Fluent but Empty: Why Polished Copy Gets Skipped by AI
Fluency alone produced far smaller GEO gains than statistics, quotations and citations. Why polished generalities give engines nothing, and how to run the extraction test.
Polished, fluent copy gets skipped by AI engines all the time, because fluency isn’t what makes a passage citable — substance is. The GEO research found that improving fluency alone produced far smaller visibility gains than adding statistics, quotations and citations. An engine composing an answer needs something concrete to ground it in; beautifully-written generalities give it nothing. Plenty of well-crafted marketing content fails this test entirely, reading impressively while saying nothing an engine could quote.
Key takeaway
- Fluency alone produced much smaller GEO gains than adding statistics, quotations and citations.
- Engines need something concrete to ground an answer in — polished generalities provide nothing quotable.
- Test your passages by asking what specific fact an engine could lift; if nothing, add substance rather than polish.
Why fluency isn’t enough
During generation, an engine assembles an answer from retrieved passages and prefers material it can confidently stand behind — a specific figure, a named expert’s statement, a concrete claim with a source. Elegant prose that asserts something general offers no such anchor. The passage may be pleasant to read and perfectly on-topic, but there’s nothing in it the engine can point to as the substance of an answer. This is why the research found fluency optimisation delivering modest returns compared with the substance-adding methods.
Substance > style
In the KDD 2024 GEO research, fluency improvements produced far smaller visibility gains than adding statistics, quotations and citations — the methods that give an engine something concrete to ground its answer in.
Source — Aggarwal et al., KDD 2024
What empty-but-polished looks like
You know it when you apply the test. “Businesses today face an increasingly complex digital landscape where standing out requires a thoughtful, strategic approach.” Grammatically flawless, comfortably on-topic, and containing precisely zero quotable content. Compare: “Only around 17% of AI Overview citations came from organic top-ten results in early 2026, down from roughly 76% in mid-2024.” One of these an engine can use. The difference isn’t writing quality — it’s whether a specific, checkable claim is present.
The extraction test
Read each important passage and ask: if an engine wanted to answer a question using this, what exactly would it quote? If you can point to a specific figure, a named person’s statement, a concrete finding or a definite claim, the passage has substance. If your honest answer is “the general idea,” it doesn’t. Apply this to your key sections rather than every sentence — introductions and transitions can be general, but the passages carrying your main points need something an engine could lift verbatim.
Fluency still matters — just not alone
None of this argues for writing badly. Clear, well-structured prose helps readers understand you, helps engines parse your meaning, and makes the substance you’ve included accessible. The research found fluency improvements did help, just far less than substance additions. The failure mode isn’t good writing; it’s good writing used as a substitute for having something specific to say. The strongest content is both — fluent prose that carries real, sourced, concrete claims. Polish the sentences after you’ve made sure they contain something worth polishing.
Frequently asked questions
Why doesn’t well-written content get cited by AI?
Because engines need something concrete to ground an answer in, and polished generalities provide nothing quotable. During generation, an engine reaches for specific figures, named statements and sourced claims it can confidently stand behind. Elegant prose asserting something general offers no such anchor, which is why the GEO research found fluency alone delivering far smaller gains than adding substance.
Does writing quality matter for AI visibility at all?
Yes, but less than substance. Clear, well-structured prose helps engines parse your meaning and makes your substance accessible, and the research found fluency improvements did help — just considerably less than statistics, quotations and citations. The failure mode isn’t good writing; it’s good writing used as a substitute for having something specific to say.
How do I tell if a passage is citable?
Apply the extraction test: if an engine wanted to answer a question using this passage, what exactly would it quote? If you can point to a specific figure, a named person’s statement, a concrete finding or a definite claim, the passage has substance. If the honest answer is “the general idea,” it doesn’t — add something specific.
What should I add to make polished content citable?
The three proven levers: relevant sourced statistics, quotations from named credible experts, and citations to authoritative sources — placed on the passages carrying your main claims. Keep the prose clear and well-structured, but ensure each key section contains something specific and checkable an engine could lift verbatim rather than only a well-expressed general point.
The bottom line
Fluent copy that says nothing specific gives AI engines nothing to work with, which is why polish alone produced modest gains in the research while statistics, quotations and citations moved the needle. Run the extraction test on your key passages: name what an engine could quote. If you can’t, the fix is substance, not another editing pass.
We write content that’s both readable and genuinely quotable, with substance on every key claim. Part of our AI Visibility service.
