Zyntlox
Evidence-based GEO audit

Can AI answer engines read your website?

Zyntlox checks what your site permits AI crawlers to read and how clearly it describes itself: robots.txt rules, simulated crawler requests, sitemaps and structured data. Every finding comes with its evidence and a way to check it yourself. No AI model writes the results.

https://

Free during the public beta, no account needed. Zyntlox shows what a site permits; it cannot show whether any AI engine cites it.

What Zyntlox checks

Four scored categories, and signals we only report

28 tests move the score, each with a fixed weight and severity. A further 33 are reported for your information and never change it.

30% of the score

Crawler access

Whether search and retrieval crawlers are allowed by robots.txt and actually served when they ask: a simulated request for each one, HTTPS and certificate, redirects, and whether the page is marked noindex.

15 scored tests

35% of the score

Entity graph

Whether the site says clearly who it is in structured data: Organization details, sameAs links to your other profiles, FAQPage markup, and whether the JSON-LD is valid.

5 scored tests

20% of the score

Machine context

How easy the site is to map and read: a valid sitemap that is declared in robots.txt and lists healthy URLs, and a page title, meta description and a single main heading.

7 scored tests

15% of the score

Citability

Whether the page text contains concrete, quotable figures such as percentages, amounts and counts, which are easier to cite than general statements.

1 scored tests

Reported, never scored

33 signals

Blocking training crawlers (GPTBot, ClaudeBot, Google-Extended), /llms.txt, page weight, response time and compression, heading and image-alt structure, and checks across the pages Zyntlox crawls. They are shown with the same evidence as every other test, but they stay out of the score because they are a site owner's choice or have not been confirmed by AI engines.

How it works

Three steps, no setup

From an address to a list of findings you can verify.

01

Enter a public address

Paste the address of any website a visitor can open. Zyntlox refuses private and internal addresses.

02

Zyntlox runs the tests

It reads robots.txt, sends a simulated request as each crawler, and checks the sitemap, the server's responses, structured data and a sample of linked pages (25 by default). No AI model is involved.

03

Review the evidence

Every finding shows what was expected, what was observed and how to check it yourself. Download the findings as CSV or JSON, or export a PDF report.

Example report

What a finished audit looks like

This is an example for a fictional business, northwind-studio.example. It was produced by the same engine and tests as a live audit, from fixed inputs, and it is not a real site. Open any test to see its evidence.

Example report · fictional site

ZGI™ Master Score

67/ 100
Grade: Good

Category Score Distribution (% of each category)

How this score was calculated

ZGI v2.7 · tested 2026-11-01
Coverage
100%
Passed
21
Warnings
3
Failed
2
Not verified
0
CategoryCategory scorePoints earned
Crawler access99%29.6 / 30.0
Entity graph25%8.8 / 35.0
Machine context97%19.4 / 20.0
Citability60%9.0 / 15.0
Total67 / 100

Each test has a fixed weight inside its category; tests that do not apply to this site, or could not be verified, are excluded from the score and shown in coverage instead. Blocking training crawlers (GPTBot, ClaudeBot, Google-Extended) is reported but does not lower the score. Full method: docs/zgi-methodology.md.

AI crawler access by engine

Read from robots.txt
ChatGPT SearchPartially blocked

OAI-SearchBot: allowed · ChatGPT-User: allowed · GPTBot: blocked

! BLOCKER:GPTBot blocked in robots.txt
Perplexity AICrawl allowed

PerplexityBot: allowed

Google AI OverviewsCrawl allowed

Googlebot: allowed. Google-Extended is a separate control token for Gemini training and is not part of this status

Claude (Anthropic)Crawl allowed

Claude-SearchBot: allowed · Claude-User: allowed · ClaudeBot: allowed

Simulated crawler access

2026-11-01

Each row is one request sent to https://northwind-studio.example/ with that crawler's user-agent string, compared with a normal browser request. It shows how the server answers a request that looks like the crawler. It does not prove the real crawler visits, and real crawlers may be treated differently (for example, firewalls that check provider IP ranges).

