I scan websites for a living. 42% of domains in my most recent batch returned zero valid JSON-LD. 82% served no llms.txt file. Finance, healthcare, government, e-commerce. The regulated industries.
Every retrieval pipeline assumes the source delivers structured data. I stopped assuming and started measuring. Five signals, one binary result, a CSV for batch runs. Below is the tool and every line of reasoning behind it.
Five signals determine whether a domain provides the minimum infrastructure an automated system can rely on:
application/ld+json
blocks? Which Schema.org types?llms.txt
file at the root?Content-Type
and security headers?A domain passes if at least 3 of 5 checks succeed. The threshold is configurable in the code.
Python 3.8+ and the requests
library:
pip install requests
That is all. Everything runs locally, with zero external dependencies beyond requests
.
The homepage gets one GET request. Every <script type="application/ld+json">
block goes through json.loads
. If the JSON parses and carries an @type
, the check passes.
def check_json_ld(domain: str, session: requests.Session) -> dict:
"""Fetch the homepage and look for valid JSON-LD blocks."""
url = f"https://{domain}"
result = {
"signal": "json_ld",
"found": False,
"valid": False,
"types": [],
"error": None,
}
try:
resp = session.get(url, timeout=TIMEOUT, allow_redirects=True)
resp.raise_for_status()
except requests.RequestException as e:
result["error"] = str(e)[:120]
return result
pattern = re.compile(
r'<script[^>]+type=["\']application/ld\+json["\'][^>]*>'
r'(.*?)</script>',
re.DOTALL | re.IGNORECASE,
)
matches = pattern.findall(resp.text)
if not matches:
return result
result["found"] = True
types_found = []
for raw in matches:
try:
data = json.loads(raw.strip())
items = data if isinstance(data, list) else [data]
for item in items:
t = item.get("@type", "")
if t:
types_found.append(
t if isinstance(t, str) else str(t)
)
except (json.JSONDecodeError, AttributeError):
continue
if types_found:
result["valid"] = True
result["types"] = types_found
return result
168 domains had a JSON-LD block in the HTML. The block existed. It contained empty types, broken nesting, truncated strings. A malformed block is the more dangerous case: your pipeline sees data, tries to parse it, and extracts garbage.
The llms.txt
standard (proposed by Jeremy Howard) gives language models a machine-readable summary of a website. Adoption sits at 18% across 559 scanned domains in my most recent batch run.
def check_llms_txt(domain: str, session: requests.Session) -> dict:
"""Check whether the domain serves an llms.txt file at the root."""
url = f"https://{domain}/llms.txt"
result = {
"signal": "llms_txt",
"found": False,
"size_bytes": 0,
"error": None,
}
try:
resp = session.get(url, timeout=TIMEOUT, allow_redirects=True)
if (
resp.status_code == 200
and "text" in resp.headers.get("content-type", "")
):
body = resp.text.strip()
if len(body) > 20 and not body.startswith("<!"):
result["found"] = True
result["size_bytes"] = len(body.encode("utf-8"))
except requests.RequestException as e:
result["error"] = str(e)[:120]
return result
Two guard clauses matter here. First, the content-type check filters out HTML error pages served with a 200 status. Second, body.startswith("<!")
catches soft-404 pages that return an HTML document instead of a plain-text file.
Two questions matter here. Can a generic bot crawl the site at all? And does the site explicitly shut the door on known AI crawlers like GPTBot, CCBot, ClaudeBot, or PerplexityBot?
KNOWN_AI_BOTS = [
"gptbot", "chatgpt-user", "claudebot", "anthropic",
"google-extended", "ccbot", "bytespider", "cohere-ai",
"perplexitybot", "amazonbot",
]
def check_robots_txt(domain: str, session: requests.Session) -> dict:
"""Parse robots.txt for bot-related directives."""
url = f"https://{domain}/robots.txt"
result = {
"signal": "robots_txt",
"found": False,
"allows_generic_bots": True,
"blocks_ai_bots": False,
"ai_bot_rules": [],
"error": None,
}
try:
resp = session.get(url, timeout=TIMEOUT, allow_redirects=True)
if resp.status_code != 200:
return result
if "text" not in resp.headers.get("content-type", ""):
return result
except requests.RequestException as e:
result["error"] = str(e)[:120]
return result
result["found"] = True
lines = resp.text.lower().splitlines()
current_agent = None
for line in lines:
line = line.split("#")[0].strip()
if line.startswith("user-agent:"):
current_agent = line.split(":", 1)[1].strip()
elif line.startswith("disallow:") and current_agent:
path = line.split(":", 1)[1].strip()
if path == "/" and current_agent == "*":
result["allows_generic_bots"] = False
if path == "/" and current_agent in KNOWN_AI_BOTS:
result["blocks_ai_bots"] = True
result["ai_bot_rules"].append(current_agent)
return result
KNOWN_AI_BOTS
will grow. I update the list every time a new crawler shows up in my server logs.
A domain that ships Strict-Transport-Security
and X-Content-Type-Options
runs a maintained stack. I track 10 sector lists. The pattern repeats in every single one: domains that fail on structured data also fail on security headers.
