Filler phrases analyser

By Jack Stovell · published 2026-09-24

Filler phrases are the casual expressions a writer reaches for without noticing, such as "to be fair" or "at the end of the day". A ScriptGrain voice profile lists up to five per analysis, quoted verbatim from your samples. When a draft is scored, code counts hits on that list, with weight 2, alongside your other signature phrases.

Analyse your filler phrases free

Your free voice profile measures filler phrases among all 45 attributes, from a sample of your own writing. No card needed.

Try the six-signal check first

At a glance

What filler phrases measures

Filler phrases measures the casual expressions you keep reaching for, catalogued exactly as you wrote them rather than picked from a fixed list. The app calls them "filler expressions you reach for repeatedly", and that's the whole definition: no scoring rubric hiding inside it, just a record of habit.

Here's the thing about how it's built. Claude Sonnet 5 reads each sample in pass one, then synthesises everything into one profile in pass two. For this attribute specifically, the phrases are collected verbatim; whatever you actually typed is what ends up in the list. You can edit that list by hand on the Profile page afterwards, so if the model catches something you'd rather drop, you drop it. It sits in the Lexical group with preferred words, the other hand-editable list, and beside contraction frequency, another habit that makes prose sound spoken.

A separate free tool counts five common fillers in code. The free AI cliché checker counts these "systematic casual markers" ("I mean", "you know", "right?", "sort of" and "kind of") only when they recur. Its note reads: "Fine once. A tell when they arrive at regular intervals." That's a different job from the profile attribute itself, which just wants your list, not a verdict on whether you overuse it.

How filler phrases is scored

Filler phrases moves the voice match score through the shared "signature phrases" feature, with weight 2. The pool draws from four sources: your preferred_words, discourse_markers, filler_phrases and preferred_phrasings. From that pool, the system keeps multi-word phrases and single words of six or more letters, up to fifteen entries total.

For a similarity score to be calculated at all, the pool needs at least three entries. Once it has them, similarity is hits divided by the smaller of two numbers: four, or the pool size. That result is capped at 1. When a draft is scored, code counts the hits. With a pool of three, three hits max it out; with four or more, any four do.

If you want the full mechanics of how voice match adds up across all 28 scoring attributes, that lives on the voice measurement framework page, and the voice match glossary entry defines the score. The app labels the result: 90%+ is Excellent, 75 to 89% is Good, 60 to 74% is Fair, under 60% is Low.

How to analyse filler phrases in your own writing

You can find your own filler phrases by hand with a few unedited pieces and a highlighter, metaphorical or otherwise.

  1. Pull three or four pieces you wrote without much editing (emails count, so do first drafts).
  2. Read them aloud. Filler phrases hide on the page but announce themselves the moment you speak them.
  3. Mark every phrase that feels like a verbal tic rather than a deliberate word choice, things like "so", "fair enough", "at the end of the day".
  4. Count occurrences per piece. A phrase that shows up once is a choice. A phrase that shows up in every piece is a habit.
  5. Keep the five that recur most and write them down as your list.

ScriptGrain's free voice profile reads this straight from a writing sample, as one of all 45 attributes.

Examples of filler phrases in real writing

Filler phrases show up as small, recognisable expressions that a writer leans on across multiple pieces, not one-off word choices. Here are three, each labelled:

Example: "To be fair, the numbers were never going to hit target this quarter."

Example: "So that's the plan, more or less."

Example: "At the end of the day, nobody remembers the slide deck."

None of these are wrong on their own. The tell is repetition across samples, the same phrase turning up in the newsletter, the pitch email and the blog post. That pattern is what the extraction model catalogues, up to five phrases per analysis.

How to make AI use your filler phrases

Getting an AI assistant to sound like you, fillers included, means telling it exactly which fillers are yours and where they belong.

  1. List the fillers you actually use, not the ones you wish you used. Be honest; check your sent emails if you're not sure.
  2. Ask the model to place them only where a speaker would naturally pause, never at fixed intervals like the start of every paragraph.
  3. In a ScriptGrain profile, edit Filler phrases directly on the Profile page so the list reflects your voice, not a guess.

When a ScriptGrain profile writes for you, filler_phrases sits inside the voice profile JSON block sent to the model, treated as a hard constraint rather than a suggestion, and any phrase on the list is exempt from the generator's bans on stock AI phrases. The profile reaches ChatGPT, Claude and other tools through the API and the MCP server, so they read the same list.

How to make AI use fewer filler phrases

Cutting fillers from AI writing starts with naming them, then banning them outright.

  1. Name the specific fillers you want gone, such as "you know" and "sort of", instead of a vague "sound more professional".
  2. In ScriptGrain, add those phrases to Banned terms, so the generator is told to avoid them while it writes.
  3. Ask for an editing pass that deletes any filler whose removal leaves the sentence's meaning intact. Most of them qualify.
  4. Scan the finished draft yourself for recurring casual markers such as "you know" and "sort of", since even a banned list can miss a paraphrase.

The free AI cliché checker counts those casual markers when they recur, so it works as a final check on a draft.

Questions

What are filler phrases in writing?

Filler phrases are the casual expressions a writer repeats across pieces without deliberate intent, things like "fair enough" or "at the end of the day". They're not errors. They're habits, and a voice profile catalogues them verbatim from your samples rather than matching them against a generic list, capped at five entries per analysis in ScriptGrain's system.

How is filler phrase use scored?

Filler phrases enters voice match through the shared signature phrases pool, combined with preferred_words, discourse_markers and preferred_phrasings. The pool keeps multi-word phrases and words of six-plus letters, up to fifteen entries, needing at least three to score at all. Similarity is hits divided by the smaller of 4 or pool size, capped at 1, weighted at 2 overall.

How do I find filler phrases in my own writing?

Read your own drafts aloud rather than silently; fillers hide on the page. Pull a few unedited pieces, mark phrases that sound like verbal tics rather than deliberate choices, and count how often each recurs across samples. A phrase appearing once is coincidence. One appearing in every piece is your filler. ScriptGrain's free voice profile reads the list from your samples as one of its six lexical attributes.

How do I make ChatGPT use fewer filler words?

Name the specific fillers explicitly rather than asking for "more professional" writing. Ban them by name, then request a pass that removes any filler whose deletion doesn't change the meaning. In ScriptGrain, adding them to Banned terms puts them among the generator's hard constraints, and a banned term overrides the voice profile.

How do I make AI writing sound more conversational?

Allow your genuine fillers back in, placed where you'd naturally pause when speaking, not at mechanical intervals. Pair that with more contractions and fewer rare words, since a conversational tone rarely comes from one habit alone. In ScriptGrain, the Conversational register tells the generator to use contractions freely.

Related

Sources