Citability — writing to be quoted

ExpertDuration ~20 min video + 45 min hands-onTools A page of your own content to rewrite, ChatGPT / Perplexity to test-quote it, The downloadable citability_scorer.py

Why this lesson

An AI will only quote text it can lift cleanly and trust. Most web writing fails that test: the real answer is buried three paragraphs down, the section leans on context from earlier on the page, and there’s not a single concrete number to grab. So the model finds someone else’s page that is quotable and cites them instead. Citability is the craft of writing passages an answer engine can extract, understand in isolation, and repeat with confidence. This lesson gives you a five-part score to diagnose any passage, the one writing pattern that fixes most of the gap, and the evidence for which specific moves actually raise your odds. It’s the most controllable lever in all of AEO — you don’t need permission or budget, just a rewrite.

Watch

Watch for: The four structural principles — BLUF (bottom-line-up-front), atomic (self-contained) content, entity-rich writing, and simple declarative sentences. Note the data: content length barely correlates with citations (0.04), and 53% of cited pages are under 1,000 words. It's about answering the question, not word count.

Watch for: The BID formula for vetting keywords (Business potential, Intent, Difficulty) plus the AI filter — can AI fully satisfy this query without a click? — and the shift from keyword research to prompt research, where the same question asked ten ways yields ten different answers. This tells you which questions to write answer-targets for.

The explainer

The five citability dimensions

Score each content block on five dimensions; a page’s citability is the average of its best five blocks. On a 0–100 scale, ≥70 is “citation-ready” and ≤30 is “citation-unlikely.”

1. Answer-block quality (25%) — does the block answer a real question in 1–3 sentences, cleanly enough to quote verbatim?
2. Self-containment (20%) — does it make sense on its own, with no dependency on surrounding paragraphs?
3. Structural readability (20%) — lists, tables, bolded key terms, scannable formatting the model can parse.
4. Statistical density (20%) — specific numbers, dates and percentages, not vague claims.
5. Uniqueness (15%) — original data or insight the model can’t get from ten other pages. This is the same idea as information gain: content that adds something new.

The answer-target pattern

Most citability problems dissolve with one move: the answer-target pattern. Write a heading shaped like the question people actually ask (“How much does local SEO cost?”), then follow it immediately with a self-contained answer of about 40–60 words that states the conclusion first. This is BLUF — bottom-line-up-front, borrowed from military writing. Don’t open with “Over the past few years, pricing has evolved due to…”; open with “Local SEO typically costs $500–$2,000 per month in 2026, averaging around $1,557.” Humans scan pages in an F-pattern — beginning closely, middle lightly — and large language models weigh the start and end of a passage more heavily than the middle. Bury the answer and both readers miss it.

An answer block must stand alone

AI systems chunk your content into pieces before processing, and different models cut at different places — you can’t control where. So every section needs self-containment: pull one H2 out and read it with zero surrounding context. Does it still make sense, or does it lean on “as we discussed above”? If it can’t stand alone, it can’t be quoted alone. A concrete test: copy a single section into a fresh doc and ask a colleague (or an AI) whether it’s fully understandable. Rewrite until it is.

Load it with statistics and entities

Statistical density is the highest-value ingredient the evidence supports. The Princeton GEO study found that swapping vague language for specific statistics raised visibility inside AI answers by roughly 37–41% — the single biggest lever it tested. Adding cited sources and expert quotes added another 30–40%. So replace “this tool is fast” with “this tool returns results in under 400ms.” The same goes for entities: instead of “the tool helps with research,” write “Ahrefs Keywords Explorer surfaces low-difficulty, high-traffic keywords.” Named entities and hard numbers give the model concrete, quotable material — and signal that you have first-hand data rather than generic filler.

Write for humans, aim at the right questions

There is no secret AI format — models are trained on what humans find useful, so writing that genuinely serves a reader is what gets cited. The strategic half is choosing which questions to answer. Vet keywords with BID (Business potential, Intent, Difficulty) and add the AI filter: if an AI Overview already answers the query completely, ranking for a click is a trap — target that query for a mention instead. Then do prompt research: real people don’t type keywords into ChatGPT, they hold conversations (“I run a two-person agency, which CRM should I pick?”), and the same intent phrased ten ways yields ten different answers. You’re not optimizing one prompt — you’re building answer-target passages across a whole topic so you surface however the question gets asked.

Do it

1. Take one real page you own. Pick its five most important content blocks.
2. Run the downloadable citability_scorer.py on the page (or score each block by hand on the five dimensions). Record the starting citability score.
3. Choose your weakest block and rewrite it into the answer-target pattern: a question heading + a self-contained 40–60-word BLUF answer.
4. Inject at least two specific statistics or dates, and name at least two entities (real products, tools, or sources) with a citation for any stat you didn’t generate.
5. Apply the self-containment test — read the rewritten block cold, out of context. Fix anything that only makes sense with the rest of the page.
6. Re-score. Then paste the question into ChatGPT or Perplexity and see whether your new passage gets quoted. Log the before/after in the Level 3 workbook.

Level 3 workbook — citability scorecard & answer-target rewrite templatelevel-3-workbook.pdf113 KBOriginal course material — free to use

Terms introduced

Terms introduced in this lesson

• Citability
• Answer block
• Self-containment
• Statistical density
• Answer-target pattern
• Information gain
• Prompt research

Each term links to the glossary, where it is defined by the way you just used it.

Check

Check yourself

1. What is the "answer-target" pattern for a citable passage?

   A paragraph that saves the answer for the very endA 3,000-word essay that covers everythingA question-shaped heading followed immediately by a self-contained 40–60-word answer that states the point up front (bottom-line-up-front)A keyword-stuffed paragraph

2. The Princeton GEO study found the single biggest lever for AI visibility was…

   Changing the fontSwapping vague language for specific statistics — a ~37–41% lift — with cited quotes and sources close behindMaking the article much longerAdding more keywords

3. Why does "self-containment" matter for getting cited?

   It makes the page load fasterIt has no real effect; it just reads nicelyIt improves your backlink profileAI chunks your page into pieces and you can’t control where the cuts fall — a section that only makes sense with three paragraphs of prior context can’t be quoted on its own

Check answers

You can move on when you can… score a passage on the five citability dimensions, rewrite a weak block into the answer-target pattern with real statistics and self-containment, and explain why stats-and-quotes (not word count) are the evidence-backed levers for getting quoted.

Mark this lesson complete

Go deeper

• Princeton / Georgia Tech / Allen Institute — “GEO” paper (KDD 2024): the source for the +statistics and +quotes findings you just applied.
• Ahrefs — “How to create content that gets cited by AI”: the written companion, with the BLUF / atomic / entity-rich principles and the cited-page data.
• Next: 3.3 · AI crawler access — none of this matters if the crawlers can’t reach your page in the first place.