SECURITY_HEADERS = [
"strict-transport-security",
"x-content-type-options",
"x-frame-options",
"content-security-policy",
]
def check_headers(domain: str, session: requests.Session) -> dict:
url = f"https://{domain}"
result = {
"signal": "headers",
"status_code": None,
"has_security_headers": False,
"content_type_valid": False,
"server": None,
"error": None,
}
try:
resp = session.head(url, timeout=TIMEOUT, allow_redirects=True)
result["status_code"] = resp.status_code
ct = resp.headers.get("content-type", "")
result["content_type_valid"] = "text/html" in ct
present = sum(
1 for h in SECURITY_HEADERS if h in resp.headers
)
result["has_security_headers"] = present >= 2
result["server"] = resp.headers.get("server", "")[:60]
except requests.RequestException as e:
result["error"] = str(e)[:120]
return result
The shortest function in the file. If the HTTPS handshake breaks, everything else is academic.
def check_ssl(domain: str, session: requests.Session) -> dict:
url = f"https://{domain}"
result = {"signal": "ssl", "valid": False, "error": None}
try:
resp = session.head(url, timeout=TIMEOUT, allow_redirects=True)
result["valid"] = True
except requests.exceptions.SSLError as e:
result["error"] = f"SSL error: {str(e)[:100]}"
except requests.RequestException as e:
result["error"] = str(e)[:100]
return result
A domain passes when at least 3 of 5 checks succeed. The threshold lives in a constant at the top of the script. Adjust it based on your pipeline's tolerance.
PASS_THRESHOLD = 3
def evaluate_domain(domain: str) -> dict:
session = requests.Session()
session.headers.update({"User-Agent": USER_AGENT})
checks = {
"json_ld": check_json_ld(domain, session),
"llms_txt": check_llms_txt(domain, session),
"robots_txt": check_robots_txt(domain, session),
"headers": check_headers(domain, session),
"ssl": check_ssl(domain, session),
}
passing = 0
if checks["json_ld"]["found"] and checks["json_ld"]["valid"]:
passing += 1
if checks["llms_txt"]["found"]:
passing += 1
if (
checks["robots_txt"]["found"]
and checks["robots_txt"]["allows_generic_bots"]
):
passing += 1
if (
checks["headers"]["content_type_valid"]
and checks["headers"]["has_security_headers"]
):
passing += 1
if checks["ssl"]["valid"]:
passing += 1
verdict = "PASS" if passing >= PASS_THRESHOLD else "FAIL"
return {
"domain": domain,
"verdict": verdict,
"passing": passing,
"total": 5,
"checks": checks,
}
python infra-check.py example.com
python infra-check.py example.com another.com third.com
Create a domains.txt
with one domain per line:
example.com
another.com
third.com
Then:
python infra-check.py --file domains.txt --csv results.csv
python infra-check.py --file domains.txt --json-out results.json
python infra-check.py --file domains.txt -q --csv results.csv
Prints only the verdict line per domain. Useful for CI pipelines.
Checking 3 domain(s)...
❌ example-bank.de [FAIL] (2/5 checks passed)
JSON-LD: missing
llms.txt: missing
robots.txt: open
Headers: HTTP 200, security headers: yes
SSL/TLS: valid
✅ example-saas.com [PASS] (4/5 checks passed)
JSON-LD: valid (Organization, WebSite)
llms.txt: found (2340 bytes)
robots.txt: open, blocks 2 AI bots
Headers: HTTP 200, security headers: yes
SSL/TLS: valid
❌ example-gov.de [FAIL] (1/5 checks passed)
JSON-LD: missing
llms.txt: missing
robots.txt: blocks all bots
Headers: HTTP 200, security headers: incomplete
SSL/TLS: valid
Total: 3 | Passed: 1 | Failed: 2
Five binary signals. One question: can an automated system extract structured data from this source?
This is a pre-flight check. My production scanner (SOVP) runs 180+ signals across 21 audit clusters and produces cryptographically signed attestations. This tool extracts 5 of those signals into a standalone script you can run without an account, an API key, or my infrastructure.
Content quality, factual accuracy, and freshness sit on a higher layer. This layer sits beneath all of them. If a domain fails here, your pipeline will burn tokens on extraction and get noise in return.
The user-agent string identifies itself honestly. The script waits 1 second between domains, follows redirects, and respects a 15-second timeout.
The complete infra-check.py
with CLI argument parsing, CSV/JSON export, and deduplication:
infra-check is a single-file command-line tool that inspects any domain for machine-readable infrastructure signals — the kind of things that determine how well a site can be understood by crawlers, AI agents, and automated clients It runs five checks per domain and produces a binary PASS / FAIL verdict with optional CSV or JSON export for batch runs.
| # | Check | What it looks for |
|---|---|---|
| 1 | JSON-LD | |
Presence of valid application/ld+json structured data on the homepage, and the @type s it declares. |
||
| 2 | llms.txt | |
Whether the domain serves a non-trivial /llms.txt file at the root. |
||
| 3 | robots.txt | |
Whether /robots.txt exists, whether generic bots are allowed, and which known AI crawlers (GPTBot, ClaudeBot, Google-Extended, CCBot, …) are blocked. |
||
| 4 | Headers | |
HTTP status, a valid text/html content type, and the presence of common security headers (HSTS, X-Content-Type-Options, X-Frame-Options, CSP). |
||
| 5 | SSL/TLS | |
| That the domain responds over HTTPS without certificate |
You feed domains into your pipeline. How many of them pass all five checks?
Run the script. Look at the CSV. That number is the actual size of your usable source list.
If you run it, drop your numbers in the comments. I am genuinely curious how other people's source lists hold up.
Thorsten Litzki is the founder of Litzki Systems LLC and builds a cryptographically signed infrastructure verification engine. He scans regulated websites for a living and writes about what the data reveals.