Skip to content
AvocadoScore

AvocadoScore audit · stripe.com

stripe.com

Unripe

Returning visitors get a workable experience, but cold loads, new-visitor performance and SEO need significant, prioritized work.

Real users:Almost ripeCold lab:Almost ripe(10 of 10 pages below Ripe)

Mobile perf
73/100
Almost ripe
SEO ready
88/100
Almost ripe
AI ready
83/100
Almost ripe

How “AI ready” is scored: 18 checks across three pillars — Access (can AI crawlers fetch you), Comprehension (can they parse & chunk you) and Trust (can they attribute & cite you).

Scanned 16 days ago10 pages4 views

Overview

Key findings

The headline takeaways, in plain English.

  1. 1In cold lab tests, 5 of 10 scanned pages render in the "Unripe" band (the overall verdict above weights real-user data, which is faster).
  2. 2Median mobile Total Blocking Time is 594 ms — about 3.0× the 200 ms target.
  3. 3The biggest opportunity: Address accessibility gaps — affects 9 pages.
  4. 4The heaviest page ships 2.1 MB of JavaScript — a primary cause of slow loads.
  5. 5Strengths: accessibility and SEO score well — focus effort on performance.
See the prioritized action plan →
Performance
73/100
12 issue types
Accessibility
96/100
3 issue types
On-page SEO
97/100
1 issue type
Best Practices
77/100
3 issue types
0Ripe5Almost ripe5Unripe

AI & search visibility

Can AI answer engines find, read and cite your site?

More people now ask ChatGPT, Perplexity and Claude instead of Googling. This section answers whether those answer engines can reach your content, understand it, and cite you — the audit other tools skip.

AI visibility verdict

AI agents can read your site; a few quick additions would make it easier to cite.

Fix next: add structured data — see the action plan for copy-paste fixes.

OG image quality

100/100

https://b.stripecdn.com/mkt-ssr-statics/assets/_next/static/media/stripe.37119086.jpg

Open Graph preview image as social platforms and AI answer engines see it
3 checks · all pass▸
  • ✓
    Reachable
    HTTP 200
  • ✓
    Content-Type
    image/jpeg
  • ✓
    File size
    305.5 KB

Crawl-as-bot diff

Homepage · GPTBot UA

GPTBot sees 12,792 chars · browser sees 12,792 chars

100%

GPTBot sees 100% of the content a browser sees — essentially identical.

llms.txt content

305 links · 31 sections

90/100
5 checks · 1 to review▸
  • ✓
    /llms.txt present
    HTTP 200
  • ✓
    Has main heading
    # Title found
  • ✓
    Link coverage
    305 markdown links
  • ✓
    Sections
    31 `##` sections
  • !
    /llms-full.txt
    Optional but recommended — expanded file with full body text of key pages
# Stripe > Stripe is a technology company that provides financial infrastructure for businesses. Businesses of every size—from new startups to established public companies—use Stripe to accept payments online and in person, embed financial services, and create custom revenue models. ## Payments -…

Robots.txt advisor

3 groups · 1 sitemap ref

80/100
5 checks · 1 to review▸
  • ✓
    robots.txt present
    3 user-agent groups
  • ✓
    Allows crawlers
    `User-agent: *` permits standard crawling
  • ✓
    Sitemap declared
    1 sitemap ref
  • ✕
    AI bot policies
    No explicit policy for GPTBot, ClaudeBot, Google-Extended, etc. — decide and declare
  • ✓
    Crawl-delay
    No crawl-delay — crawlers use their own pacing

Schema.org coverage

2/10 pages (20%)
Schema typePagesCoverage
FAQPage110%
Organization110%
WebSite110%

Readiness

SEO & AI Readiness

Two scorecards derived from the homepage HTML, robots.txt and AI-specific well-known files. The differentiating analysis no other tool does.

