# Metaphor usage analyser: how it is scored and how to change it > Metaphor usage is how often a writer reaches for figurative comparison instead of literal description, expressed as metaphors per 1,000 words. Canonical: https://scriptgrain.com/attributes/metaphor-usage # Metaphor usage analyser *By Jack Stovell ยท published 2026-09-24* Metaphor usage is how often a writer reaches for figurative comparison instead of literal description, expressed as metaphors per 1,000 words. In a ScriptGrain voice profile the extraction model (Claude Sonnet 5) estimates it from your samples rather than counting it, and it moves the voice match score at a weight of 0.75. ## Analyse your metaphor usage free Your free voice profile measures metaphor usage among all 45 attributes, from a sample of your own writing. No card needed. [Try the six-signal check first](https://scriptgrain.com/#sg-measure) ## At a glance - API field: **metaphor_usage_rate** - How it is measured: **Judged by an AI model** - Group: **Rhetoric** (ScriptGrain voice profile, 45 attributes in 8 groups) - Unit: **per 1,000 words** (ScriptGrain attribute definitions) - Voice match: **Scored, weight 0.75** (ScriptGrain voice-match engine (published method)) ## What metaphor usage measures Metaphor usage measures how densely a piece of writing leans on figurative comparison rather than plain statement. Say "the market is drowning" instead of "the market is shrinking fast" and you've spent one metaphor. Do that ten times in a thousand words and your rate is 10; never, and it's zero. Neither is wrong. A technical manual should probably sit near zero. A brand manifesto might run hot with it, and that's fine too, so long as it's consistent with the rest of the voice. Metaphor usage lives in the [Rhetoric group](https://scriptgrain.com/attributes/rhetoric) of a ScriptGrain voice profile, alongside things like [repetition as emphasis](https://scriptgrain.com/attributes/repetition-as-emphasis) and [argument structure](https://scriptgrain.com/attributes/argument-structure). It's one small piece of a much bigger picture: 45 attributes across 8 groups, all read from the same samples. On its own it won't tell you much. Paired with the rest, it starts to describe an actual voice. ## How metaphor usage is scored Metaphor usage does enter your voice match score, at a weight of 0.75. The mechanics: Claude Haiku (claude-haiku-4-5) reads your draft and judges its metaphors per 1,000 words, and the engine compares that figure with your profile's stored rate. The gap between the two gets divided by a tolerance (whichever is larger: 3 per 1,000 words, or your profile's own rate), and that fraction is subtracted from 1, floored at zero. Miss by the full tolerance or more and this attribute scores 0; miss by half of it and it scores 0.5. One honest caveat: judged values vary slightly run to run. Claude Haiku isn't running a script that counts metaphors like commas; it's making a call, the same kind of call a sharp editor would make reading your paragraph twice. So the draft's figure is an estimate, not a precise count. Full mechanics are on the [voice measurement framework](https://scriptgrain.com/reference/voice-measurement-framework) page, and the [voice match](https://scriptgrain.com/glossary/voice-match) glossary entry defines the score itself. ## How to analyse metaphor usage in your own writing You don't need software to get a rough read on this. Grab a sample, at least a few hundred words, and work through it by hand. 1. Read the piece once for meaning only. Don't hunt yet. 2. Read it again, this time underlining every figurative comparison: anything that isn't literally true but stands in for something that is. 3. Count the underlines. 4. Divide the count by the word count, then multiply by 1,000. That's your rate per 1,000 words. 5. Do this across three or four samples and average it. One blog post can mislead you; a pattern across several won't. That's the manual version. ScriptGrain's free voice profile reads metaphor usage from a writing sample automatically, among all 45 attributes. ## Examples of metaphor usage in real writing Here's what the model is looking for, in practice. Example: "The negotiation stalled because both sides dug in." (one metaphor: "dug in") Example: "Her inbox was a battlefield by nine a.m." (one metaphor: the whole comparison) Example: "Revenue dropped eleven percent in the third quarter." (zero metaphors; entirely literal) That's the whole idea in miniature: some writing describes things directly, some reaches for a picture instead, and most real writing sits somewhere between the two, sentence by sentence. ## How to make AI use more vivid comparisons If your AI drafts read flat and you want more figurative colour, be specific about where it should come from. 1. Ask for one concrete comparison, placed only where an abstract point genuinely needs a picture to land. 2. Tell it to pull metaphors from your own field, the way your samples do (for example, courtroom images for a lawyer, machinery for an engineer). 