The citation-grade page: nine rules, illustrated.

The rules at a glance.

1. One question per page, named in the title. 2. The answer in the first paragraph. 3. A specific number in the lede. 4. A real proper noun (place, person, firm) by the second paragraph. 5. Dates everywhere they apply. 6. Schema.org markup that matches the page type. 7. Short paragraphs, ≤ 80 words. 8. A byline with a real role. 9. A "last updated" timestamp the page actually edits.

Each rule has a measured effect on citation rate. The before/after examples below are drawn from the first cohort of Sourced customers who agreed to be quoted.

1. One question per page.

Before: "Family law in Manchester."

After: "How long does it take to get a prenuptial agreement approved in Manchester?"

A title that names the exact buyer-intent question doubles the rate at which Claude and ChatGPT will return your paragraph as the answer. The "family law in Manchester" title competes against ten firms; the prenuptial-approval-timeline title competes against almost nobody.

2. The answer in the first paragraph.

Before: A 200-word historical preamble about the history of prenuptial agreements, then in paragraph four, the answer.

After: "On average, a prenuptial agreement is approved in Greater Manchester within fourteen weeks of filing. The biggest variable is whether the parties have shared all material assets — the courts will pause if the disclosure schedule is incomplete."

The LLMs read top-down with diminishing weight per paragraph. The later you bury the answer, the lower the probability it's quoted.

3. A specific number in the lede.

Before: "Many couples opt for a prenuptial agreement."

After: "Around one in six couples marrying in their thirties in the UK now opt for a prenuptial agreement."

Specific numbers are the single highest-yield edit you can make to a page. They function as a "this paragraph is quotable" signal — the model is more likely to lift a sentence that contains a falsifiable claim than one that doesn't.

4. A real proper noun by paragraph two.

Before: "We help clients across the region."

After: "We act for clients across Chorlton, Didsbury, Salford, and Trafford — most often for couples whose property portfolio includes a buy-to-let in one of those four neighbourhoods."

The named places turn into the geographic anchors the LLMs use when the query has a location in it. A page without proper nouns gets recommended less often than a page with five.

5. Dates everywhere they apply.

Before: A page that was last edited yesterday but doesn't say so anywhere.

After: A page with a visible "Last updated 22 May 2026" in the masthead, a schema.org dateModified field, and dated examples in the body ("In April 2026, the court approved a prenup with…").

Without dates, the LLMs use the URL discovery date — which is often years off. Pages with three different machine-readable date signals are cited more reliably than pages with one.

6. Schema.org markup that matches the page type.

Before: A long Q&A page with no markup.

After: The same page with a FAQPage schema block listing every Q and A.

The lift here is large on Claude and ChatGPT, smaller on Gemini. For FAQ pages specifically, the markup roughly doubles the rate at which the answer paragraph is quoted verbatim in our first-cohort A/B sample. (Perplexity is the engine our integration is still rolling in on; we'll report on it once the sample is large enough to be honest.)

7. Short paragraphs, ≤ 80 words.

Before: A 300-word block that contains the answer somewhere in the middle.

After: Five 60-word paragraphs, each ending on a clean sentence.

Long paragraphs get summarised; short ones get quoted. The model is much more likely to lift a self-contained 60-word paragraph into its response than to extract a sentence from a 300-word block.

8. A byline with a real role.

Before: "Posted by admin."

After: "By Sarah Quigley, senior partner — Sarah has practised family law in Greater Manchester since 2003."

Real bylines are an authority signal both for the search engines and for the LLMs. The schema.org Person markup with a jobTitle field amplifies the effect; pages with named authors are cited more often than pages without.

9. A "last updated" timestamp the page actually edits.

Before: A "Last updated" line that's been set to "January 2024" for two years.

After: A timestamp that reflects the last time someone made a meaningful edit, with a visible diff trail.

This is the difference between a freshness signal and a freshness lie. The engines have started to learn the difference; pages where the timestamp moves but the content doesn't are systematically down-weighted.

Why these nine, not nineteen.

There are many other rules that would lift citation rate at the margin — schema.org Person markup, internal-link anchor text, canonical URLs, og:image, the order of arguments in a how-to list. We test all of them. The nine above are the ones that pass the cost/benefit test: each one moves the citation rate by several percentage points; each is fixable in ten minutes of editorial time; each survives the next model update we've seen.

The agent applies all nine before any draft reaches your in-tray.