SEO Readiness
88/100
10 pass · 1 warn · 1 fail
AI Readiness
83/100
12 pass · 6 warn · 0 fail

A few readiness gaps to close — the checklists below name what’s missing or weak on the homepage and in robots.txt.

AI readiness vs the Index

At 83/100, this site scores higher than 62% of 922 well-known sites in the AI Readiness Index. The Index is a curated set of big names, not a sample of the web.

62nd
percentile
WorstIndex median 78Best

SEO checklist

10/12 passing
  • ✓
  • ✕
  • ✓
  • ✓
  • ✓
  • ✓
  • ✓
  • ✓
  • ✓
  • !
  • ✓
  • ✓
  • ✕
    Sitemap
    No sitemap.xml found
  • !
    H1 heading
    2 H1s — should be one
▸ 10 passing checks — show all
  • ✓
    HTTPS
    Served over HTTPS
  • ✓
    robots.txt
    Present
  • ✓
    Title tag
    54 chars
  • ✓
    Meta description
    149 chars
  • ✓
    Canonical URL
    https://stripe.com/
  • ✓
    Open Graph
    Complete (title, description, image)
  • ✓
    Twitter Card
    summary_large_image
  • ✓
    Structured data (JSON-LD)
    1 block
  • ✓
    Language declaration
    <html lang="en-US">
  • ✓
    Viewport meta
    Set

AI readiness — three pillars

12/18 passing
Access83

Can AI crawlers fetch the site at all?

  • !
  • ✓
  • !
  • ✓
  • ✓
  • ✓

4/6 passing

Comprehension85

Once fetched, can they parse and chunk it?

  • ✓
  • !
  • !
  • ✓
  • ✓
  • ✓
  • ✓
  • ✓
  • !
  • !
  • ✓

7/10 passing

Trust75

Can they attribute and date the content?

  • ✓
  • !

1/2 passing

Access — Can AI crawlers fetch the site at all?

4/6 passing
  • !
    AI answer-engine access
    OAI-SearchBot, ChatGPT-User, Claude-User, PerplexityBot, Perplexity-User are restricted in robots.txt — ChatGPT search, Perplexity and Claude can't read or cite the site through these crawlers
  • !
    Search-index crawlers (AI surfaces)
    Googlebot, Bingbot partially restricted — AI surfaces built on these indexes see a reduced site
▸ 4 passing checks — show all
  • ✓
    Training-crawler stance
    No training opt-outs declared — content may be used for model training (5 tokens checked)
  • ✓
    CDN / firewall gate
    GPTBot sees 100% of the content a browser sees — essentially identical.
  • ✓
    Snippet directives
    No nosnippet / max-snippet / noai restrictions
  • ✓
    Agent-friendly 404s
    Returns 404 for unknown paths

Comprehension — Once fetched, can they parse and chunk it?

7/10 passing
  • !
    /llms-full.txt
    Not present — optional, but useful for richer LLM context
  • !
    Markdown for agents (bonus)
    No text/markdown variant — optional, but agents parse markdown more reliably than HTML
  • !
    RSS/Atom feed
    No feed advertised in <head> — several AI systems ingest feeds for freshness
  • !
    Sitemap freshness
    No sitemap.xml — crawlers fall back to link discovery with no freshness signals
▸ 7 passing checks — show all
  • ✓
    /llms.txt
    Present — AI-friendly content map
  • ✓
    Structured data (validated)
    1 block, all required properties present (Organization, Person, WebSite)
  • ✓
    Server-rendered content
    10344 words in initial HTML
  • ✓
    Semantic HTML
    Uses main, header, footer, nav
  • ✓
    Heading outline
    57 headings in a clean hierarchy
  • ✓
    Open Graph metadata
    Complete
  • ✓
    Consent wall
    No consent-manager wall detected

Trust — Can they attribute and date the content?

