SEOSite Audit
Site Audit
anthropic.com · 31 pages crawled · 8 transparent sub-scores · live technical signals
SEO suite · module 2 of 7
01

The verdict — can the engines crawl, render & trust this site

A whole-site technical audit across 31 crawled pages, scored on 8 measured sub-scores that sum, with their weights, to a single number — no black box. The headline: Site Health 70 / 100, MODERATE. The foundations an AI crawler needs are excellent — 100% server-rendered, every bot allowed — but the page is bleeding on speed, structured data and security headers. Every issue below names the exact pages and ships the fix.

MODERATEwhole-crawl health

A health score is only honest if you can see the math. Every sub-score below shows its weight, its value, and the one-line reason — and every issue points at the pages it lives on.

The math — Site Health 70 = the weighted sum of 8 measured sub-scores

no black box · tap any sub-score
How thorough this is vs a one-page checker

02

Issues by severity — what to fix, and where it lives

critical and warnings across the crawl, grouped by type and sorted by severity. Each card names the affected pages — this isn't "improve your site," it's "fix these exact URLs." pages are already perfectly clean.

03

Crawlability & discovery — can every bot find everything

The front door. Before anything can rank or be cited, the crawlers — search and AI — have to be let in and pointed at the content.

AI & search crawler access

all allowed

robots.txt is a permissive User-Agent: * / Allow: / — every crawler below is allowed. The nuance: there are no bot-specific rules at all, so AI access rides entirely on the wildcard. That works today; it's worth making the AI allow-list explicit.

Discovery files

llms.txt today
404 — not found. Claude & other answer-engines look here first for an AI-readable map.
once added
A clean /llms.txt index pointing AI crawlers straight at your canonical content (the fix is in §12).
04

AI rendering — what the bots actually see

Server-side rendering — your strongest plane

SSR 100%

All 31 crawled pages serve real HTML before JavaScript runs. This matters more than anything else on this page: GPTBot, ClaudeBot and PerplexityBot do not execute JavaScript — a client-rendered page returns an empty <div id="root"> and is invisible to them. Anthropic clears this gate cleanly, which is why it gets cited at all. Protect it: every framework change should be re-checked for SSR.

05

Indexability — clean signals to the index

Crawlable isn't the same as cleanly indexable. Three signal-dilution problems repeat across the crawl.

06

Architecture — where the link equity flows

Internal links are how authority moves through a site. The structure is sound at the top — zero orphans, strong hubs — but the defining content is starved.

Top hubs — equity sources

inbound links, 31-page sample

The under-linked — starved pages

equity gap

Your flagship safety & research writing — the Responsible Scaling Policy, Core Views on AI Safety, the Anthropic Institute — sits at 1–3 inbound links. The exact content that should define the entity is the hardest for crawlers to reach.

07

Speed & Core Web Vitals — site-wide

The most measurable ranking signals Google ships — and there are two truths most tools only half-show: the field data is what Google actually ranks on (real users), the lab data is the diagnostic that tells you where to optimize. We show both, and recommend the fix for each.

Core Web Vitals — field data

real users · 28-day CrUX · the ranking signal

Supporting metrics

diagnostic — not directly ranked

The lab diagnostic — where the headroom is

Page weight by type

Optimization opportunities — lab diagnostics

with the recommended fix

The three Core Web Vitals — thresholds, causes & fixes

your value vs the bands

Per-page response & weight — all 31 crawled pages

Per-page, the crawl gives two continuous speed signals — server response (TTFB) and HTML weight; the full Core Web Vitals are origin-level (the cards above) plus a desktop-lab pass on 3 templates. Thresholds — TTFB ≤800ms good / ≤1.8s / >1.8s poor. Field TTFB (real users) is a healthy 752 ms; these crawl numbers are cold render waits behind the CDN cache. Click any page for its full vitals — field, lab, crawl signals and issues.

Why speed now feeds AI citation directional

Across monitored reference sites in early 2026, pages that pass Core Web Vitals show a 3–4× higher AI-citation rate than failing pages in the same topic. Not a published ranking factor — but the mechanism is plausible: fast, well-structured pages are more reliably crawled, parsed and trusted by AI answer engines. Speed is no longer just a Google-organic play; it's table stakes for being quotable.

