Passage Visibility Index

Passage Visibility Index (PVI) is not a Google metric. It is a custom score SEOs build to estimate how likely a 40-250 word passage is to surface for a query, even when the full page is not the strongest result. Useful in practice. Easy to misuse.

What PVI actually measures

PVI is usually a 0-1 or 0-100 score assigned to a paragraph, list, or short section. The model tries to predict whether that block has the structure, relevance, and context to benefit from Google’s passage-level understanding. Google introduced passage ranking in 2020, and Google’s John Mueller has repeatedly clarified that Google does not have a separate “passage index” you can optimize against directly. So treat PVI as an internal prioritization layer, not a ranking factor.

The best use case is pages already ranking in positions 8-30 in Google Search Console. Those pages often have enough authority to compete, but weak answer blocks. Tightening one passage can move a URL from 0 clicks on long-tail variants to meaningful incremental traffic.

How teams build it

Most teams extract paragraph-level HTML with Screaming Frog custom extraction, Python, or BeautifulSoup, then map each block to its parent H2 or H3. Features usually include passage length, query-term overlap, semantic similarity to top-ranking snippets, heading alignment, internal link context, and entity coverage compared with competing pages from Ahrefs or Semrush exports.

For modeling, simple beats fancy more often than people admit. Logistic regression is usually enough if you have a clean labeled set from GSC plus SERP snapshots. XGBoost can help on larger sites with 10,000+ passages, but only if your labels are reliable. That is the weak point. Passage-level labels are noisy because GSC reports at page-query level, not paragraph-query level.

A practical benchmark: if your model cannot beat random by a wide margin and hold an AUC above roughly 0.75 in back-testing, it is probably not production-ready.

What to optimize when PVI is low

• Answer shape: Put the direct answer in the first sentence. Then support it with specifics in 40-80 words.
• Heading match: Rewrite H2s and H3s to reflect the actual query framing shown in GSC and Semrush.
• Context signals: Add nearby entities, examples, and internal links so the passage is not semantically isolated.
• Formatting: Lists and short explanatory blocks often outperform bloated paragraphs for comparison and how-to intent.

Surfer SEO and Clearscope-style content scoring can help with entity gaps, but they are not passage models. Different job.

Where PVI breaks down

The caveat is simple: Google ranks pages, not detached paragraphs floating in a vacuum. A strong passage on a DR 18 page with 12 referring domains will still lose to a weaker passage on a DR 70 page with 5,000 referring domains for competitive terms. PVI is most useful on sites that already have baseline authority and stable indexing.

It also gets messy on JavaScript-heavy pages, templated content, and pages with poor heading hierarchy. If Screaming Frog cannot extract clean sections, your score will be garbage. Start there. Not with machine learning theater.