1/2 passing
  • !
    Content dates
    No datePublished/dateModified in structured data — answer engines can't judge freshness
▸ 1 passing check — show all
  • ✓
    Author / Organization markup
    Person/Organization schema present

AI bot policies (robots.txt)

10 bots tracked

All 10 tracked AI bots have mixed rules.

Answer & search crawlers — blocking these removes the site from AI answers

  • OAI-SearchBotsome paths allowed, others disallowed
  • ChatGPT-Usersome paths allowed, others disallowed
  • Claude-Usersome paths allowed, others disallowed
  • PerplexityBotsome paths allowed, others disallowed
  • Perplexity-Usersome paths allowed, others disallowed

Training crawlers — blocking is an opt-out from model training and does not affect citations

  • GPTBotsome paths allowed, others disallowed
  • ClaudeBotsome paths allowed, others disallowed
  • Google-Extendedsome paths allowed, others disallowed
  • Applebot-Extendedsome paths allowed, others disallowed
  • CCBotsome paths allowed, others disallowed

mixed

▸ Vendor and status detail
BotVendorStatus
GPTBotOpenAIMixed
OAI-SearchBotOpenAIMixed
ChatGPT-UserOpenAIMixed
ClaudeBotAnthropicMixed
Claude-UserAnthropicMixed
Google-ExtendedGoogleMixed
PerplexityBotPerplexityMixed
Perplexity-UserPerplexityMixed
Applebot-ExtendedAppleMixed
CCBotCommon CrawlMixed

Declare a policy either way so the intent is clear — see the action plan for a copy-paste robots.txt block.

Readiness is sampled on the site homepage; deeper pages may differ. AI bot policies follow the llms.txt proposal and the established robots.txt convention.

📩 Save this report to your free dashboard

Everything below is already unlocked — this emails you the link and keeps every scan in one place. Free, no card. We may send an occasional reminder to re-scan; unsubscribe in one click.

Deep dive

Performance analysis

Where this site stands against its peers, which viewport is the weak link, the slowest pages, and the issues that recur most.

Performance vs peers

At 73/100, this site is well above the Next.js site (peer median 55).

79th
percentile (est.)
WorstNext.js siteBest
▸Show detailsviewport gap, slowest pages, frequent issues

Slowest pages

  • 1
    /
    perf 40/100
    6.5 s
  • 2
    /about
    perf 63/100
    4.1 s
  • 3
    /climate/legal
    perf 71/100
    3.8 s

Ranked by mobile LCP (Largest Contentful Paint).

Most frequent issues

Performance: 12Best practices: 3Accessibility: 3SEO: 1
  • Issues were logged in the `Issues` panel in Chrome Devtools
    10 pages · 100%
  • Legacy JavaScript
    10 pages · 100%
  • Reduce unused JavaScript
    10 pages · 100%
  • Uses third-party cookies
    10 pages · 100%
  • Use efficient cache lifetimes
    10 pages · 100%
  • Render-blocking requests
    9 pages · 90%
  • Background and foreground colors do not have a sufficient contrast ratio.
    9 pages · 90%
  • Elements with visible text labels do not have matching accessible names.
    9 pages · 90%

Core Web Vitals

How fast your site really is

Real-user experience from the Chrome UX Report (p75, 28-day window). Lab metrics below show cold first-visit timings.

Real users: your Core Web Vitals pass Google's thresholds.

Every measured metric is in the green — re-check after each release.

Real users · page-level
LCP
≤ 2.5 s
1.4 s
FAST
CLS
≤ 0.1
0.000
FAST
INP
≤ 200 ms
79 ms
FAST
Lab · median of scanned pages▸
LCP
≤ 2.5 s
2.9 s
CLS
≤ 0.1
0.000
TBT
≤ 200 ms
651 ms
FCP
≤ 1.8 s
2.3 s
TTFB
≤ 800 ms
49 ms
SI
≤ 3.4 s
2.5 s

Page Weight

Resource breakdown

Pages average 1.5 MB across 166 requests. Third-party scripts add 1.5 MB on top.

