SEO Audit + AI Visibility — is your site citable by ChatGPT?

A bulk SEO audit and AI visibility checker in one run: it crawls your site, checks every page for technical SEO issues, verifies every link, and answers the question no other audit answers — can AI systems reach and cite you? It reads your robots.txt the way each AI crawler does, diffs every run against your previous one, and hands you ready-to-paste fixes. Built for bulk, schedules, API, Make, Zapier and MCP.

The question nobody else's audit answers: are you citable?

Your readers increasingly get your answer from ChatGPT, Claude or Perplexity instead of from a blue link. Whether you appear there is decided by your robots.txt — and by a distinction almost nobody gets right:

Crawler type	Examples	Blocking it means
Training	GPTBot, ClaudeBot, Google-Extended, Applebot-Extended, CCBot	Your content is not used to train models
Retrieval	OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot	You can never be cited in an AI answer

Most sites that "blocked the AI bots" used one blanket rule and quietly made themselves invisible in AI answers — which is the opposite of what they wanted. This Actor tells you exactly where you stand:

"aiVisibility": {
  "summary": { "retrievalAgentsAllowed": "6/7", "trainingCrawlersAllowed": "4/8",
               "citable": false, "pagesServerRendered": "19/19",
               "pagesWithAnswerShapedHeadings": 17, "pagesWithAuthorAndDate": 0 },
  "findings": [
    { "severity": "critical", "code": "ai-worst-of-both",
      "message": "Training crawlers are allowed while retrieval agents are blocked — your content may be used to train models but can never be cited back to you." },
    { "severity": "critical", "code": "ai-conflicting-robots-rules",
      "message": "robots.txt gives GPTBot, ClaudeBot both \"Disallow: /\" and \"Allow: /\". Which one wins is decided by each vendor's own tie-breaking… Typical cause: a CDN injected a managed block above your own rules." }
  ]
}

Three things this catches that a browser-based audit structurally cannot:

1. Contradictory robots.txt rules for the same crawler. Very common now that CDNs inject "managed" AI-blocking blocks above hand-written rules — the site then genuinely does not know what it is telling that crawler, because the outcome depends on each vendor's tie-breaking.
2. Content-Signal declarations, which are an express reservation of rights under Article 4 of the EU DSM Directive — a legal statement, flagged so you can check it says what you intend, especially if you did not add it.
3. Pages that are empty without JavaScript. AI retrieval agents do not execute scripts. This Actor reads the served HTML, so it sees your site exactly the way they do — a page Google ranks fine (Google renders) can be blank to every AI system, and only a non-rendering crawler can tell you.

Plus the signals AI answers rely on to lift and attribute a passage: question-shaped headings, FAQPage markup, an author and a date.

The robots.txt matcher implements Google's documented semantics — longest match wins, Allow beats Disallow on equal length, a group naming the agent beats * — and is covered by its own test suite (npm run test:robots, 14 cases). A wrong answer here would tell you that you are blocked when you are not, so it is asserted rather than assumed.

---

Beyond that, three more things ordinary SEO crawlers don't do:

1. It compares against your previous run. Every run stores a baseline per site, so the output says which pages got worse since last time, which improved, which are new and which disappeared. That is the difference between a report you read once and a monitor you can act on.
2. It hands you the fix, not just the finding. Each page carries a fixes array with the exact markup to paste: the title shortened at a word boundary, a meta description drafted from the page, the canonical, viewport, hreflang and Open Graph tags, ready to ship.
3. It finds contradictions, not just defects. Pages your sitemap asks Google to index while the page says noindex. Canonicals pointing at 404s or at pages that canonicalise onwards. Redirect chains and loops, traced hop by hop. Pages in your sitemap that nothing on your site links to. None of these are visible when you look at one page at a time — they only appear when you join the crawl, the sitemap, the canonicals and the link graph.
4. It validates structured data against what Google actually requires. Not "does the JSON-LD parse" — whether the block is eligible for a rich result at all. An Article without a headline is valid JSON and earns nothing.

One charge per page covers all of it — the AI-visibility audit, the comparison, the patches, the contradiction checks and everything below.

Plus everything you'd expect: broken links, titles, meta descriptions, headings, canonicals, indexability including the X-Robots-Tag header, hreflang, mobile viewport, thin content, mixed content, TLS certificate expiry, duplicate titles across the whole site, robots.txt, sitemap.xml, llms.txt — with a 0–100 score and a ranked recommendation list per page.

