Conversion (CRO) · Methodology
How we score where your page leaks conversions.
A heuristic audit — the friction a senior CRO spots by eye from real desktop and mobile screenshots, not statistical A/B testing. Eleven weighted dimensions, scored 0–10.
The 11 conversion dimensions
| # | Dimension | Weight | What it scores |
|---|---|---|---|
| 01 | clarity_5s | 15% | Within 5 seconds, is it obvious what this is and what to do next? Judged from the above-the-fold screenshot. 10 = headline + subhead + one CTA tell the whole story. |
| 02 | single_primary_cta | 12% | Is there one primary action, visually dominant? 0 = five equal buttons (= no priority). |
| 03 | inp_cwv | 12% | Core Web Vitals — CrUX p75, INP heaviest, then LCP, CLS. 10 = all “good” (INP<200ms, LCP<2.5s, CLS<0.1). |
| 04 | form_friction | 10% | Field count + required fields + perceived effort. 10 = one field (email); 0 = a 12-field gauntlet for a newsletter. |
| 05 | mobile_experience | 10% | From the mobile screenshot: tap-target size, legibility, CTA reachability, no horizontal scroll, viewport meta. |
| 06 | traffic_source_match | 8% | Does the page match the promise of how visitors arrive (ad → message match, email → continuity)? 0 = a generic homepage for a specific-intent visitor. |
| 07 | social_proof | 8% | Presence and quality of proof near the decision. 0 = none, or a vague “trusted by thousands.” |
| 08 | ai_visitor_readiness | 8% | Can an AI agent parse the offer + act? Clear text, semantic structure, machine-legible pricing/CTA. 0 = everything locked in images/JS. |
| 09 | readability | 6% | Flesch-Kincaid grade + scannability. 10 = grade ≤8, scannable; 0 = dense, jargon-heavy walls. |
| 10 | objection_cta_repetition | 6% | Are objections handled and the CTA repeated at decision points down a long page? 0 = one CTA at top, none after. |
| 11 | nav_visual_noise | 5% | Competing links / clutter vs focus. 0 = a 30-link mega-nav stealing the click. |
Overall (0–100) = the weighted mean ×10. We score mobile and desktop separately and the headline takes the worst-of per dimension — a page that converts on desktop but not mobile is broken, because most traffic is mobile. A persona pre-score (founder / marketer / developer / agency / enterprise) conditions every dimension but is not itself weighted.
The 5 traffic-source rule sets
| Source | How it re-weights the score |
|---|---|
email | Warm — clarity, CTA and objection handling carry it. |
ai-referred | Up-weight ai_visitor_readiness, readability, clarity (AI sends users expecting a fast, legible answer). |
paid-search | Expensive, high-intent — protect CTA, clarity, form, message-match. |
organic | Researching — readability, social proof, clarity. |
social | Cold, distracted, mostly mobile — instant clarity, mobile, proof. |
The cold-buyer persona
The cold-buyer reaction is CRO’s equivalent of AEO’s engine receipts — visceral evidence behind the scorecard. We construct a single LLM persona representing the page’s ideal cold buyer for the detected ICP, then run it against the captured screenshots and page text and ask for its verbatim first-five-seconds reaction. The card carries the persona description, the exact prompt used, the model, the verbatim quote, and a verdict pill — bounce-likely / engaged / converts — plus a pointer to the fix that would change the verdict. We never invent the reaction; it is the model’s actual output, retained with the report.
The 2026 page-pattern flags
Six yes/no checks for the 2026 baseline patterns most audits skip, each with a measured 2026 lift cited inline:
| Pattern | Definition | 2026 signal |
|---|---|---|
| Autoplay video hero | An auto-playing hero video above the fold. | −7% conversion (Digital Applied 2026) |
| One-question intent routing | A single qualifying question that routes the visitor to the right path. | +35% (Jimo 2026 onboarding) |
| Interactive demo embed | An embedded interactive product demo, not a static screenshot. | +30% (Storylane benchmarks) |
| Transparent pricing | Real prices on the page, not “contact sales.” | high-intent capture |
| Server-side tagging | Analytics events sent server-side, not client-only. | +25% event capture (Transcend 2026) |
| Consent Mode v2 compliant | Consent signals wired to Google Consent Mode v2. | 67% of sites non-compliant (2026 audit) |
Revenue at stake — and the bench
The headline dollar figure always ships with its stated assumption (“at an assumed 10,000 visits/mo and a 2% baseline conversion, ~$X/mo is on the table”) — never as a measured number. The cover bench card sets your conversion rate against an industry benchmark (median + top quartile) so the CFO gets context before the scroll. Benchmarks are sourced from CXL’s B2B SaaS conversion benchmarks and the Digital Applied 2026 bounce study; CWV thresholds are web.dev’s “good” bands.
Myths we refuse to score
- A/B button-color tests (red vs green).
- Heatmaps without a funnel behind them.
- Exit-intent popups as a conversion strategy.
- Live chat for cold traffic.
- Generic trust seals / badges.
- Hick’s Law quoted as gospel.
See also the AEO and Brand Perception methodologies, or browse the audit library.