harvey.ai website review

About 953 KiB of JavaScript is downloaded but never used on this page. Google estimates it would make the page about 2.7 s faster on a phone.

23 files are served with short cache lifetimes, so returning visitors download them again (581 KiB).

A phone's processor spends 12.1 s on main-thread work while loading, 6.1 s of it running JavaScript. Above 2 s, taps and scrolls feel sluggish.

Your mobile performance score is 34. Under 50 is Google's “poor” band: many phone visitors leave before the page is usable.

The page has 6,268 HTML elements. Above 1,500, rendering and every interaction get slower on phones.

8 third-party services load with your homepage (Google Tag Manager, transcend-cdn.com and Facebook and others), adding 1.2 MB and 1.4 s of main-thread time.

20 long tasks block the main thread while loading, the longest for 1.1 s. During a long task the page ignores taps.

Google Tag Manager costs a phone 611 ms of main-thread work and 678 KiB of downloads while your page loads. Loading it after the page is usable, or only on the pages that need it, takes that off the critical path.

7 files block the first paint until they finish downloading. Deferring or inlining them could save about 650 ms.

transcend-cdn.com costs a phone 347 ms of main-thread work and 161 KiB of downloads while your page loads. Loading it after the page is usable, or only on the pages that need it, takes that off the critical path.

Facebook costs a phone 287 ms of main-thread work and 248 KiB of downloads while your page loads. Loading it after the page is usable, or only on the pages that need it, takes that off the critical path.

Polyfills for old browsers add 45 KiB JavaScript that modern phones do not need. Google estimates it would make the page about 150 ms faster on a phone.

2 images could be 147 KiB smaller with modern formats (WebP or AVIF) and sizes that fit a phone screen.

Critical files load one after another instead of in parallel (3 chains). Preconnecting to key origins and preloading the hero image shortens the chain.

Scripts make the browser recalculate the layout repeatedly while the page loads, which wastes processor time on phones.

5 KiB could be saved by minifying JavaScript.

Real Chrome users on phones do not pass Google's Core Web Vitals on this page (28-day field data from the Chrome UX Report (opens in a new tab)). Google uses this in ranking.

Real visitors wait 878 ms for the page to respond to a tap (75th percentile). Good is under 200 ms.

Your largest element takes 22.4 s to render on a phone. Google's threshold for “good” is 2.5 s. The largest element is the image “Harvey’s Command Center tracking a firm’s adoption rate, ac…”.

The page becomes fully interactive after 26.7 s. Good is under 3.8 s.

Visually, the page fills in at 12.9 s (Speed Index). Good is under 3.4 s.

Scripts block taps and scrolling for 2.5 s in total while the page loads. Good is under 200 ms.

The first text or image appears after 2.3 s on a phone. Good is under 1.8 s.

Over the last 25 four-week periods to 2026-11-07, real Chrome visitors' largest content (lcp) went from 1.9 s to 2.0 s at the 75th percentile (Chrome UX Report (opens in a new tab) history).

Your homepage downloads 5.7 MB in 253 requests on a phone: scripts 2.7 MB (73), other files 1.8 MB (76) and images 726 KiB (91). Under 2 MB keeps phones on slow connections fast.

By the Sustainable Web Design model, a visit to your homepage emits about 1.61 g of CO₂e: 192.6 kg a year at 10,000 visits a month. It scales with page weight, so every byte cut lowers it.

ARIA dialog and alertdialog nodes should have an accessible name: 1 element on the page fails this check, rated serious by axe-core.

Landmarks should have a unique role or role/label/title (i.e. accessible name) combination: 1 element on the page fails this check, rated moderate by axe-core.

Structured data found (Corporation, ItemList, VideoObject and WebPage) but with 1 problem, for example “Corporation: missing name”.

7 links use generic text like “click here” or “learn more”, which tells Google nothing about the page behind them.

The homepage links to 164 pages on harvey.ai and 19 external pages.

There is no https://www.harvey.ai/llms.txt, a short Markdown guide to your key pages for AI tools (llmstxt.org). It is optional and few AI tools read it yet, so treat it as a cheap extra.

We fetched 15 pages linked from your homepage (of 54 found) and read each the way a search engine does.

3 errors were logged to the browser console while the page loaded, for example “Cannot read properties of undefined (reading 'getRegistrations') <anonymous>:3:39”.

Large scripts ship without source maps.

Your homepage shows a street address but no LocalBusiness JSON-LD, so Google and AI assistants have to guess your address, hours and phone from the page text.

A local customer looking to call finds no phone number on the homepage. Add one, ideally as a tap-to-call link.

No phone number or email link on the homepage, so ready-to-buy visitors have one fewer way to reach you.

The homepage shows no opening hours. If customers visit or call, show when.

There is no form on the homepage, so every lead has to click through to another page first.

There is no link to Google or Apple Maps. A directions link is one tap for a phone visitor.

No guarantee or risk reducer (money-back, free trial, no credit card) is mentioned, which is often what tips a hesitant visitor.

The homepage carries about 934 words of visible text.

Before a visitor clicks anything, the page loads Amazon Ads, Google Ads, Google Analytics, Google Tag Manager and LinkedIn Insight and 4 mores (measured from a US location before any click). No consent tool was detected. Visitors from the EU, UK and several US states must be asked first.

Opening the homepage sets 7 cookies (measured from a US location before any click).

No known cookie consent tool loads with the page. Some sites build their own banner, and some only show one to visitors from the EU.

Loading the homepage contacts 26 hosts run by other companies (fonts, scripts, analytics, ads), each of which sees the visitor's IP address.