3. Paste in two or three of your own figurative lines as the reference point, not generic examples. A ScriptGrain profile handles this by sending `metaphor_usage_rate` in the voice profile JSON of the system prompt, under an instruction to treat every attribute as a hard constraint. That same profile reaches ChatGPT, Claude and other tools through the API and MCP server, so your rate travels with you rather than living in one chat window. ## How to make AI use fewer metaphors If the output feels overwritten, purple, too pleased with itself, the fix is blunter than people expect. 1. Ask explicitly for literal statements that describe the thing itself, not what it resembles. 2. Ban the specific offenders by name: "the architecture of", "a constellation of" and similar abstract metaphor constructions. Naming them gives the model something to check; a vague "be less flowery" doesn't. 3. Tell it your own writing samples are the ceiling for figurative language, not the floor: it should never sound more poetic than you actually are on the page. When ScriptGrain writes from your profile, a register anchor in the system prompt says "Do not reach for a metaphor when metaphor_usage_rate is low", and nominalised abstract metaphors such as "geography of" are banned unless they appear in your profile's own lists. The AI-isms in marketing copy reference covers more stock phrasing of this kind. ## Questions ### How do you measure metaphor use in writing? Read a sample, underline every figurative comparison (anything standing in for a literal statement), count the underlines, divide by total words, multiply by 1,000. That gives metaphors per 1,000 words. Do it across several samples and average the result; one piece rarely tells the full story. A ScriptGrain voice profile does this same read automatically, judged by a model rather than counted by a script. ### How is metaphor usage scored? It contributes to your voice match score at a weight of 0.75. Claude Haiku judges your draft's metaphor rate; the engine compares it with your profile's stored rate and converts the gap into a similarity score using a tolerance (3 per 1,000 words, or your profile's rate, whichever is bigger). A gap as large as the tolerance scores 0 for this attribute; a smaller gap loses a proportionate share. ### How do I make AI use fewer metaphors? Ask directly for literal description rather than comparison. Name the specific abstract metaphors to ban, phrases like "the architecture of" or "a constellation of", since vague instructions like "less flowery" rarely land. Tell it explicitly that your own writing samples set the ceiling: it shouldn't sound more poetic than you actually write, ever. ### How do I make ChatGPT less flowery? Give it literal-language instructions, name the exact constructions you want gone, and tell it your own samples are the ceiling for figurative language. A ScriptGrain profile does this structurally rather than as a one-off request: it stores your metaphor_usage_rate, which ScriptGrain's own drafts treat as a hard constraint and which reaches ChatGPT, Claude and other tools via the API and MCP server. ### Is there a metaphor counter? There's no code that literally counts metaphors the way a word processor counts words; it's a judgement call, made by a model reading the text for figurative comparisons versus literal ones. ScriptGrain's extraction model (Claude Sonnet 5) makes that call across your samples, and Claude Haiku makes a separate judgement when scoring a draft against your profile. Both are estimates, not tallies. ## Related - [Rhetoric attributes](https://scriptgrain.com/attributes/rhetoric) - [Opening style](https://scriptgrain.com/attributes/opening-style) - [Closing pattern](https://scriptgrain.com/attributes/closing-pattern) - [Repetition as emphasis](https://scriptgrain.com/attributes/repetition-as-emphasis) - [The anatomy of a writing voice: 45 measured attributes across 8 layers](https://scriptgrain.com/reference/writing-voice-attributes) - [AI-isms in marketing copy: 30 phrases and 10 sentence shapes, counted per 1,000 words](https://scriptgrain.com/reference/ai-isms-in-marketing-copy) ## Sources - [How ScriptGrain scores voice match](https://scriptgrain.com/reference/voice-measurement-framework)