CrawlerResultHTTPTimerobots.txtEvidence
GPTBotBlocked by robots.txt200360 msblockedrobots.txt disallows this crawler. For comparison the simulated request got HTTP 200. (server: cloudflare)
OAI-SearchBotAccessible200360 msallowedHTTP 200 for the simulated crawler request. (server: cloudflare)
ChatGPT-UserAccessible200360 msallowedHTTP 200 for the simulated crawler request. (server: cloudflare)
ClaudeBotAccessible200360 msallowedHTTP 200 for the simulated crawler request. (server: cloudflare)
Claude-SearchBotAccessible200360 msallowedHTTP 200 for the simulated crawler request. (server: cloudflare)
Claude-UserAccessible200360 msallowedHTTP 200 for the simulated crawler request. (server: cloudflare)
GooglebotAccessible200180 msallowedHTTP 200 for the simulated crawler request. (server: cloudflare)
Google-ExtendedNot tested——allowedThis is a robots.txt control token, not a crawler. It sends no requests, so only its robots.txt permission applies.
BingbotAccessible200180 msallowedHTTP 200 for the simulated crawler request. (server: cloudflare)
PerplexityBotBlocked by server/CDN403180 msallowedThe crawler user agent got HTTP 403 while a normal browser user agent got HTTP 200. Response carried: server: cloudflare. (server: cloudflare)

Test results

52 tests in total: 26 are scored (these match the score breakdown) and 26 are reported but never change the score. Click a row for the full explanation.

52tests
  • Passed: 44
  • Warning: 3
  • Failed: 5
  • Not verified: 0

Emerging signals

Not scored

Conventions that no major AI engine has confirmed it uses yet. They are checked so you can prepare, and they never change the score.

/llms.txt
/llms.txt returned HTTP 404.
Not present

Page weight for AI

Estimate · not scored

The rendered page is about 423 tokens (0.3% of an illustrative 128,000-token context window). The readable text in it is about 234 tokens, 55% of the HTML.

Readable text~234 tokens · 55%
HTML tags and whitespace~149 tokens · 35%
JSON-LD structured data~41 tokens · 10%

Tokens are estimated as characters / 4; real tokenizers differ by model and language. Many AI tools extract the text before the model sees it, so the HTML figure is what an agent reading the raw page would load, not a measured bill. No AI provider documents a size threshold, so this is never scored.

Site crawl

Sample · not scored

Starting from the audited page, Zyntlox followed internal links and tested up to 26 pages without running scripts. Every page it found was tested. Account, login and cart links are left out.

6
URLs found
6
Tested
0
Not tested
5
Answered OK
0
Redirected
1
Not found
0
No usable answer
Page testedTypeHTTPIssues
https://northwind-studio.example/old-servicesOther404missing page

URL-level findings (1)

URLTypeHTTPIssueTestEvidence
https://northwind-studio.example/old-servicesOther404Link leads to a missing page
Informational
GEO-801-INTERNAL-LINKSHTTP 404. Linked from https://northwind-studio.example/services.

Page type is guessed from the URL path alone (for example /blog/, /products/), so a page whose address gives no hint is shown as Other. Redirecting links are recorded, not followed. Pages are fetched without a browser, so links that scripts add are not seen. Tested 2026-11-01T10:00:00.000Z.

Priority fixes

Issues found on this site that lower its ZGI score.

[GEO-120-SIMULATED-ACCESS] Server refuses some crawler user agents (High). PerplexityBot (HTTP 403 vs browser HTTP 200) refused while a browser was served.

[GEO-301-ORG-SCHEMA] No Organization structured data (Medium). No Organization structured data found.

[GEO-204-SITEMAP-URLS] Some sitemap URLs redirect or do not exist (Medium). Sampled 6 of 6 URL(s): 5 OK, 0 redirected, 1 not found.

[GEO-401-STAT-DENSITY] Few quantitative figures in page text (Low). 3 quantitative figure(s) found in the page text (full credit at 5).

[GEO-303-FAQ-SCHEMA] No FAQPage structured data (Low). No FAQPage structured data found.

Detailed findings

9 findings

Every finding states what was observed, what was expected, the evidence, and how to check it yourself. Severity is fixed per test by published rules.