08

Structured data — the machine-readable layer

The lowest sub-score, and the highest-leverage fix on the site — because it's the one change that moves three planes at once (technical, on-page E-E-A-T, and entity/GEO).

Schema coverage 29% — and it's the wrong schema

34 / 100

22 of 31 pages carry no structured data at all. The 9 that do carry only FAQPage markup — there is no Organization, Article, Product, Person or Breadcrumb schema anywhere on the domain. That's the entity-proof an answer-engine needs, missing.

what a crawler extracts now
An unlabelled HTML page. The machine has to guess what Anthropic is, who wrote this, and how pages relate.
with Organization + Article JSON-LD
Declared facts — the org, its sameAs identity graph, authorship, and breadcrumbs — extractable verbatim. Drop-in code in §12.
09

Security & technology stack

Transport and email authentication are strong; the gap is a set of missing HTTP response headers — cheap, edge-level one-liners that also signal engineering rigor.

Transport & response headers

74 / 100

DNS & email authentication

strong

Served behind Cloudflare with HTTP/3 and Brotli. Email is DMARC p=reject (strongest) — but SPF ends in a soft ~all; tighten to -all.

10

Defensive scan — AI prompt-injection (IDPI)

Indirect prompt-injection / AI-poisoning scan

31 pages clean

As AI crawlers read your pages, hidden text can hijack them. We scanned all 31 pages for the December-2025 Unit-42 payload patterns — zero-size text (font-size:0), off-screen positioning (left:-9999px), transparency tricks (opacity:0, color-on-background), and HashJack URL-fragment payloads. All clean. No rival SEO audit even looks for this — it's the defensive half of being AI-readable.

11

Per-page crawl health — every page, worst first

The whole crawl, ranked by issue count. Click any page for its full technical breakdown and the fixes it needs. 2 pages (/product/claude-code, /product/claude-cowork) are flawless — proof the platform can ship clean pages.

Page Issues Status TTFB Weight Schema Canon H1 Skips
12

Deployable fixes — drop-in code, impact-ordered

The repair, shipped. These are edge-level and template-level changes — most ship in one deploy.

① Security response headers — Cloudflare / edge

Headers HSTS 1h · no X-Frame / nosniff / Referrer → full set, HSTS 1yr + preload74 → ~95
_headers (Cloudflare Pages) / equivalent edge rulecopy
# lock down the response — all currently missing or weak
/*
  Strict-Transport-Security: max-age=31536000; includeSubDomains; preload
  X-Frame-Options: SAMEORIGIN
  X-Content-Type-Options: nosniff
  Referrer-Policy: strict-origin-when-cross-origin
  Permissions-Policy: geolocation=(), microphone=(), camera=()

② Self-referential canonical — 12 pages incl. homepage

Canonical missing on 12 pages → self-referential on every templateindexability 61 → ~85
every template · <head>copy
<!-- emit the page's own absolute URL; also de-dupes -->
<!-- /responsible-scaling-policy, currently crawled via two URL forms -->
<link rel="canonical" href="https://www.anthropic.com/responsible-scaling-policy">

③ llms.txt — the AI-readable site map (currently 404)

Discovery /llms.txt → 404 → AI crawlers get a clean mapcrawlability 90 → ~96
/llms.txt (site root)copy
# Anthropic — AI-readable index
> Anthropic builds Claude, a family of frontier AI models.

## Core
- [Research](https://www.anthropic.com/research): safety & interpretability
- [Claude models](https://www.anthropic.com/claude): Opus, Sonnet, Haiku
- [Responsible Scaling Policy](https://www.anthropic.com/responsible-scaling-policy)

Fix priority

impact-ordered

Ordered by health-score lift × ease. The structured-data block (§08) is listed first because it's the only fix that also moves the On-Page and GEO planes — one deploy, three scores.

Bliss Site Audit · anthropic.com · 100% real — transparent sub-scores from a live crawl + headers/DNS + lab CWV; owner-only gaps (logs, full index) excluded, never faked.