Built for bulk and for schedules: hundreds or thousands of pages, results in a dataset, reachable from the API, Make, Zapier or an MCP client.

What gets checked on every page

Check	Finding
Every <a href> link, internal and external	broken-links with the exact status code per dead URL
Title tag	missing-title, title-too-long (>65), title-too-short (<10)
Duplicate titles / descriptions across the whole site	duplicate-title, duplicate-meta-description — pages competing with each other
Meta description	missing-meta-description, meta-description-too-long (>170)
H1 structure	missing-h1, multiple-h1
Heading outline	heading-hierarchy-skip (h2 → h4 and similar jumps)
Canonical	missing-canonical, invalid-canonical, cross-domain-canonical, multiple-canonicals
Indexability	not-indexable with the reason — including the X-Robots-Tag HTTP header, the blocker that is invisible in the HTML
Meta robots	noindex (flags pages you may be de-indexing by accident)
Mobile readiness	missing-viewport
Language & hreflang	missing-lang, invalid-hreflang, hreflang-missing-self
Content depth	thin-content (< 250 words of real body text, scripts stripped)
Social cards	incomplete-social-tags (og:title/description/image, twitter:card)
Redirects	redirected (the crawled URL differs from the final URL)
Images	images-missing-alt with counts
JSON-LD structured data	structured-data-errors with the exact parse error
HTTPS hygiene	mixed-content (http resources on https pages)
TLS certificate	issuer, expiry date, days remaining; certificate-expiring under 30 days

Contradictions — findings that need two sources joined

Finding	What it means
sitemap-but-noindex	Your sitemap asks Google to index a page that tells Google not to
canonical-target-broken	The canonical points at a URL that answers 404 — the page is dropped entirely
canonical-chain	The canonical points at a page that canonicalises somewhere else again
redirect-chain / redirect-loop	Traced hop by hop; a loop means no user or crawler ever arrives
orphan-page	In the sitemap, linked from nowhere on the site
sitemap-url-redirects	The sitemap lists a URL that redirects instead of answering

Orphan detection is suppressed rather than guessed when the crawl could not support the claim — if maxPages cut the crawl short or the depth was too shallow, missing inbound links prove nothing. The summary says so explicitly in orphanDetection.note instead of reporting noise.

Fixes you can paste

"fixes": [
  { "issue": "title-too-long", "action": "replace", "selector": "head title",
    "snippet": "<title>XRechnung-Fehlercodes im Klartext: BR, BR-CO und BR-DE</title>" },
  { "issue": "missing-meta-description", "action": "insert-into-head",
    "snippet": "<meta name=\"description\" content=\"…\" />",
    "note": "Drafted from the first paragraph of this page — review the wording before shipping." }
]

Only deterministic fixes are emitted, and anything drafted from page content says so. No invented copy presented as fact.

Every page item carries a score from 0 to 100 and a recommendations array — the findings translated into plain sentences, sorted so the most damaging fix is first. That is what goes into a client report without rewriting.

Plus one site summary item per run: average score, robots.txt (exists? disallow-all accident? references sitemap?), sitemap.xml, llms.txt presence, the full issue histogram, every unreachable page, all certificates — and orphanPages: URLs that are in your sitemap but that no page on your site links to. Those are the pages search engines discover and then quietly ignore.

Example output (one page item)

{
    "type": "page",
    "url": "https://example.com/pricing",
    "statusCode": 200,
    "title": "Pricing – Example",
    "metaDescription": null,
    "h1Count": 1,
    "canonicalStatus": "self",
    "imagesWithoutAlt": 2,
    "jsonLdBlocks": 1,
    "jsonLdErrors": [],
    "linkCount": 58,
    "brokenLinkCount": 1,
    "brokenLinks": [{ "url": "https://example.com/old-docs", "status": 404 }],
    "issues": ["missing-meta-description", "images-missing-alt", "broken-links"],
    "issueCount": 3
}

How to audit a website in 3 steps

1. Enter the start URL — the Actor crawls all internal pages from there (same host), up to your maxPages limit.
2. Run it — every page is audited and every distinct link target is verified (HEAD with GET fallback, redirects followed).
3. Use the results — filter the dataset to issueCount > 0, export CSV for the client report, or wire it to Make/Zapier/Slack for alerts.

Example input

{
    "startUrls": [{ "url": "https://example.com" }],
    "maxPages": 50,
    "auditAiVisibility": true,
    "compareWithPreviousRun": true,
    "includeHealthyPages": false
}

That is a full technical audit of up to 50 pages, the AI-visibility check, and a diff against your previous run — issues-only output, ready for a client report.

