Why it matters

When a claim, its evidence, and its source are scattered across a long paragraph, an AI system has to reassemble them — and often extracts the assertion without the support, or attributes it to the wrong source. Formatting important claims as explicit Claim / Evidence / Source blocks keeps the three together, so the model can lift a fully-provenanced statement in one piece.

The discipline also forces clarity on you: one claim per block means you can't smuggle three half-supported assertions into a single sentence, and it makes the difference between established fact, sourced interpretation, original inference, and speculation explicit.

What good looks like

Important claims are structured, atomic, and provenanced:

Each significant claim is a three-line block: Claim → Evidence → Source.
One claim per block — no mixing multiple claims in a single paragraph.
The evidence is specific (data or observation), and the source is a named organisation + year.
The block distinguishes established fact, sourced interpretation, original inference, and speculation.
Blocks are used for the page's load-bearing claims, not every sentence.

Step by step

Isolate one claim Take a single load-bearing assertion. If a paragraph contains several, split them — each gets its own block. One claim per block is the rule.
State the claim plainly Write the claim as a short, direct statement — the conclusion you want a reader (or model) to take away. "GEO improves citation visibility."
Add the specific evidence Give the concrete data or observation that supports it — a number, a measured effect, an experiment result. "Adding quotations, statistics or citations raised a source’s visibility in AI answers by up to ~40%."
Name the source Attribute the evidence to a named organisation and year, hyperlinked where possible (Phase 4.4). "Aggarwal et al., GEO (KDD 2024)."
Label the claim's epistemic status Where it isn't obvious, mark whether the block is established fact, sourced interpretation, original inference, or speculation — so the model (and reader) knows how much weight it carries (links to Phase 4.6).

Worked examples

1 Before / after: paragraph vs. C/E/S block

Before A long paragraph mixing the claim, some narrative, a half-remembered statistic, and an unnamed source — the assertion and its provenance drift apart.

After

Claim:    Adding evidence to a page makes it more
          visible in generative answers.
Evidence: Quotations, statistics and citations
          raised visibility by up to ~40%.
Source:   Aggarwal et al., GEO (KDD 2024).

2 Labelling epistemic status

The same structure, with the claim's status made explicit so its weight is unambiguous.

Claim:    Answer engines will keep favouring pages
          that carry their own evidence.
Evidence: The GEO study found evidence-adding
          rewrites raised visibility on two engines.
Source:   Aggarwal et al., GEO (KDD 2024).
Status:   Sourced interpretation - one study,
          not yet an established trend.

Common mistakes

Mistake	Fix
Multiple claims in one block	One claim per block; split the rest.
Evidence that just restates the claim	Use concrete data or an observation, not a paraphrase.
Unnamed source ("a study")	Name the organisation and year (Phase 4.4).
Speculation formatted like established fact	Label the epistemic status explicitly.
Forcing every sentence into a block	Reserve blocks for load-bearing claims.

Limits of this advice

The block format improves clarity and provenance; no study measures its effect on citation directly.
Use it for load-bearing claims only; a page of nothing but blocks is hard to read.

Checklist

Each load-bearing claim is a Claim / Evidence / Source block
One claim per block — no mixing
Evidence is specific data or observation
Source names an organisation and year
Epistemic status labelled where it isn't obvious

How to Format Claims as Claim / Evidence / Source