aiseo 0.7.1

Agent-first CLI for SEO, GEO (generative engine optimisation), and AEO (answer engine optimisation) audits.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95

//! Position-bias analysis. Kevin Indig's Feb 2026 Growth Memo write-up
//! (growth-memo.com) on ChatGPT citation distribution argues citations
//! concentrate in the first third of the body. The exact percentages
//! reported in secondary summaries (~44% in the first 30%) have not been
//! cross-checked against the primary dataset; treat the 30% boundary as
//! engineering judgement, not a hard number. Flag when high-leverage
//! signals — TL;DR, first statistic, first credential mention — sit
//! below that mark.

use once_cell::sync::Lazy;
use regex::Regex;
use serde::Serialize;

static TLDR_RE: Lazy<Regex> = Lazy::new(|| Regex::new(r"(?i)TL;?DR:?\s*").unwrap());
static STAT_RE: Lazy<Regex> = Lazy::new(|| {
    Regex::new(
        r"(?i)(\d+(?:,\d{3})*(?:\.\d+)?)\s*(%|percent|years?|months?|patients?|people|users?|cases?)",
    )
    .unwrap()
});
static CREDENTIAL_RE: Lazy<Regex> = Lazy::new(|| {
    Regex::new(r"\b(MD|PhD|Ph\.D\.|M\.D\.|MBA|MSc|MPH|DDS|DMD|JD|RN|DO|DPM|OD|PharmD|DVM|EdD|PsyD)\b")
        .unwrap()
});

#[derive(Serialize)]
pub struct PositionBias {
    pub total_words: usize,
    pub tldr_position_pct: Option<f64>,
    pub first_stat_position_pct: Option<f64>,
    pub first_credential_position_pct: Option<f64>,
    pub warnings: Vec<String>,
}

pub fn analyze(body_text: &str) -> PositionBias {
    let words: Vec<&str> = body_text.split_whitespace().collect();
    let total = words.len();

    if total == 0 {
        return PositionBias {
            total_words: 0,
            tldr_position_pct: None,
            first_stat_position_pct: None,
            first_credential_position_pct: None,
            warnings: Vec::new(),
        };
    }

    let tldr_off = word_offset(body_text, &TLDR_RE);
    let stat_off = word_offset(body_text, &STAT_RE);
    let cred_off = word_offset(body_text, &CREDENTIAL_RE);

    let pct = |off: Option<usize>| -> Option<f64> {
        off.map(|o| ((o as f64 / total as f64) * 1000.0).round() / 10.0)
    };

    let tldr_pct = pct(tldr_off);
    let stat_pct = pct(stat_off);
    let cred_pct = pct(cred_off);

    let mut warnings = Vec::new();
    // The 10% TL;DR threshold is engineering judgement; Indig's data supports
    // 30% as the primary citation-window boundary.
    if let Some(p) = tldr_pct
        && p > 10.0
    {
        warnings.push(format!(
            "TL;DR sits at {p}% of body. Move into the first 10%."
        ));
    }
    if let Some(p) = stat_pct {
        if p > 30.0 {
            warnings.push(format!(
                "First statistic at {p}% of body. Front-load citation-worthy numbers."
            ));
        }
    } else if total >= 200 {
        warnings.push(
            "No statistics detected. Named numerical claims lift AI citation.".to_string(),
        );
    }

    PositionBias {
        total_words: total,
        tldr_position_pct: tldr_pct,
        first_stat_position_pct: stat_pct,
        first_credential_position_pct: cred_pct,
        warnings,
    }
}

fn word_offset(text: &str, re: &Regex) -> Option<usize> {
    let m = re.find(text)?;
    Some(text[..m.start()].split_whitespace().count())
}