Monitor continuously — this is the point

Schedule the Actor and every run answers "what changed":

"changes": {
  "comparedWith": "2026-07-22T06:11:03.000Z",
  "regressedCount": 1,
  "regressed": [{ "url": "https://example.com/pricing",
                  "newIssues": ["broken-links"], "scoreBefore": 98, "scoreNow": 83 }],
  "improvedCount": 4,
  "newPages": ["https://example.com/pricing-enterprise"],
  "pagesGone": ["https://example.com/old-offer"]
}

Wire regressedCount > 0 to Slack or email through an Apify integration and you have a regression alarm for your own or your clients' sites: the release that broke a canonical shows up the next morning, not in a quarterly audit. The baseline lives in a named key-value store per host, so it survives across runs and schedules. The first run has nothing to compare against and says so.

Turn off includeHealthyPages for a compact issues-only report.

Use with AI agents (MCP)

This Actor is available as a tool through the Apify MCP server, so agents in Claude, ChatGPT or any MCP client can run audits and read the results without a line of glue code. Prompts that map directly onto this Actor:

• "Audit example.com and list every broken link with its status code."
• "Check whether example.com is citable by ChatGPT and Perplexity, and tell me which robots.txt lines to change."
• "Compare example.com against last week's audit and summarize what got worse."

The dataset is plain JSON with stable field names, and the site summary item states its own caveats (suppressed orphan detection, skipped link checks) in prose — so an agent reading the output knows what the numbers do and do not claim.

What this adds over a standard SEO audit actor

Capability	Standard audit actors	This one
AI visibility: which AI crawlers your robots.txt actually allows, training vs. retrieval	—	✔ with per-vendor verdicts and conflicting-rule detection
Pages empty without JavaScript (what AI retrieval agents see)	— (browser-based audits structurally cannot)	✔
llms.txt check	—	✔
Diff against your previous run (changes)	—	✔ baseline per host, survives schedules
Ready-to-paste fixes (fixes[])	findings only	✔ exact markup snippets
Cross-source contradictions (sitemap vs. noindex, canonical chains, orphan pages)	—	✔
On-page checks, broken links, redirects, structured data	✔	✔
Core Web Vitals / page speed	some	— deliberately (see FAQ), use the companion Actor

FAQ

Does it use a headless browser? No — it audits the server-rendered HTML (what search-engine crawlers see first), which keeps runs fast and cheap. JavaScript-only content is out of scope for this version.

Does it measure page speed or Core Web Vitals? No, and deliberately not: LCP, CLS and TBT need a real Chrome run, which costs roughly ten times as much per page as reading HTML. Putting it in here would make every audit expensive for everybody. Use the Core Web Vitals Auditor for that — same author, built to run alongside this one: audit the site here, feed the URLs that matter into the measurement there.

How is the score calculated? Each finding subtracts a fixed weight from 100 — a broken link or a missing title costs far more than a slightly long meta description. Additive, capped at 0, no hidden model: you can reproduce any score from the issues array, and sort a dataset by it without having to trust us.

Will it hammer my site? Concurrency is capped at 10 requests and the Actor identifies itself with a clear User-Agent. It audits the site you point it at — intended for your own sites or sites you have permission to audit.

What counts as a broken link? Any link target answering HTTP ≥ 400, timing out, or failing DNS/TLS. Servers that reject HEAD requests are automatically re-checked with GET before being reported.

Can AI agents use this? Yes — via Apify's MCP server, agents can run audits and read results programmatically ("audit example.com and list all broken links").

Pricing

$0.05 per page audited. One charge covers everything for that page: all on-page checks, every link on it verified, the redirect chain traced, structured data validated against Google's requirements, the contradiction checks, the score, the ready-to-paste fixes and the comparison against your previous run. The site summary item is free, and maxPages caps the cost exactly.

No cheaper-than-the-others pitch: this is priced in the same band as the established bulk audit tools, and the argument for choosing it is the capabilities above that they don't have.

More from this publisher

• Core Web Vitals Auditor — bulk Lighthouse for a list of URLs: LCP, CLS, TBT, ranked fix opportunities. The measurement companion to this audit.
• TED Tender Scraper — search and monitor EU public procurement notices via the official TED API.
• E-Invoice Parser — XRechnung, ZUGFeRD/Factur-X and UBL e-invoices to clean, validated JSON.