Measures how "sticky" each sentence is — how densely packed with function words (prepositions, articles, auxiliaries) versus content words (nouns, verbs, adjectives). High glue = the reader is doing a lot of work to extract meaning from mostly-structural words.
Endpoint: /glueindex — no word cap; runs on the full manuscript.
What it measures
For every sentence in your manuscript the endpoint computes a glue percentage — the ratio of function words (prepositions like "of / in / by / on", articles like "the / a", auxiliaries like "is / was / have", pronouns, conjunctions) to total words. Content words (nouns, main verbs, adjectives, adverbs of manner) are the load-bearing parts of a sentence; function words are the connective tissue.
A sentence like "It was of the sort of a thing that one might have been said to have been considering" is nearly all glue — the reader has to trudge through ten function words before extracting any meaning. A sentence like "The dog barked" is almost all content — the reader gets the picture immediately. Most healthy prose sits at roughly 45–55% glue; academic and legal writing runs 55–65%; deliberately spare fiction can drop below 40%.
Each sentence gets categorised into one of three buckets by its glue percentage:
- Sticky — glue percentage is high (typically above ~60%). The sentence is doing a lot of structural work the reader has to unpick before the meaning emerges.
- Semi-sticky — glue is elevated but not extreme (typically 55–60%). Individually fine; in aggregate a sign of drift toward flabby prose.
- (uncategorised) — normal glue level. Not flagged.
The endpoint also returns an overall glue index for the whole manuscript, with a target range (usually ≤35–40% depending on genre calibration) and a plain-English verdict. The overall number is a rough average; the per-sentence list is where the actionable content lives.
This endpoint has no word cap — it runs on the full manuscript.
Why it's useful
Sticky sentences are the ones that read as clumsy without you being able to say exactly why. You know something is off — the sentence feels like it's taking too long to land — but the individual words are all correct English, so it's hard to spot the pattern on re-read. Glue index gives you the pattern: too many function words per content word means the sentence is structural, not communicative.
Almost all sticky sentences rewrite trivially into cleaner prose. "It was of the sort of a thing that one might have been said to have been considering" collapses to "One might have considered it" or "It was worth considering" — same meaning, a fraction of the words, all glue removed. Doing a single revision pass through the top twenty sticky sentences flagged by this report typically tightens the prose more than a whole-manuscript re-read.
The report is especially useful for writers whose voice drifts toward academic register — anyone with a PhD, a legal background, a corporate day-job, or years of writing in a formal register brings glue-heavy patterns into fiction without noticing. This report surfaces the pattern so you can strip it.
Use alongside Sentence analysis (which flags hard sentences by multiple dimensions, not just structural stickiness) and Passive voice (many sticky sentences are also passive; the two lists partially overlap and reading both catches the full pattern).
How to read it
The glue output appears as the Sticky sentences collapsible issue list near the bottom of the panel, marked with an amber dot. The row header shows the total count of flagged sentences (sticky + semi-sticky combined). Click to expand.
Inside, every flagged sentence is a row rendered as a small card:
- An amber dot at the top-left (the layer's swatch color)
- The sentence text on the right — line-clamped to 2 lines
- A meta line below in mono/tabular type:
NN% glue · <category>— the exact glue percentage rounded to an integer, and the category label (stickyorsemi-sticky)
The list is not sorted by percentage in the current version; sentences appear in their order in the manuscript. Sticky sentences (the more severe category) are visually indistinguishable from semi-sticky in the row layout except for the meta line — the amber dot is the same for both.
Click any row to lock it as focused. When the list is open, the semi-difficult highlight layer paints in the editor across every flagged sentence, so you can see the full pattern in place. Clicking a specific row additionally puts a focused-style outline on that exact span, and the editor scrolls to it. Click the same row again to deselect. Collapse the list to clear the paint.
The overall glue index (whole-manuscript percentage) does not render as its own header on the sticky-sentences card — it feeds into the Glue index metric row inside the Style scores radar chart higher in the panel. Hover the "Glue index" axis on the radar to see the current percentage and the ideal range.
When to ignore it
Formal prose genuinely needs more function words. Legal writing, academic writing, technical documentation, and certain kinds of essay all rely on connective tissue that fiction can do without — "with respect to", "in accordance with", "for the purposes of" are load-bearing in a contract, wasteful in a novel. If you're writing in a formal register, the report will flag every third sentence and every flag will be technically correct but practically wrong. Ignore.
Second: certain literary effects use glue deliberately. Henry James, Marcel Proust, and the tradition of the long-clause introspective sentence build meaning through connective tissue — the sentence is the network of qualifications. A Jamesian sentence will always score sticky and always be right to. If you're chasing that register, the report is descriptive, not prescriptive.
Third: dialogue in certain characters is supposed to be glue-heavy — an evasive character who never gets to the point, a bureaucrat, a lawyer, a professor. Character voice can require exactly the pattern the report flags as a defect. Filter accordingly.
Fourth: sentences with a lot of prepositional phrases stacking spatial or temporal information ("on the third shelf of the second cabinet by the door at the back of the study on the ground floor of the townhouse in Bloomsbury") will read as sticky. If you meant to give that much spatial detail, the sentence is doing its job even if the glue percentage is high.
Finally, glue is a surface-feature metric that cannot see meaning. A perfectly rewritten sentence in the same low-glue form as a botched original will read completely differently to a human reader. Use the flag as a pointer to look, not as an instruction to cut.
How to run it
- Click Write in the left sidebar and open the document you want to analyze.
- Open the right rail: click the Reports button (bar-chart icon) in the editor's top toolbar.
- In the rail header, click the Rust tab.
- All 8 Rust endpoints run in parallel on tab open — typically completes in 2–5 seconds. There is no card to click; everything renders in one composite panel.
- The glue output appears as the Sticky sentences collapsible issue list near the bottom of the panel. Each row shows the sentence text plus a meta line with the glue percentage and its category (sticky / semi-sticky).
- Click a row to jump to the sentence in the editor. The semi-difficult highlight layer paints while the list is open so every flagged sentence is visible at once; the clicked row gets an extra focused outline.
What to read next
- Sentence analysis — for hard sentences overall (not just structurally sticky)
- Reports — back to the panel overview