Methodology

Every AI answer starts with a prompt.

Scroll through the story of how a single question becomes a reproducible visibility score across the 8 engines shaping how buyers discover you.

01Chapter 1 — The prompt graph

Buyers don't ask one question.

We build a category-specific prompt graph from your industry, sub-vertical, geography and competitor set. Every prompt is tagged by buying-stage intent — so the scoring weights match what actually drives pipeline.

Discovery
"best AI SEO tool for B2B SaaS"
intent weight
Comparison
"best AI SEO platform for growth teams"
intent weight
Evaluation
"does citerank support Gemini and Claude"
intent weight
Conversion
"citerank pricing for growth teams"
intent weight
02Chapter 2 — Multi-LLM execution

The same prompt, replayed 10–30 times across 8 engines.

LLMs are non-deterministic — ask the same question twice and you'll get different answers. We replay every prompt 10–30 times per engine to control for temperature and answer drift, then average. Per plan: Basic runs 10 replays, Growth 15, Premium 15–20; Enterprise scales further on request.

8 engines × 12 replays shown
ChatGPTGeminiClaudePerplexityGoogle AIOCopilotGrokMeta AI
cited in response not cited
03Chapter 3 — Why one query isn't enough

Answer drift is real. Confidence bands make it honest.

We report every score with a 95% confidence interval so single-run noise doesn't drive decisions. If the band is wide, you know before you act.

Visibility per replay
Same prompt · Same engine · 10 runs
52 ± 9
95% CI
run 1run 10
04Chapter 4 — Citation detection

Every URL, every mention, resolved to one entity.

Responses are parsed for explicit citations, inline brand mentions and entity-level references. Aliases, typos and domain variants all collapse to the same brand so nothing is double-counted or missed.

parsed response
citerank.in
URL citation
CiteRank AI
brand mention
CiteRank
alias resolved
cite-rank
typo variant
competitor.io
entity separated

One brand, many surfaces.

We deduplicate near-mentions and resolve corporate aliases so "CiteRank", "CiteRank AI" and the .in domain all match a single entity. Sentiment and recommendation-order are extracted per mention.

05Chapter 5 — The composite score

Four signals become one score, 0–100.

Intuition first, math second. Each signal has a plain-language explanation — expand any card to see the formula, why it exists and how it's weighted.

share_b = Σ mentions_b / Σ mentions_all

Why it exists — Baseline signal — presence in the answer at all.

order_b = Σ (1 / position_b,i)

Why it exists — Buyers act on what's recommended first.

sent_b ∈ [-1, +1]

Why it exists — Being cited badly is not the same as being cited well.

w_p = 1 / (1 + σ_p)

Why it exists — Stable signals should count more than lucky runs.

Composite
score_b = Σ_p w_p · ( α·share + β·order + γ·sent )

Weights (α, β, γ) are published in-product per intent bucket. Every score is reproducible from the same prompt graph and seed range.

06Chapter 6 — Gaps, re-runs, attribution

A score without a fix is a vanity metric.

Every brand score has matched competitor scores. Where you lose, we ship a gap-fix brief — the specific prompts you don't appear on, the sources cited instead, and the entity, schema or content changes that would close the gap.

Gap-fix briefs

Prompt-level briefs listing the sources currently cited and the schema/entity moves to close the gap.

Weekly re-runs

Automated re-runs (daily on Growth+) so you see movement, not a snapshot.

Downstream attribution

Optional GA4, Search Console and HubSpot links tie AI visibility shifts to sessions and pipeline.

07Chapter 7 — Who it's for

Built for teams who need to defend a number.

Marketing leaders

Prove AI visibility movement to the board with reproducible scores and CIs.

SEO / GEO teams

See exactly which prompts and sources are winning — and what to ship next.

Agencies

White-label reports, per-client prompt graphs, and gap-fix briefs baked in.

Product & PR

Track how launches and press cycles land in the actual AI answer surface.

What we don't do

No single-shot prompts. No black-box scoring. No fabricated samples.

  • · Every score is averaged across 10–30 replays per engine (Basic 10 · Growth 15 · Premium 15–20).
  • · The composite formula and weights are published in-product.
  • · Marketing samples are labelled "Illustrative"; customer reports use only that customer's real runs.