  • JavaScript1.2 MB77%
  • Images98 KB6%
  • CSS60 KB4%
  • Fonts61 KB4%
  • Other143 KB9%
▸Show detailsheaviest page

Heaviest page: / at 3.2 MB over 185 requests.

Deep dive

Resource intelligence

Where time and bytes are going across your site — the specific vendors and code paths behind the totals.

Show▸

Top third-party vendors

Site-wide totals
VendorTotal bytesBlockingPages
Google Tag Manager6.6 MB6.1 s10
reddit262 KB245 ms10
Twitter Online Conversion Tracking189 KB154 ms10
Bing Ads123 KB98 ms7
LinkedIn Ads187 KB92 ms8
DemandBase105 KB63 ms5

Main-thread time

Average per page · 2.8 s
  • Script evaluation1.5 s52%
  • Other461 ms16%
  • Script parse / compile408 ms14%
  • Style + layout292 ms10%
  • Garbage collection118 ms4%
  • Paint + composite59 ms2%
  • Parse HTML38 ms1%

Image format breakdown

77% modern (WebP / AVIF)
  • webp762 KB
  • png79 KB
  • gif60 KB
  • svg55 KB
  • plain13 KB
  • javascript9 KB
  • html5 KB

Technology

Detected stack

What we recognised from the homepage HTML, response headers and DNS. Used to tailor stack-specific recommendations.

Show▸
Framework
Next.js

Action Plan

Root-cause groupings, sequenced by effort. Each card opens a ready-to-paste task you can create in Jira or Linear with one click.

Quick wins
Under a day each — start here.
  1. high severity:01Address accessibility gaps⚡ quick win
  2. high severity:02Unblock AI answer-engine crawlers in robots.txt⚡ quick win
  3. medium severity:03Fix crawlability issues
  4. low severity:04Set long cache lifetimes for static assets
Moderate effort
A few days each — book this sprint.
  1. critical severity:01Unblock the initial render
  2. critical severity:02Audit and trim third-party scripts
  3. critical severity:03Right-size and modernize images
  4. medium severity:04Add Schema.org structured data
  5. low severity:05Reduce total page weight
Larger projects
A week+ each — scope and plan.
  1. critical severity:01Cut and defer JavaScript
Fix these with AI

Bundle any of the 10 fixes into one ready-to-paste prompt for Claude Code, Cursor or v0 — it carries the site-specific evidence and exact snippets.

Filter·Sort
01 / 10

Unblock the initial render

Critical·Medium effort·Affects 10 pages·~1.9 s faster
Create ticket →

CSS and JavaScript in the critical path delay first paint — visitors stare at a blank screen, and Largest Contentful Paint (a Google ranking signal) suffers.

Impact — Largest Contentful Paint is a Google ranking signal. Sites in the slowest quartile of their category see 30–50% less organic search traffic than faster peers.

Found on this site~1.9 s faster if fixed
  • ·Render-blocking requests — Est savings of 2,810 ms
Seen on:/contact/pricing//about/legal+5 more
▸ Steps (4)
  1. 1.Inline critical above-the-fold CSS and defer the rest.
  2. 2.Add async or defer to render-blocking scripts.
  3. 3.Preload the LCP image and key fonts so the browser discovers them early.
  4. 4.Remove unused CSS rules from above-the-fold stylesheets.
  5. For your stack
  6. 1.Use next/font so fonts do not block render.
  7. 2.Avoid importing large global CSS; prefer per-component styles.
Paste into Claude Code, Cursor or v0 to apply this fix.
02 / 10

Audit and trim third-party scripts

Critical·Medium effort·Affects 10 pages
Create ticket →

Third-party code — tag managers, analytics, chat widgets, ad pixels — runs on every page and is a leading cause of slow, janky sites. It also changes without warning, outside your control.

Impact — Third-party scripts cause ~57% of JavaScript-related main-thread delays on a typical site (Web Almanac). They also change without warning, outside your control.

