AI Citability Audits
A page can hold position one and offer an answer engine nothing it can lift. Ivan Sp audits pages for citability, meaning whether an answer engine can lift three to five sentences and present them as a complete answer. Citability is a separate measurement from ranking, and a page can rank well in Google while failing it entirely.
What Makes a Sentence Quotable
A qualifying sentence meets three conditions at once. The sentence names the business or entity explicitly rather than using a pronoun. The sentence states a specific claim rather than a general benefit. The sentence resolves every reference inside itself, so it still makes sense with no surrounding paragraph attached.
“We pride ourselves on fast response times” fails all three. “Metrics Rule audits sites between 50 and 2,000 pages” passes all three. The difference is not writing quality, it is self-containment.
The Three-Sentence Threshold
Pages carrying fewer than three qualifying sentences get flagged as weak on citability. Three is the working threshold because an answer engine assembling a response needs enough material to choose from. A page with one usable sentence competes badly against a page with eight.
The audit counts qualifying sentences per page and reports the number alongside the total sentence count. A page with 60 sentences and two qualifying ones is a common finding, and it explains why an otherwise strong page never gets cited.
The Definitional Sentence Check
One check runs on every page. Does the content carry a clean definitional sentence in the form “X is a Y that Z” for the core entity?
Answer engines reach for that pattern constantly, and most business websites never write one. Marketing copy tends to open with a benefit claim or a question instead of a definition. A page selling storage units that never states what a storage unit is has no sentence to offer when a model needs one.
What the Audit Delivers
The audit returns a per-page report covering four items:
- The count of qualifying sentences, with each one quoted
- Whether a definitional sentence exists for the core entity
- The specific sentences that fail, with the reason each one fails
- Rewritten versions of the highest-value failures
Rewrites stay inside the facts the page already states. Where a sentence is vague because the underlying fact is missing, the audit flags the gap and requests the detail rather than inventing a number.
Where Citability Fits
Citability work sits downstream of structure and upstream of authority. A page has to be crawlable and organized before its sentences matter, and it has to be trusted before a model prefers it over an alternative. The full sequence is covered on the AI search optimization service page.
What the Audit Measures
- The unit
- A passage of roughly three to five sentences, rather than the page as a whole.
- The test
- Whether the passage means the same thing when read with nothing around it.
- Subject naming
- Whether the entity is named explicitly rather than carried by a pronoun.
- Claim completeness
- Whether the core claim finishes inside one sentence.
- The definitional sentence
- A single sentence stating what the thing is, early in the section.
- Antecedent distance
- How far back the referent of each pronoun sits.
- Heading dependency
- Whether the passage still makes sense without its heading attached.
- What the audit outputs
- A pass, borderline, or fail grade per passage, with the failing sentence named.
Where the Standard Content Advice Breaks Down
Could you lift three sentences from your strongest page and have them make sense?
- The usual advice
- Write comprehensively and in depth, since thorough content is what gets cited.
- What holds up in practice
- Depth and citability are independent. A four-thousand-word article can contain no passage that stands alone, while a nine-hundred-word page with clean definitional sentences is quoted repeatedly.
Consider a guide running to four thousand words on a technical subject. The research is genuine and the page ranks well.
Every claim in it is built across paragraphs, with each sentence depending on the one before. No three-sentence window states anything complete.
Ivan Sp audits at passage level rather than page level. Length was never the variable, and self-containment always was.
A comprehensive page that cannot be excerpted earns its ranking and none of the citations. The research was real and the format wasted it.
Every answer engine describing your topic is choosing sources right now. A page nothing can quote is not in the running, whatever it ranks.
Four Things Commonly Assumed About Citability
Which of these describes your strongest content?
I meet these four assumptions on most content audits.
| Commonly assumed | What actually holds |
|---|---|
| Longer content is more citable. | Length and self-containment are independent. A long page built on chained sentences offers nothing extractable. |
| Ranking well implies citability. | Ranking measures relevance across a whole document. Citation depends on whether a short passage stands alone. |
| Writing for extraction reads badly. | Naming the subject and completing a claim in one sentence is clearer prose. Both audiences benefit from the same edit. |
| Bullet lists are the most extractable format. | A bullet fragment usually depends on its stem. A complete sentence extracts more reliably than a list item. |
| Adding an FAQ section makes a page citable. | An FAQ helps where the answers are complete sentences. Answers that reference the question without restating it fail the same test. |
The passage an engine could quote from your page either exists or it does not. Writing one is an editing job rather than a research job.
