Information Per Word: The GEO Metric Nobody's Measuring

Most pages don’t get skipped by AI search because they’re wrong. They get skipped because they’re padded. AI answer engines pull the passage that delivers the most verifiable fact in the fewest words. That’s a ratio worth naming and measuring on its own: Information Per Word (IPW).

Why IPW matters for AI search

Traditional SEO rewards pages that rank, meaning pages that satisfy a search algorithm’s sense of relevance, authority, and keyword match. AI answer engines do something different: they extract and cite a specific passage, not a whole page. That passage has to stand on its own, because it’s often lifted out of context and dropped into a generated answer.

This changes what “good copy” means. A paragraph that reads smoothly for a human, with transitions, throat-clearing, and brand-voice flourishes, can still fail an AI system if the actual extractable fact is buried in filler. A denser, blunter competitor paragraph wins the citation instead, even if it “reads worse” to a person skimming the page.

IPW is a way to name that gap: how much verifiable, citable information a passage delivers, divided by how many words it takes to say it.

How I estimate it

There’s no official tool that spits out an IPW score. This isn’t a metric with an industry-standard calculator. But you don’t need one to use it. Here’s the gut-check I run on a page during a review:

• Count the discrete facts. Read a paragraph and count how many standalone, verifiable claims it contains: numbers, named specifics, concrete outcomes.
• Count the words it took to say them. Total word count for that same paragraph.
• Ask the extraction question. Could you pull 3 bullet-point facts out of this paragraph without rewriting a single word? If yes, IPW is high. If you’d have to paraphrase and infer to get anything usable, IPW is low.

You’re not computing a precise ratio. You’re training your eye to spot the difference between a paragraph that contains information and one that’s actually built to deliver it.

Before and after

Low IPW:

“Our team is committed to providing best-in-class solutions that are tailored to meet the unique needs of every customer we work with, ensuring a seamless experience from start to finish.”

Zero extractable facts. No numbers, no named specifics, nothing an AI system could cite with confidence. It’s 29 words that say “we try hard.”

High IPW:

“We complete custom orders in 3–5 business days and back every part with a 2-year warranty.”

18 words, two independently verifiable claims. An AI system can lift this sentence whole and use it as a direct answer to “how long does a custom order take” or “what’s the warranty.”

Same topic, same intent to reassure a buyer, completely different citability.

Where IPW fits with the other five levers

IPW isn’t a sixth lever. It’s a working test for Factual Density, one of the five levers in the GEO framework (alongside Citability, Conversational Alignment, Authority Signals, and Structured Clarity). When I’m auditing a page against all five levers and want a fast way to spot Factual Density problems specifically, IPW is the shortcut: read a paragraph, ask what’s extractable, count the words it took to get there.

FAQ

Does higher IPW always mean better copy? No. Some context and framing is still necessary. A page that’s nothing but bare facts with zero connective tissue reads like a spec sheet, not a page a human wants to engage with. The goal is cutting filler, not cutting all narrative.

How is this different from keyword density? Keyword density measures how often a term appears. IPW measures how much verifiable information a passage delivers regardless of which words are used. A page can have perfect keyword density and still say almost nothing.

Can I calculate an exact IPW score? Not with any standardized tool. This is a practical rule of thumb, not an industry metric with a formula behind it. Treat it as a diagnostic lens, not a number to optimize toward a decimal point.

Where should I apply this first? Start with your highest-traffic pages: product pages, service pages, anything already getting AI Overview or answer-engine impressions. Those are the pages where a low-IPW paragraph is most likely costing you a citation right now.

Want a second pair of eyes on this?

If you’re not sure whether your own pages would pass the extraction test, that’s exactly what I look at in a page review. See how SF5 page reviews work.

Written by David Cox, GEO consultant at Small Factory 5. IPW is the same Factual Density check applied in every page review, made explicit enough to run yourself.