HighServer refuses some crawler user agentsGEO-120-SIMULATED-ACCESS · −0.4 pts
Expected
The server answers crawler user agents the way it answers a browser.
Observed
PerplexityBot (HTTP 403 vs browser HTTP 200) refused while a browser was served.
Evidence
Test
OAI-SearchBot access simulation
Requested URL
https://northwind-studio.example/
HTTP response
HTTP 200 (browser control: HTTP 200)
Result
Accessible
Timestamp
2026-11-01T10:00:00.000Z
Test
ChatGPT-User access simulation
Requested URL
https://northwind-studio.example/
HTTP response
HTTP 200 (browser control: HTTP 200)
Result
Accessible
Timestamp
2026-11-01T10:00:00.000Z
Test
Claude-SearchBot access simulation
Requested URL
https://northwind-studio.example/
HTTP response
HTTP 200 (browser control: HTTP 200)
Result
Accessible
Timestamp
2026-11-01T10:00:00.000Z
Test
Claude-User access simulation
Requested URL
https://northwind-studio.example/
HTTP response
HTTP 200 (browser control: HTTP 200)
Result
Accessible
Timestamp
2026-11-01T10:00:00.000Z
Test
Googlebot access simulation
Requested URL
https://northwind-studio.example/
HTTP response
HTTP 200 (browser control: HTTP 200)
Result
Accessible
Timestamp
2026-11-01T10:00:00.000Z
Test
Bingbot access simulation
Requested URL
https://northwind-studio.example/
HTTP response
HTTP 200 (browser control: HTTP 200)
Result
Accessible
Timestamp
2026-11-01T10:00:00.000Z
Test
PerplexityBot access simulation
Requested URL
https://northwind-studio.example/
HTTP response
HTTP 403 (browser control: HTTP 200)
Result
Blocked by server/CDN
Timestamp
2026-11-01T10:00:00.000Z
What was tested
The page was requested once per scored crawler user-agent string and once with a normal browser user agent, and the answers were compared.
Why it matters
A site can allow a crawler in robots.txt and still refuse it at the server, CDN or firewall (commonly with HTTP 403 or 429).
What this result means
Some crawler user agents were refused while a browser was served.
GEO implications
If a CDN refuses AI and search crawler user agents, those engines may be unable to fetch the site even though robots.txt allows them.
What to verify
That the server returns the same status to crawler user agents as to browsers, and that any bot-protection rule lets legitimate crawlers through (providers publish IP ranges for this).
How to verify
Run `curl -I -A "<crawler user-agent>" https://northwind-studio.example/` and compare the status with `curl -I https://northwind-studio.example/`. Check CDN or WAF bot-protection settings and logs.
Limits of this test
This is a simulation using the crawlers' user-agent strings from our servers. Real crawlers come from published IP ranges that a firewall may treat differently, so it does not prove what the real crawler experiences.
Affected URL: https://northwind-studio.example/
MediumNo Organization structured dataGEO-301-ORG-SCHEMA · −21.9 pts
Expected
The page declares the organization behind the site in JSON-LD.
Observed
No Organization structured data found.
Evidence
Test
Organization structured data
Requested URL
https://northwind-studio.example/
Result
No Organization structured data found.
Timestamp
2026-11-01T10:00:02.580Z
What was tested
The page's JSON-LD was searched for an Organization or Corporation entity.
Why it matters
Structured data lets a machine read who is behind a site without inferring it from prose.
What this result means
No Organization or Corporation entity was found in the page's JSON-LD.
GEO implications
An explicit Organization entity gives search and answer systems a declared identity to associate with the site. This test checks presence only; no engine guarantees that markup leads to citations or rankings.
What to verify
That valid JSON-LD with an Organization (or Corporation) type is present in the page HTML as served.
How to verify
Paste the URL into Google's Rich Results Test or validator.schema.org and look for an Organization item; or search the page source for `"@type": "Organization"`.
Limits of this test
The page is read without running scripts in the command-line tool; markup injected by JavaScript may be missed there. A match is a text match on the JSON-LD, not a full schema validation.
Affected URL: https://northwind-studio.example/
MediumSome sitemap URLs redirect or do not existGEO-204-SITEMAP-URLS · −0.6 pts
Expected
Sitemap URLs return HTTP 2xx directly (no redirect, no 404/410).
Observed
Sampled 6 of 6 URL(s): 5 OK, 0 redirected, 1 not found.
Evidence
Test
Sitemap URL request (broken)
Requested URL
https://northwind-studio.example/old-services
HTTP response
HTTP 404
Result
broken
Timestamp
2026-11-01T10:00:00.000Z
What was tested
Up to 10 evenly spaced URLs from the sitemap on the site's own host were requested without following redirects.