A Worked Example: Three Passages Graded
Consider three passages from one page, graded against the audit. The assessment below is illustrative, written to show what each grade means. No client engagement is described here.
| Passage opens with | Grade | Why |
|---|---|---|
| It depends on how the site is built | Fail | No subject named, and the claim is incomplete |
| This approach works better in most cases | Fail | Demonstrative with no antecedent in the passage |
| Log files record every request a crawler made | Pass | Subject named, claim complete in one sentence |
| The tool exports the data as a CSV | Borderline | Subject named vaguely, and the tool is not identified |
| Screaming Frog exports crawl data as a CSV file | Pass | Named entity, complete claim, no dependency |
I grade every passage rather than sampling. The two failing rows are single-word fixes, and nobody finds them without reading for this specifically.
A citability audit is reading your own page through a straw. Only three sentences are visible at a time, and most writing collapses under that.
The borderline row is the common case. Naming the tool turns a vague sentence into a quotable one, and it costs two words.
How a Citability Audit Runs
I read each passage in isolation rather than reading the page in order.
- Split the page into passages of roughly three to five sentences each.
- Read each passage with everything around it hidden, including its own heading.
- Mark every passage where the subject is carried by a pronoun rather than named.
- Mark every claim that finishes across two or more sentences rather than inside one.
- Check whether a definitional sentence appears early in each section.
- Grade each passage as pass, borderline, or fail, and name the specific failing sentence.
- Rewrite failing passages by naming the subject and completing the claim in one sentence.
- Replace pronouns whose antecedent sits outside the passage.
- Re-read each rewritten passage in isolation to confirm the grade changed.
- Record the pass rate per page, so the next audit compares against a number.
How This Engagement Is Scoped
I scope citability audits by page count, since the work is per passage rather than per template.
- What arrives
- A passage-level grade sheet, the failing sentence named per passage, and rewritten replacements.
- Typical duration
- One to two weeks for a set of twenty pages.
- What the client supplies
- The page list, and subject review of any rewritten claim.
- Included
- Passage splitting, isolation reading, grading, rewriting, and a recorded pass rate.
- Not included
- Any promise that a rewritten passage will be cited.
- What changes the estimate
- Very long pages, or highly technical content needing subject review.
- How it is quoted
- After a short scoping call, once the page list is agreed.
Tools and Primary Sources
The references below cover extraction behaviour and structure. Each opens in a new tab.
- Google AI features documentation describes how AI surfaces draw on and attribute web content.
- Creating helpful, people-first content is the underlying quality standard a citable passage still has to meet.
- Google featured snippet documentation describes passage selection, which follows similar self-containment logic.
- Schema.org vocabulary supplies the entity names that make a passage unambiguous to a machine.
Grade One Passage Yourself
I suggest a test that takes about two minutes per passage.
- Open your most important page and pick a paragraph from the middle.
- Copy three consecutive sentences into a blank document.
- Read them with no heading and no surrounding context.
- Ask whether a stranger could state the claim from those three sentences alone.
Most pages fail on the first passage tried. The fix is usually naming a subject that the surrounding paragraphs were carrying for it.
Common Questions About AI Citability Audits
Can a page rank in Google and still fail a citability audit?
Yes, and that combination is common. Google ranks whole documents on relevance and authority signals, while answer engines extract individual passages. A page can satisfy the first system and offer the second nothing usable.
How many pages does a citability audit cover?
Scope depends on the site. Audits typically start with the homepage, the primary service pages, and the pages already earning impressions in Google Search Console, because those carry the most upside.
Does fixing citability hurt readability for people?
No. The changes remove pronoun ambiguity, replace vague claims with specific ones, and put answers before context. Those edits make pages easier for people to read as well.
Does writing for extraction make content worse for humans?
No. The changes are naming the subject explicitly and completing a claim inside one sentence. Both are ordinary clarity edits, and readers benefit from them as much as machines do.
Discuss This Work
The first call is a free 30-minute consultation with no pitch attached. You receive an honest assessment of where the site stands and what can realistically be improved.