The SwiftGeo Citation Score measures how often and how reliably AI engines cite your brand when buyers ask relevant questions — not when they search for you by name. Every number on this page is reproducible, auditable, and tied to real engine output. Nothing is estimated or inferred.
SwiftGeo Citation Score is reproducible within ±10 points across five independent runs. Zero-citation brands score zero. Inconsistently-cited brands score lower than reliably-cited brands regardless of fame. Every component only scores on organic buyer-intent queries where the brand was genuinely mentioned — not dismissed, not echoed from a comparison query.
The Formula
Your Citation Score is computed in two stages: a raw score across five weighted components, then multiplied by a Consistency Coefficient (Cc) that penalizes inconsistency across scan runs.
raw = (
mention_rate × 10 × 0.35 // Are you being cited?
+ avg_prominence × 0.25 // Where in the response?
+ sentiment_normalized × 0.20 // How positively?
+ cross_engine × 0.15 // Across how many engines?
+ avg_accuracy × 0.05 // Are facts about you correct?
) × 10
// Stage 2 — Apply Consistency Coefficient
Citation Score = round(min(100, max(0, raw × Cc)))
All five components only score on organic queries — discovery and problem-type questions where the brand was genuinely mentioned. Comparison queries (e.g. "SwiftGeo vs Semrush") are excluded from all scoring because AI engines echo brand names back in fabricated responses, inflating mention rates for brands with zero real AI presence.
The Five Components
| Component | Weight | What It Measures |
|---|---|---|
| mention_rate | 35% | The share of organic queries where the engine genuinely cited your brand — not dismissed, not echoed. This is the primary signal. A brand cited in 75% of buyer-intent queries has real AI presence. A brand cited in 0% has none, regardless of how famous it is. |
| avg_prominence | 25% | Where in the response your brand appears. Being the first recommendation scores higher than being buried at the end of a list. Scored 0–10 per mention, averaged across all organic citations. |
| sentiment_normalized | 20% | Whether the engine's language about your brand is positive, neutral, or negative. Scored 0–10 per citation. Zero-mention brands score 0.0 — not neutral. Neutral framing scores 0.0, not a positive bump. |
| cross_engine | 15% | The share of the four expected engines (ChatGPT, Perplexity, Gemini, Claude) that cited your brand. Always normalized against four — a dropped engine contributes 0%, not an absence from the denominator. Being cited across all four engines scores higher than one engine alone. |
| avg_accuracy | 5% | Whether factual claims the engines make about your brand are accurate. Verified against your website content. Unknown accuracy scores 0.0 — not 5.0. This is a secondary signal; primary score integrity comes from mention and prominence. |
The Consistency Coefficient (Cc)
The Cc is SwiftGeo's core differentiator. AI engines are non-deterministic — the same query returns different responses on different runs. A brand cited 80% of the time is a different business asset than a brand cited randomly in some scans and not others, even if both show the same average mention rate on a single scan.
Each scan runs three independent passes. For every organic query, Cc is computed as:
Cc_q = 1.0 − (2 × |mention_rate_q − 0.5|)
// Final Cc = average across all organic queries
Cc = mean(Cc_q for all organic queries)
Cc=1.0 means the engine's behavior is perfectly consistent — either always cited or never cited across all three runs. Cc=0.5 means pure noise, a coin flip. The final Citation Score multiplies raw by Cc, so an inconsistently-cited brand is penalized even if its average mention rate looks reasonable.
A brand never cited across all three runs has Cc=1.0 — mathematically correct, since 0 every run is perfectly consistent. But raw=0 because mention_rate=0, so final score = 0 × 1.0 = 0. The Cc card is hidden in the dashboard when mention_rate=0 to avoid a misleading "Cc: 1.0" display for invisible brands.
Score Interpretation Guide
These ranges reflect real calibration data across known brands. They are not theoretical — they are anchored to actual engine behavior on buyer-intent queries as of March 2026.
Calibration Data
We run calibration scans against known brands to validate that scores reflect reality. The table below shows official calibration results from March 30, 2026 — post all methodology fixes. Pre-fix numbers exist only in our audit trail and are not published here.
| Brand | Score | Expected Range | Cc | Mention % | Engines | Pass |
|---|---|---|---|---|---|---|
| HubSpot | 66 | 60–80 | 1.000 | 75% | 4 / 4 | ✓ YES |
| Salesforce | 57 | 50–70 | 1.000 | 65% | 4 / 4 | ✓ YES |
| Surfer SEO | 36 | 25–45 | 1.000 | 18% | 4 / 4 | ✓ YES |
| Pintas & Mullins | 0 | 0–10 | 1.000 | 0% | 4 / 4 | ✓ YES |
| SwiftGeo | 0 | 0–5 | 1.000 | 0% | 4 / 4 | ✓ YES |
All five brands pass. HubSpot's score of 66 is notable: one of the most recognized SaaS brands in the world, yet AI engines don't recommend them in 25% of buyer-intent queries. That gap is the market SwiftGeo addresses. Our own score of 0 reflects honest measurement — we built this platform while invisible, deployed our own fixes, and rescanned to prove it worked.
AI engines are non-deterministic. A brand can score 63 on one calibration run and 69 on another — both correct. Expected ranges reflect the ±10 point reproducibility window that defines a reliable score. If your score moves more than 10 points between scans without any fix deployed, that is a signal worth investigating.
Known Limitations
We document these openly. A methodology that doesn't acknowledge its limits isn't a methodology — it's marketing.
The Closed Loop
Measurement alone doesn't improve your score. What changes it is deploying the right signals to the right places — and then proving it worked with a rescan.
After every scan, SwiftGeo identifies the specific gaps: which queries missed you, which engines didn't cite you, what the engines said instead. The deployment layer then pushes targeted fixes directly to your domain — Schema.org JSON-LD, llms.txt signals, content rewrites, directory submissions — without requiring CMS access. Then we scan again.
That before/after proof is the product. The Citation Score is how you know the fix worked.
See Your Score
Run a free scan and get your Citation Score across ChatGPT, Perplexity, Gemini, and Claude. Results in under three minutes.
Scan My Brand Free →