Why it matters
A sitemap should list the final, working address of each page. Redirecting or missing URLs waste crawl effort and signal an outdated sitemap.
What this result means
Some sampled URLs redirected or returned 404/410.
GEO implications
Crawlers that read the sitemap may spend requests on redirected or missing URLs instead of live pages. The effect on AI visibility is indirect and not quantified by any provider.
What to verify
That each URL in the sitemap returns HTTP 200 directly, and that redirected URLs are replaced by their destination.
How to verify
Run `curl -sI <url>` on a listed URL and check for HTTP 200 without a Location header.
Limits of this test
Only a sample is checked, not every URL. 403, 429, 5xx and timeouts are shown as unclear and are not scored, because they do not prove a URL is broken.
Affected URL: https://northwind-studio.example/sitemap.xml
LowFew quantitative figures in page textGEO-401-STAT-DENSITY · −6 pts
Expected
The page contains concrete figures that can be quoted.
Observed
3 quantitative figure(s) found in the page text (full credit at 5).
Evidence
Test
Quantitative figures in page text
Requested URL
https://northwind-studio.example/
Result
3 quantitative figure(s) found in the page text (full credit at 5).
Timestamp
2026-11-01T10:00:02.580Z
What was tested
The page text was scanned for percentages, dollar amounts and numbers with thousands separators.
Why it matters
Concrete figures are easy to quote and to check, so answer engines can attribute them.
What this result means
Some quantitative figures were found, fewer than the 5 that earn full credit.
GEO implications
Pages with specific, sourced figures may be easier to cite than pages with only general claims. This test is a rough proxy: it counts figures, not their accuracy or originality.
What to verify
That the page states specific, accurate figures where they are relevant.
How to verify
Read the page text and count the quantitative claims; the audit counts matches of %, $ amounts and comma-separated numbers.
Limits of this test
It counts patterns, not meaning, and does not measure information gain. Do not add numbers just to raise the score.
Affected URL: https://northwind-studio.example/
LowNo FAQPage structured dataGEO-303-FAQ-SCHEMA · −4.4 pts
Expected
Question-and-answer content is marked up when the page has any.
Observed
No FAQPage structured data found.
Evidence
Test
FAQPage structured data
Requested URL
https://northwind-studio.example/
Result
No FAQPage structured data found.
Timestamp
2026-11-01T10:00:02.580Z
What was tested
The JSON-LD was searched for FAQPage markup.
Why it matters
FAQPage markup states questions and answers in machine-readable form.
What this result means
No FAQPage markup was found.
GEO implications
It can make answerable content easier to extract. Google has limited where FAQ rich results are shown, so treat it as optional.
What to verify
That pages with real question-and-answer content mark them up as FAQPage.
How to verify
Use Google's Rich Results Test or validator.schema.org on the page.
Limits of this test
A page without FAQs is not defective; this test has a low weight for that reason.
Affected URL: https://northwind-studio.example/
InformationalGPTBot (training) is disallowed in robots.txtGEO-101-GPTBOT · not scored
Expected
GPTBot is permitted to fetch "/".
Observed
Disallow: / (line 6, under User-agent: GPTBot)
Evidence
Test
robots.txt evaluation for GPTBot
Requested URL
https://northwind-studio.example/robots.txt
HTTP response
HTTP 200
Matched rule
Disallow: / (line 6, under User-agent: GPTBot)
Result
fail
Timestamp
2026-11-01T10:00:02.580Z
Test
GPTBot access simulation
Requested URL
https://northwind-studio.example/
HTTP response
HTTP 200 (browser control: HTTP 200)
Result
Blocked by robots.txt: robots.txt disallows this crawler. For comparison the simulated request got HTTP 200.
Timestamp
2026-11-01T10:00:00.000Z
What was tested
The site's robots.txt was parsed per RFC 9309 and evaluated for the GPTBot user-agent token against the site root (/).
Why it matters
GPTBot is OpenAI's crawler for content that may be used to train its models.
What this result means
robots.txt disallows GPTBot from fetching "/". A crawler that honours robots.txt will not fetch the site's pages.
GEO implications
Blocking GPTBot opts out of training use. It is a policy choice and does not affect eligibility for ChatGPT search, which is governed by OAI-SearchBot.
What to verify
That the robots.txt group applying to GPTBot does not disallow "/".
How to verify
Run `curl -s https://northwind-studio.example/robots.txt` and find the group whose User-agent line matches the token (or `*` if none does). Check that no Disallow rule covers `/`, remembering that the longest matching rule wins and Allow wins a tie (RFC 9309).
