Quotable claims and evidence blocks

A quotable claim is a sentence an AI assistant (or a hurried human) can lift out of your page and repeat without inventing context, inflating the meaning, or mistaking a joke for a fact. An evidence block is the adjacent number, named source, method note, or link that makes that claim defensible to repeat. Together they are the writing pattern behind two AEO judgments Silktide already scores: Claims backed by evidence and Tone cannot be misread as fact.

This technique sits on top of Answer-first page structure. Answer-first decides where the point goes. Quotable claims decide how each factual sentence is built so extraction does not corrupt it.

Throughout this page, suppose Fernwood publishes a guide on reimbursement speed and a comparison page against Ledgerly.

Why it works

Assistants do not "read the page and form an opinion." They retrieve passages, score them for usefulness, and quote or paraphrase the ones that look like answers. Research and practitioner reporting on AI citations consistently find the same pattern Silktide encodes in its AEO checks: passages with specific figures, named sources, and self-contained wording get reused; vague marketing assertions and ironic asides either get skipped or - worse - get repeated as if they were literal.

Three failure modes show up constantly:

1. Unsupported assertion. "Fernwood is the fastest expense tool" - nothing to verify, high liability if repeated.
2. Context-dependent fragment. "That drops to two days after approval" - "that" means nothing in isolation.
3. Tone trap. "Ledgerly reimburses sometime in the next geologic era" - a human hears sarcasm; a model may emit it as a factual claim about Ledgerly.

Your job is to make the sentences you want repeated impossible to misunderstand, and the sentences you do not want repeated impossible to mistake for facts.

Anatomy of a quotable claim

Every load-bearing factual sentence should survive this test: paste it into a blank document with no surrounding paragraph. Is it still true, specific, and attributable?

1. Name the subject; kill the pronoun

Restate the noun. Include the unit. Include the scope (US, direct deposit, after approval).

2. Put the number in the sentence, not only in a chart

Charts and hero animations often never reach AI crawlers - see Content readable without JavaScript. Even when they do, models prefer a sentence they can quote. Duplicate the key figure in plain text:

Median approval-to-payout time: 1.2 business days (Fernwood customer benchmark, Q1 2026, n = 412).

3. Attach evidence in the same block

An evidence block is not a bibliography at the footer of the site. It is the immediate support a skeptical reader needs:

• Primary data you own - link to the research page that has method and tables (Original research and statistics).
• Third-party source - name it and link it ("Bureau of Labor Statistics…", "Gartner…", the statute, the vendor docs).
• Product fact - point at the settings screen, docs page, or policy that defines the behavior ("documented in Fernwood Docs → Payments").

Pattern:

Fernwood reimburses approved US direct-deposit expenses in the next
business-day payment run (Fernwood Docs, Payments overview, updated
2026-06-12).

Or, for a researched claim:

Among 412 mid-workspace finance teams surveyed in Q1 2026, median
approval-to-payout time on Fernwood was 1.2 business days
([Fernwood Reimbursement Benchmark 2026](https://fernwood.example/research/reimbursement-times-2026)).

One well-sourced claim beats a paragraph of adjectives. Silktide's AEO guidance is blunt about the same tradeoff on Claims backed by evidence.

4. Prefer structures machines extract cleanly

When the claim is enumerable, use a list or table - the same rule as answer-first extractable structures:

Each row is already a quotable mini-claim.

Evidence quality bar

Treat evidence the way you would want a journalist to treat it - because assistants increasingly behave like careless interns with perfect memory:

• Specific over vague. "Industry-leading" is not evidence. "1.2 business days median" is.
• Named over anonymous. "Experts say" is worthless. "Maria Chen, Fernwood's head of accounting products" or "BLS CES 2025" can be checked.
• Dated. Numbers without a when go stale silently. Pair claims with Machine-readable dates on the page.
• Honest about method. If n = 412 self-selected customers, say so on the research page. Hiding the method is how citations turn into reputational debt.
• Match the visible page. Do not put a statistic only in JSON-LD or only in an image alt. Structured data may repeat a fact; it must not invent one.

Keeping tone from becoming fake fact

Personality is allowed. Literal-looking falsehoods are not.

Rewrite rules

Practical tests before publish:

1. Literal reading test. If a model quoted this sentence into a buyer's brief, would you still stand behind it?
2. Opposite-tone test. If someone removed all typographic cues (no italics, no emoji, no surrounding jokes), would the sentence still be true?
3. Competitor-harm test. Would this sentence be defamatory or misleading if repeated as fact about a named competitor? If yes, rewrite or cut - comparison pages have stricter rules under Comparison pages.

Irony that inverts meaning ("Sure, losing three days to expense busywork is everyone's favorite hobby") is the highest-risk form, because the literal parse is the opposite of the intended one. Do not put inverted jokes in the same paragraph as real product facts.

Worked example: Fernwood reimbursement guide

Before (common marketing draft):

We've always believed getting paid back should feel magical. That's why Fernwood is the fastest, smartest expense platform on the workspace - finance teams tell us they'll never go back. Receipts? Handled. Approvals? Instant. Your CFO will wonder how you lived without it.

After (quotable + evidenced):

Fernwood reimburses approved US direct-deposit expenses in the next business-day payment run, with funds typically arriving in 1–2 business days (Fernwood Docs → Payments, updated June 12, 2026).

In Fernwood's Q1 2026 customer benchmark (412 mid-workspace finance teams), median time from manager approval to payout was 1.2 business days. Full method and tables: Reimbursement Benchmark 2026.

What can delay a reimbursement?

• Missing receipt image after a card spend
• Approver out of office without a delegate
• International wire payout instead of US direct deposit

Every sentence in the revision can stand alone. The humor is gone from the factual spine; if Fernwood wants voice, it can live in a clearly labeled opinion section below the facts - not inside the claims assistants will grab first.

How Silktide helps

• Claims backed by evidence - flags bare figures and superiority claims without support.
• Tone cannot be misread as fact - flags humor and exaggeration that could be quoted literally.
• Key points quotable standalone / Content in a quotable shape / Specific rather than promotional - neighboring judgments this technique also improves.
• Offers original information - when the evidence is data only you have.

Related

• Answer-first page structure - where claims sit on the page
• Internal linking for topical authority - link claims to the proof pages that back them
• Original research and statistics - producing primary evidence worth citing
• Comparison pages - high-stakes claims about named competitors
• Credible proof on conversion pages - placing attributable proof next to the conversion ask
• Author pages and bylines - who stands behind advisory claims
• Machine-readable dates - dating the page the claims live on
• Creating helpful content (Google Search Central)