Found on this site
  • ·Third-party code is 100% of total page weight — 1.5 MB of 1.5 MB
Seen on:/contact/pricing//about/legal+5 more
▸ Steps (5)
  1. 1.Inventory every third-party script and tie each one to a clear business owner and purpose.
  2. 2.Remove anything not actively used; consolidate overlapping tools.
  3. 3.Load non-critical third parties lazily — after interaction or on idle.
  4. 4.Use tag-manager triggers to fire tags only where needed, not site-wide.
  5. 5.Self-host critical third-party scripts where licensing allows.
Paste into Claude Code, Cursor or v0 to apply this fix.
03 / 10

Right-size and modernize images

Critical·Medium effort·Affects 2 pages·~600 ms faster
Create ticket →

Oversized images waste bandwidth and slow the largest element on the page. On mobile connections this is often the single biggest load-time cost.

Impact — Images account for roughly 25% of an average mobile page (HTTP Archive). Every MB of unoptimized images adds 1–2 seconds of mobile load time.

Found on this site~600 ms faster if fixed
  • ·Improve image delivery — Est savings of 94 KiB
Seen on://about
▸ Steps (4)
  1. 1.Serve images at the dimensions they are displayed, with srcset for different screen sizes.
  2. 2.Use modern formats (WebP/AVIF) with fallbacks.
  3. 3.Lazy-load below-the-fold images.
  4. 4.Compress images — aim well under 200 KB for most photos.
  5. For your stack
  6. 1.Use the next/image component — it handles sizing, modern formats and lazy-loading automatically.
Paste into Claude Code, Cursor or v0 to apply this fix.
04 / 10

Cut and defer JavaScript

Critical·Large effort·Affects 10 pages·~2.4 s faster
Create ticket →

Large scripts block the browser’s main thread, so the page looks loaded but does not respond to taps or clicks. This is the most common cause of a slow mobile experience and directly drives up bounce rate.

Impact — Heavy JavaScript is the #1 driver of slow mobile sites — 53% of mobile users abandon pages that take longer than 3 seconds to load (Google mobile-speed research).

Found on this site~2.4 s faster if fixed
  • ·Reduce JavaScript execution time — 3.7 s
  • ·Minimize main-thread work — 6.7 s
  • ·Reduce unused JavaScript — Est savings of 546 KiB
  • ·Legacy JavaScript — Est savings of 19 KiB
Seen on:/contact/pricing//about/legal+5 more
▸ Steps (4)
  1. 1.Audit the bundle to find the largest dependencies (a bundle analyzer or source-map-explorer).
  2. 2.Code-split so route- and below-the-fold code loads on demand instead of upfront.
  3. 3.Remove or replace heavy third-party scripts; load non-critical ones with async/defer or after first interaction.
  4. 4.Drop legacy polyfills and transpilation targets that modern browsers no longer need.
  5. For your stack
  6. 1.Use next/dynamic to lazy-load heavy components.
  7. 2.Run @next/bundle-analyzer to find oversized chunks.
  8. 3.Set a modern browserslist so Next.js stops shipping legacy transforms.
Paste into Claude Code, Cursor or v0 to apply this fix.
05 / 10

Address accessibility gaps

High·Small effort·Affects 9 pages·⚡ Quick win
Create ticket →

Accessibility issues exclude users of assistive technology and create legal exposure (ADA / European Accessibility Act). Many fixes also help SEO.

Impact — About 15% of users have a disability (WHO). Accessibility gaps both exclude these users and create ADA/EAA compliance risk in the US and EU.

