Grok Fingerprint

Technical description

F4 scores a document on two components and normalises the sum to the 0 to 5 scale. The lexical component sums weight × occurrences over a per-language dictionary of Grok-associated phrases weighted 1 to 5 and concentrated on a casual, irreverent register (bottom line, game changer, no brainer at 3; real talk, the short answer is, plot twist, fun fact at 4; let's break this down, here's the deal, buckle up, here's the kicker, let's be real, not gonna lie at 5); a matched phrase that also belongs to the shared cross-model generic set is multiplied by 0.25 first. The structural component adds informal tone, an exclamation sentence or at least three contractions (+4); short paragraphs averaging fewer than three sentences (+2); editorial opinion presented as fact (+3); and rhetorical-question density, scored as min(2.0 + (q − 0.15) × 6, 3.0) when a document of at least eight sentences carries at least three questions at a question fraction q of 0.15 or more. The raw total R maps to the reported score as min(5.0, R / 15 × 5). Twelve language dictionaries are available; the document language selects one.

How it works

The implementation is deterministic and runs at Layer 1 over compiled regular expressions.

Lexical scoring. The active per-language dictionary maps each phrase to an integer weight from 1 to 5, rising with how distinctive the phrase is of Grok output. The weight concentrates on the model's conversational, attention-grabbing openers, here's the deal, buckle up, here's the kicker, let's be real, not gonna lie at weight 5, real talk, plot twist, fun fact, the short answer is at weight 4, and milder colloquialisms such as bottom line, game changer, no brainer at weight 3. Each case-insensitive match adds its weight times its occurrence count, and a phrase held in the shared cross-model set has its weight multiplied by 0.25 first, leaving the Grok-specific residue. Matches of weight 4 or more are reported at warning severity, lighter matches at informational severity.

Structural scoring. Four signatures contribute. Informal tone, the presence of an exclamation-terminated sentence or of at least three contractions, adds 4 and captures the conversational register out of place in academic prose. Short paragraphs, a mean paragraph length below three sentences, add 2 and capture the punchy, fragmented pacing of chat answers. Editorial opinion presented as fact adds 3 and captures the model's tendency to assert a stance. Rhetorical-question density is scored on a curve: in a document of at least eight sentences with at least three questions, once the question fraction q reaches 0.15 the contribution is min(2.0 + (q − 0.15) × 6, 3.0), so a fraction at the threshold contributes 2 and rises with reader-directed questioning to a ceiling of 3.

Aggregation. The lexical sum and the four structural contributions are added into a raw score R, reported as min(5.0, R / 15 × 5). The raw score and the detected phrases with their counts and effective weights are returned in the metadata.

Score thresholds

Score	Meaning
0 to 1	Formal register, full sentences, few colloquial openers, no reader-directed questioning.
2 to 3	A concentration of casual openers, or one structural signature such as exclamations, short choppy paragraphs, or moderate rhetorical questioning.
4 to 5	The irreverent register, informal punctuation and a high density of rhetorical questions co-occur. Strongly consistent with unedited Grok output dropped into a formal context.

Why this matters

Grok is positioned as the conversational, less-filtered assistant, and its output carries that register. It opens with attention-grabbing colloquialisms, addresses the reader directly, asserts opinions, and paces its answers in short, punchy paragraphs broken up by rhetorical questions. In casual contexts none of this is remarkable; in an academic manuscript each habit is conspicuous, because formal writing rarely exclaims, contracts, or turns to the reader with think about it this way. The rhetorical-question signal is the most distinctive of the set and is scored on a graded curve rather than a flat cutoff, since occasional questions are normal but a sustained fraction of reader-directed questions is a pacing device specific to the conversational register. F4 isolates the Grok-specific part of this profile by discounting the colloquial vocabulary it shares with other assistants through the cross-model generic discount.

Limitations

Grok is the least-studied of the major models, so the lexicon rests more on observed output than on published corpus analysis and will need more frequent recalibration than the better-characterised profiles. The casual register overlaps heavily with genuinely informal human writing, blog posts, opinion columns, popular-science explainers, which can reach a moderate score without any machine involvement, so the indicator reports resemblance to a profile and weighs several signals rather than treating informality alone as decisive. The lexical signal yields to a single editing pass that formalises the openers, and the structural signals weaken under ordinary copy-editing. The dictionaries are most developed for English and thinner across the other eleven languages; the structural checks are language-independent except for the contraction pattern, which is English-oriented.

Theoretical background

F4 applies the excess-vocabulary logic shared across the F-series to a register rather than to a topical vocabulary: the markers are the conversational openers and reader-directed devices that rose with the chat-assistant era, restricted here to the irreverent variant characteristic of Grok and separated from the assistant vocabulary common to all systems by the cross-model generic discount. The structural strand draws on register analysis, in which informality is measured through countable features, contractions, exclamations, question density, paragraph length, rather than through impression; F4 turns those features into fixed and graded contributions. The reader-directed-question measure in particular follows the engagement dimension of interactional discourse, where rhetorical questions are a device for managing reader attention and so become diagnostic when their density exceeds what argued prose normally uses.

References

1. Kobak D, González-Márquez R, Horvát EÁ, Lause J. Delving into LLM-assisted writing in biomedical publications through excess vocabulary. Science Advances. 2025. https://arxiv.org/abs/2406.07016
2. Liang W, Zhang Y, Wu Z, Lepp H, Ji W, Zhao X, Cao H, Liu S, He S, Huang Z, Yang D, Potts C, Manning CD, Zou J. Quantifying large language model usage in scientific papers. Nature Human Behaviour. 2025. DOI: 10.1038/s41562-025-02273-8 https://www.nature.com/articles/s41562-025-02273-8
3. Thelwall M, Kousha K. Have LLM-associated terms increased in article full texts in all fields? arXiv preprint arXiv:2604.07565. 2026. https://arxiv.org/abs/2604.07565
4. Basani AR, Chen PY. Diversity boosts AI-generated text detection. arXiv preprint arXiv:2509.18880. 2025. https://arxiv.org/abs/2509.18880