Limits of this test
robots.txt is a request that well-behaved crawlers honour; it is not an access control, and it does not show whether the crawler visits the site.
Affected URL: https://northwind-studio.example/robots.txt
Informational/llms.txt is not publishedGEO-201-LLMS-TXT · not scored
Expected
A /llms.txt file is published.
Observed
/llms.txt returned HTTP 404.
Evidence
Test
/llms.txt
Requested URL
https://northwind-studio.example/llms.txt
HTTP response
HTTP 404
Result
/llms.txt returned HTTP 404.
Timestamp
2026-11-01T10:00:02.580Z
What was tested
A request for /llms.txt at the site root.
Why it matters
llms.txt is a proposed convention for giving language models a concise, curated description of a site.
What this result means
/llms.txt did not return HTTP 200.
GEO implications
This check is informational and does not change the score. No major AI engine (OpenAI, Google, Anthropic, Perplexity) has confirmed that it reads llms.txt, so its effect on visibility is unproven. It is reported because it is a published convention for machine-readable context that may matter in future.
What to verify
That https://northwind-studio.example/llms.txt returns HTTP 200 with useful content.
How to verify
Run `curl -I https://northwind-studio.example/llms.txt` and check for HTTP 200.
Limits of this test
Only the file's existence is tested, not its content or whether any engine reads it.
Affected URL: https://northwind-studio.example/llms.txt
InformationalMissing pages do not return a real 404GEO-510-404-HANDLING · not scored
Expected
A page that does not exist answers 404 or 410.
Observed
A made-up path answered HTTP 200 instead of 404. Google calls this a soft 404.
Evidence
Test
Missing pages return a real 404
Requested URL
https://northwind-studio.example/
Result
A made-up path answered HTTP 200 instead of 404. Google calls this a soft 404.
Timestamp
2026-11-01T10:00:02.580Z
What was tested
One request was sent to a made-up path on the site's own origin, and the HTTP status was read without following redirects.
Why it matters
Google documents that a missing page which answers 200, or redirects to another page such as the homepage, is treated as a soft 404.
What this result means
A page that does not exist did not return a real 404. This check is informational and does not change the score.
GEO implications
Soft 404s make it harder for crawlers to tell which URLs really exist. Google documents this for its own search; other providers' handling is not documented here. This check is informational and does not change the score.
What to verify
That pages which do not exist return 404 or 410, and that only real pages return 200.
How to verify
Run `curl -sI https://northwind-studio.example/some-page-that-does-not-exist/` and read the first line.
Limits of this test
One made-up path was tested, and redirects were not followed. A firewall-type answer (401, 403, 429, 5xx) is shown as inconclusive rather than counted as a result; a probe that did not complete is not tested.
Affected URL: https://northwind-studio.example/
InformationalSome internal links lead to missing pagesGEO-801-INTERNAL-LINKS · not scored
Expected
Internal links lead to pages that exist.
Observed
1 of 5 internal link target(s) answered not found (404 or 410): https://northwind-studio.example/old-services.
Evidence
Test
Internal link target
Requested URL
https://northwind-studio.example/old-services
HTTP response
HTTP 404
Result
linked from https://northwind-studio.example/services
Timestamp
2026-11-01T10:00:00.000Z
What was tested
The internal links of the page were followed, and the pages they lead to (up to 25) were requested. Targets that answered HTTP 404 or 410 were counted.
Why it matters
A link that leads to a missing page sends readers and crawlers to a dead end.
What this result means
At least one internal link leads to a page that does not exist.
GEO implications
This check is informational and does not change the score. Google's link guidance says links should lead to working pages. No major AI provider documents it as a factor, so its effect on AI visibility is unknown.
What to verify
That each listed URL is meant to be gone, or that the link to it is fixed or removed.
How to verify
Open each listed URL in a browser, then find the page that links to it (the evidence shows which one).
Limits of this test
Only the pages reached by a bounded crawl (up to 25 pages, 12 seconds) are checked, without running scripts, so links that scripts add are not seen. A refusal (HTTP 403 or 429), a server error or a timeout is not counted as broken, because it may be our request that was refused.
Affected URL: https://northwind-studio.example/old-services