▸ Steps (3)
  1. 1.Give interactive elements and dialogs accessible names (aria-label or visible text).
  2. 2.Ensure every meaningful image has descriptive alt text.
  3. 3.Check color contrast meets WCAG AA.
Paste into Claude Code, Cursor or v0 to apply this fix.
06 / 10

Unblock AI answer-engine crawlers in robots.txt

High·Small effort·Affects 1 page·⚡ Quick win
Create ticket →

These are not training bots — they are the live retrieval crawlers ChatGPT search, Perplexity and Claude use to read a page before citing it in an answer. Blocking them removes the site from AI answers entirely, while training opt-outs (GPTBot, CCBot…) can stay blocked with no citation impact.

Impact — Buyers increasingly ask ChatGPT, Perplexity and Claude for recommendations. A site these crawlers cannot read cannot be cited — the traffic goes to a competitor whose site they can.

Found on this site
  • ·Blocked/restricted: OAI-SearchBot, ChatGPT-User, Claude-User, PerplexityBot, Perplexity-User
Seen on:/robots.txt
Copy-paste fix
User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: Claude-User
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Perplexity-User
Allow: /
▸ Steps (3)
  1. 1.Open robots.txt and find the rules covering the blocked crawlers below.
  2. 2.Allow the search/citation crawlers (keep training-bot blocks if IP protection is the goal — they are separate tokens).
  3. 3.Deploy and verify with the free readiness checker.
Paste into Claude Code, Cursor or v0 to apply this fix.
07 / 10

Fix crawlability issues

Medium·Small effort·Affects 4 pages
Create ticket →

If search engines cannot follow your links or understand their text, pages will not be discovered or rank — directly capping organic traffic.

Impact — Crawlability issues directly cap organic search traffic. Search engines cannot rank what they cannot follow or understand.

Found on this site
  • ·Links do not have descriptive text — 29 links found
Seen on:/pricing//climate/legal/connect/legal
▸ Steps (3)
  1. 1.Use real <a href> elements for navigation, not click handlers on <div>s.
  2. 2.Give links descriptive text instead of “click here” or “read more”.
  3. 3.Ensure important pages are linked from somewhere crawlable.
Paste into Claude Code, Cursor or v0 to apply this fix.
08 / 10

Add Schema.org structured data

Medium·Medium effort·Affects 8 pages
Create ticket →

Schema.org JSON-LD tells search engines and answer engines what each page is — an Organization, an Article, a Product, an FAQ. Pages with structured data earn richer search results and are easier for LLMs to summarize and cite accurately.

Impact — Structured data is a low-risk SEO and AI-visibility win: it doesn’t change your design but materially improves how engines understand and present your pages.

Found on this site
  • ·Present: FAQPage (1), Organization (1), WebSite (1)
Copy-paste fix
<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "stripe.com",
  "url": "https://stripe.com",
  "logo": "https://stripe.com/logo.png",
  "sameAs": [
    "https://www.linkedin.com/company/your-company",
    "https://twitter.com/your-handle"
  ]
}
</script>
▸ Steps (4)
  1. 1.Add an Organization + WebSite schema to your homepage (identity for the whole site).
  2. 2.Add BreadcrumbList to section pages so engines understand your hierarchy.
  3. 3.Add page-type schema where it fits — Article/BlogPosting on posts, FAQPage on Q&A, Product on product pages.
  4. 4.Validate with Google’s Rich Results Test before shipping.
Paste into Claude Code, Cursor or v0 to apply this fix.
09 / 10

Set long cache lifetimes for static assets

Low·Small effort·Affects 10 pages
Create ticket →

Without long cache headers, repeat visitors re-download assets that never changed — slower return visits and wasted bandwidth.

Impact — Returning visitors slow their own experience without caching. Small fix, compounding gain: every repeat visit benefits.

Found on this site
  • ·Use efficient cache lifetimes — Est savings of 86 KiB
Seen on:/contact/pricing//about/legal+5 more
▸ Steps (3)
  1. 1.Set Cache-Control: max-age of up to a year for fingerprinted static assets.
  2. 2.Use content hashing in filenames so you can cache aggressively and still ship updates.
  3. 3.Configure caching at the CDN or host layer rather than per request.
Paste into Claude Code, Cursor or v0 to apply this fix.
10 / 10

Reduce total page weight

Low·Medium effort·Affects 1 page
Create ticket →

Heavy pages cost real money for users on metered mobile data and load slowly on weak connections — a meaningful share of every site’s traffic.

