference authoritative sources. Inline attribution like "according to Seer Interactive's 25.1M impression study" or "McKinsey projects $750B" raises both the extraction probability and the quality of the citation (the AI links the source as well as your passage).

Component 5 — A crisp conclusion

The last sentence is the takeaway. It is often the one that appears in the AI answer, because it is the line the engine uses to "close" the block semantically. Make it punchy and specific.

3. Why 40–60 words and not 80 or 100

Extraction behaviour is not uniform across engines, but the 40–60 word range is the sweet spot. Below 40 words, blocks lack the signal density engines prefer. Above 70–80 words, they exceed the context budget the engine allocates per passage and tend to get truncated — losing either the question restatement or the conclusion.

We tested deliberate 120-word blocks and saw consistent truncation on Perplexity and AI Overviews. The engine picks the middle 60 words, often the least useful ones. Tight is better.

4. Bad block vs good block, side by side

5. The ready-to-use template

Copy this template. Fill in the brackets. Publish.

Here is the same template filled in for a different category, to show the shape travels:

6. Where to put the block on your page

Placement matters. AI engines crawl the first 500 words of a page more aggressively than the rest. If your direct-answer block is in paragraph 11, the engine may never reach it before deciding what to extract. Three placement rules:

• Rule 1 — First 300 words. The primary answer block belongs in the opening of the article, typically as the second paragraph (after a brief hook).
• Rule 2 — Bold the key sentence. Visual emphasis helps screen readers, skimmers, and also appears to weight extraction. Bold the sentence containing the specific number.
• Rule 3 — Use FAQPage schema for sub-questions. Secondary answer blocks addressing related questions should live inside a FAQ block with FAQPage structured data. Each Q&A pair is itself a cited block in miniature.

7. The three mistakes we see on every client audit

Mistake 1 — Hiding the number

The strongest statistic on the page is often buried in a data table in the middle of the article. Move it up. The stat is the reason the page gets cited.

Mistake 2 — Relying on internal links to carry meaning

SEO trained writers to say "as we explored in our guide to X" with an internal link. AI engines extracting a passage do not follow the internal link. They extract the passage as-is. If the meaning depends on the linked page, the passage is un-citable. Self-contained blocks only.

Mistake 3 — No named source

A passage that says "studies show" is weak. A passage that says "the Seer Interactive 25.1M impression study shows" is strong. Every stat deserves a named source, inline, even if you also cite it formally in the bibliography. The engine reads the inline attribution.

Further reading

• GEO vs SEO: the technical overlap, the strategic divergence
• Schema for GEO: the 9 types AI engines actually read
• The 5-layer GEO programme