How to read this report

  • It shows access, not outcomes. Zyntlox reports what a site permits and what a simulated request receives. It does not show that ChatGPT, Claude, Perplexity, Gemini or any other AI engine visits, indexes or cites the site.
  • You can check every finding. Each one lists what was expected, what was observed and the evidence. No AI model writes a score or a finding.
  • Unverified tests are not counted as passed. A test that could not be run is left out of the score and lowers the coverage figure instead.
  • Some limits are our choices. Thresholds such as word counts or how many figures earn full credit are judgement calls, not facts about any AI engine.

ZGI v2.7 · tested 2026-11-01 · Read the full method and limits

How we keep it honest

Four rules every report follows

Zyntlox reports what a site permits and what a simulated request receives. It does not claim to know whether any AI engine will visit, index or cite you.

01

No AI model in the audit

Every result is a fact recorded by a test, and every explanation is written documentation. The score comes from fixed arithmetic, so the same observations always give the same score.

02

Evidence for every finding

Each finding shows what was expected, what was observed, the requested URL, the HTTP response, the rule that matched and a timestamp, plus how to check it yourself and the limits of the test.

03

Unverified is never “passed”

A test that could not be run is left out of the score and shown as lower coverage. Nothing is counted as passed because it could not be checked.

04

New conventions stay out of the score

Signals such as /llms.txt are checked and shown, but never scored, because no major AI engine has confirmed that it uses them.

Read the method, the weights and the limits →

Questions

Frequently asked questions

Plain answers about what Zyntlox checks, what it cannot tell you, and what happens to your data.

What does Zyntlox check?

Whether a website can be reached and understood by AI search and answer engines. That covers robots.txt rules for named crawlers, a simulated request as each crawler, HTTPS and redirects, the sitemap, structured data (Organization, sameAs, FAQPage), basic page structure and the quantitative figures in the text. The Method page lists every test with its weight and severity.

Does it show whether ChatGPT, Perplexity or Claude cite my site?

No, and it does not claim to. Zyntlox shows what your site permits and what a simulated request receives. Whether an AI engine visits, indexes or cites a site is decided by that engine, and Zyntlox cannot observe it.

Is an AI model involved in the audit?

No. Every result is a fact recorded by a test, and every explanation is written documentation. The score is fixed arithmetic over the test results, so the same observations always give the same score.

How is the score calculated?

The Zyntlox Generative Index (ZGI) runs from 0 to 100. Four categories carry fixed weights: Crawler access 30, Entity graph 35, Machine context 20 and Citability 15. Tests that could not be verified are left out of the score and shown as lower coverage, never counted as passed. Every report states the formula version it used.

What about /llms.txt and other new conventions?

They are checked and shown in a separate Emerging signals section, but never scored, because no major AI engine has confirmed that it uses them. If that changes and an engine documents it, the weight can be raised and the formula version updated.

How does Zyntlox scan a website?

It sends ordinary page requests. Its own checks identify themselves as ZyntloxAudit (+https://zyntlox.com). For the simulated-access test it also sends one request per crawler using that crawler's published user-agent string, so you can see how your server answers. Requests are rate limited and private addresses are refused. If you would like us to stop scanning a site you run, write to hello@zyntlox.com.

Do you store my audits?

An audit run without an account is not saved: the result goes to your browser. If you sign in, your reports are saved to your account history so you can reopen them. Server logs keep the audited domain, the score and timings, and rate limiting uses a hashed form of your IP address. The Privacy page has the details.

Do I need an account?

No. You can run an audit or a comparison without signing in during the public beta. Runs are limited per hour to keep the service available for everyone.

Can I compare my site with a competitor's?

Yes. The Compare page runs the same tests on both sites and shows, test by test, where each one stands. It compares what each site permits and how clearly it describes itself, and says nothing about how either one ranks or is cited.

Can I export the results for a client?

Yes. You can download the findings as CSV, the complete result as JSON, or a PDF report with your agency name and logo. Every file carries the formula version and the time the tests ran.