Impact — Every additional megabyte of page weight typically adds 1–2 seconds of mobile load time on a slow connection (HTTP Archive).

Found on this site
  • ·Avoid enormous network payloads — Total size was 3,255 KiB
Seen on:/
▸ Steps (3)
  1. 1.Enable text compression (Brotli or Gzip) for all text assets.
  2. 2.Trim the largest payloads — usually JavaScript and images.
  3. 3.Remove unused third-party embeds and scripts.
Paste into Claude Code, Cursor or v0 to apply this fix.

Shipped a fix? Open the task list or hit ↻ Re-scan in the header — the comparison shows what moved.

Detail

Per-page results

Mobile metrics shown, worst performance first. Click a row for the full per-page deep dive — LCP element, page weight breakdown, top failing checks.

Show▸
PagePerf
/→40
/about→63
/ssa→69
/climate/legal→71
/connect/legal→72
/pricing→77
/legal→77
/contact→82
/connect-account/legal→88
/legal/ssa→89

10 of 10 pages scanned successfully. Each row links to the full per-page report with mobile + desktop side by side.

Coverage

Scan coverage

Pages were discovered via sitemap and sampled across URL patterns. Issues found on a sampled page very likely affect the rest of its pattern group.

Show▸
Discovered
7079
Sampled
10
Scanned OK
10
Failed
0

Methodology

How we measured this

Every page measured with Google PageSpeed Insights (1 run per page per viewport — quick scan). Real-user data, where shown, comes from the Chrome UX Report.

Show▸
  • ▸Lab data is a cold first visit from Google's US infrastructure; real users with warm caches and in other regions will differ.
  • ▸Mobile is emulated on throttled slow-4G; desktop is unthrottled. Only public pages are scanned.
  • ▸Quick scan (1 run/page) — upgrade to a paid scan for median-of-3 accuracy.

Full deep-dive on every check at /methodology.

Share

Embed this badge anywhere

An SVG badge for this scan. It shows this scan’s scores and does not change if you re-scan. Embeds in your README, site footer, or blog post.

Public report URL

https://avocadoscore.io/r/09ebad3c-1f72-4534-9af6-ca29f156d16b

Markdown (for README)

[![AvocadoScore — overall verdict](https://avocadoscore.io/api/scans/09ebad3c-1f72-4534-9af6-ca29f156d16b/badge?metric=verdict)](https://avocadoscore.io/r/09ebad3c-1f72-4534-9af6-ca29f156d16b)

HTML (for site footer)

<a href="https://avocadoscore.io/r/09ebad3c-1f72-4534-9af6-ca29f156d16b"><img src="https://avocadoscore.io/api/scans/09ebad3c-1f72-4534-9af6-ca29f156d16b/badge?metric=verdict" alt="AvocadoScore — overall verdict" /></a>

Next Steps

Where to go from here

  1. 1Start this week. Knock out the 2 quick wins from the Action Plan — small effort, immediate gain.
  2. 2Re-scan after each round of fixes. Hit Re-scan in the header after each round — the comparison shows what moved and what didn’t.

You just fixed it — now keep it fixed.

Scores drift back after every deploy. For $9/mo we re-scan stripe.com monthly and email you the moment a performance, SEO or AI-readiness score slips — so you catch the regression before it costs you traffic. Cancel any time.

This audit was run on AvocadoScore

stripe.com scored 73 on Mobile Perf. What about yours?

Performance, SEO, and how Claude, GPT and Gemini see your site. Free, every fix unlocked. Same report format, your domain.

Scan my site — free →

Or see how your site stacks up against stripe.com:

Or share this report on X