Competitive evidence · Auto Genius “Speed Layer”
They sell the test score.
We measured what Google ranks on.
Auto Genius markets dramatic speed gains from a bolt-on script — scores doubling, load times “improving” up to +981%. Those are lab numbers, and lab numbers are gameable. Below is the field reality — real-user Core Web Vitals, the metric that actually drives Google rankings — for every dealer on their own sheet.
dealers measurable in the field still FAIL Core Web Vitals
average score gain Auto Genius claims across these same sites
actually pass in the field — plus 2 with too little real traffic to even measure
Lab · what AG shows you
A synthetic Lighthouse run in a clean lab. It can be inflated by streaming HTML, deferring scripts past the measurement window, or hiding content — moving work, not removing it. A high lab score can coexist with a slow page.
Field · what Google ranks on
28-day 75th-percentile metrics from real Chrome users (CrUX). It cannot be gamed by a script — it reflects what shoppers actually experience. This is the number that moves search rankings.
The receipts
Every dealer Auto Genius listed, their claimed Speed-Layer gains beside the live field result. A struck-through claim is one the field contradicts.
| Dealer | AG claim · score | AG claim · LCP | Field LCP (p75, real users) | Core Web Vitals |
|---|---|---|---|---|
| mclartymazda.com | 25 → 42 68.0% | 41.1s → 16.5s | 7.11s | FAILS FIELD |
| mclartyvw.com | 25 → 33 32.0% | 34.5s → 24.3s | 6.91s | FAILS FIELD |
| machensfordcapitalcity.net | 27 → 48 77.8% | 21.9s → 10.9s | 4.92s | FAILS FIELD |
| cdjrofcolumbia.com | 16 → 35 118.8% | 11.9s → 11.7s | 4.19s | FAILS FIELD |
| machensford.com | 27 → 72 166.7% | 17.3s → 5.7s | 3.89s | FAILS FIELD |
| mclartynissanlr.com | 28 → 38 35.7% | 20.3s → 10.2s | 3.76s | FAILS FIELD |
| mclartynissanbenton.com ★ | 31 → 43 38.7% | 20.9s → 15s | 3.73s | FAILS FIELD |
| joemachensnissan.com | 31 → 45 45.2% | 12.7s → 8.6s | 3.48s | FAILS FIELD |
| joemachenshyundai.com | 26 → 29 11.5% | 19.5s → 14.4s | 3.42s | FAILS FIELD |
| joemachensvw.net | 27 → 32 18.5% | 14.6s → 11s | 3.40s | FAILS FIELD |
| graydanielslincoln.com | 31 → 58 87.1% | 11.5s → 6.6s | 3.24s | FAILS FIELD |
| gray-danielstoyota.com | 30 → 46 53.3% | 8.3s → 9.9s | 3.07s | FAILS FIELD |
| mclartyvolvocars.com | 32 → 66 106.2% | 45.8s → 5.6s | 2.88s | FAILS FIELD |
| graydanielsford.com | 29 → 56 93.1% | 12.8s → 4.6s | 2.80s | FAILS FIELD |
| capcitycdjr.com | 28 → 45 60.7% | 15.2s → 12.7s | 2.75s | FAILS FIELD |
| mclartyhonda.com | 27 → 45 66.7% | 24.7s → 6.7s | 2.68s | FAILS FIELD |
| bmwoflittlerock.com | 40 → 42 5.0% | 28.8s → 30.9s | 2.65s | FAILS FIELD |
| machenslincoln.net | 42 → 67 59.5% | 10.3s → 4.3s | 2.48s | FAILS FIELD |
| kiaofcolumbia.com | 31 → 52 67.7% | 13.2s → 3.1s | 2.38s | FAILS FIELD |
| markmclartyford.com | 27 → 53 96.3% | 13.3s → 7.8s | 2.21s | FAILS FIELD |
| mazdaofcomo.com | 24 → 26 8.3% | 17.7s → 15.9s | 1.45s | FAILS FIELD |
| hopechevroletgmc.com | 26 → 57 119.2% | 13.4s → 8s | — | NO FIELD DATA |
| markmclartylincoln.net | 30 → 60 100.0% | 8.8s → 6.2s | — | NO FIELD DATA |
| graydanielsnissanbrandon.com ★ | 32 → 41 28.1% | 17.7s → 18.3s | 2.43s | PASSES FIELD |
| markmclartytoyota.com | 28 → 48 71.4% | 14s → 6.2s | 2.35s | PASSES FIELD |
| graydanielsnissanofjackson.com | 29 → 39 34.5% | 31.8s → 20.4s | 2.29s | PASSES FIELD |
| graydanielschevy.com | 26 → 68 161.5% | 17.3s → 1.6s | 1.96s | PASSES FIELD |
| joemachenstoyota.com | 37 → 48 29.7% | 17.4s → 4.8s | 1.84s | PASSES FIELD |
| hopeauto.com | 28 → 51 82.1% | 24.3s → 7.3s | 1.76s | PASSES FIELD |
| mclartyauto.com | 33 → 64 93.9% | 9.1s → 4.9s | 1.63s | PASSES FIELD |
| machens.com | 32 → 60 87.5% | 11.1s → 6s | 1.61s | PASSES FIELD |
| columbiahonda.com | 30 → 66 83.3% | 15.2s → 5.1s | 1.53s | PASSES FIELD |
CWV “pass” = field p75 LCP ≤ 2.5s, INP ≤ 200ms, CLS ≤ 0.1. ★ = independently deep-dived below. Click a dealer for the full field breakdown.
What's actually possible
We didn't just critique — we built the honest version and measured it the same way. The lever is real; it works where a site's architecture lets it.
What's actually possible — on a lean, server-rendered site
delmotors.com (Spif.io-built) · Same edge rewrite, same fair A/B (median of 5), image optimization OFF
Real, reproducible wins — because the HTML+CSS actually define the page. This is the lever. It works where the architecture lets it.
AG layer, isolated
With AG's Speed Layer running vs blocked, LCP was unchanged (~1.79s lab), TTI was 760ms SLOWER, and the performance score dropped 3 points. The layer relocates main-thread work; it does not remove it.
Field CrUX p75 LCP 3.74s → still FAILS Core Web Vitals.
Our rewrite, on DealerInspire
FCP −0.4s and TBT −0.4s, but Speed Index +5.6s (44% WORSE) and LCP flat. The page is client-rendered: 161 scripts + 53 stylesheets assemble it after load, so critical CSS can never be complete.
Exhibit A — Auto Genius's own marketing sheet
The source of every “AG claim” figure above. Transcribed as-is; shown here so the claim is verifiable.
Every number here is reproducible.
The exact method — field pull, lab A/B, isolation test — is documented.