Raw vs rendered
What ChatGPT sees vs what Google sees
Raw HTML is what GPTBot, ClaudeBot and PerplexityBot read. Rendered HTML is what Googlebot indexes. The Diff tab puts them side by side, line by line.
Googlebot · GPTBot · ClaudeBot · PerplexityBot
See your page the way Google and AI assistants do
Rendering, indexing, W3C, structured data and speed — every problem highlighted right on the page, in a 3D cross-section of its DOM.
Free during beta · interface in Russian for now
example.com/catalog
1,204 elements · render 3.2 s
● Googlebot · sees the full page
● AI crawlers · empty catalog
- JSON-LD: product has no price ●
- catalog text only after JS ▲
- image without alt ▲
- no catalog text in raw HTML ●
- product links only after JS ▲
AI crawlers read your page without JavaScript. The app shows exactly that: a no-JS 3D view, a raw vs rendered diff and an AI crawlers tab.
olive-tattoo.rua real check
What it looks like
A recording of a real check of a live site, sped up.Interface in Russian during beta.
Report
A report: every problem on the page
Hover a problem in the list and the element lights up in the 3D cross-section. Pick one and the source opens at the exact line.
Site crawl
A site crawl in 3D
Breadth-first from the homepage, like Googlebot. Pages sit on rings of click depth, color shows whether they’re indexable, and a site issue lights up its pages.
raw HTMLafter JavaScriptGooglebot · AI crawlers
What crawlers see that you don’t
The page looks fine in a browser, but crawlers get something else. Four cases SearcherVision finds and points to in the code.
raw HTML14 lines
<body> <div id="app"></div> <script src="/app.js"></script> </body>after JavaScript+318 lines
<h1>Sofas & armchairs</h1> <p>Delivered in 2 days…</p>▲ main text appears only after JavaScript
Content that only exists after JavaScript
Googlebot renders it. GPTBot, ClaudeBot and PerplexityBot never do — to them, that text isn’t there.
JSON-LD · FAQPage1 of 4 questions
{ "@type": "Question", "name": "How much is delivery?", "acceptedAnswer": { "text": "Free on orders over $300" } }▲ the answer isn’t in the page text
FAQ markup with no answers on the page
Google requires marked-up content to be visible. We check every answer against the page text.
<head>line 12
<meta name="yandex" content="noindex">- Googlebot
- ● indexable
- YandexBot
- ● not indexable
noindex for one crawler only
A header or meta tag that hides the page from one search engine and not the other.
after JavaScript<nav>
<nav> <a href="/catalog/chairs">Armchairs</a> <a href="/catalog/sofas">Sofas</a> </nav>▲ this link isn’t in the raw HTML
Links that appear only after JavaScript
Crawlers that don’t render will never follow them.
GooglebotGPTBotClaudeBotPerplexityBot
Google and AI crawlers, side by side
Googlebot renders JavaScript before indexing. AI crawlers — the ones behind ChatGPT, Claude and Perplexity — read only the raw HTML. SearcherVision shows both versions of your page and the exact difference between them.
Googlebot Smartphone
Googlebot/2.1 · renders in Chromium
- JavaScript
- executed
- Reads
- rendered HTML
- Catalog text
- present
●Sees the full page
AI crawlers
GPTBot · ClaudeBot · PerplexityBot
- JavaScript
- not executed
- Reads
- raw HTML only
- Catalog text
- missing
●Sees an empty catalog
Raw HTML vs rendered HTML, line by line
See exactly what exists only after JavaScript — the part AI assistants never read.
robots.txt rules for each crawler
Who is allowed in: GPTBot, ClaudeBot, PerplexityBot, Google-Extended — and whether the server really lets them through.
Also YandexBot
For sites in Russia and CIS — a separate Yandex verdict.
What we checkdeterministic rules
What we check on every page
Every check is deterministic. Each problem points to a line and column in the source and to the element in the 3D cross-section.
Indexing
- robots.txt, redirects, noindex and canonical in meta and headers.
Method: Googlebot Smartphone and YandexBot user agents
Rendering
- Mobile Chromium at 412×915, like Googlebot Smartphone.
- Resources blocked in robots.txt aren’t loaded; the page isn’t scrolled.
Method: Chromium · Playwright
Code & W3C
- Nu Html Checker errors tied to the exact element — in the code and in 3D.
Method: Nu Html Checker — the validator.w3.org engine
Structured data
- JSON-LD and Microdata against Google’s rules, shown as readable blocks: FAQ, prices, organization.
Method: schema.org vocabulary · Google rich result rules
Speed
- TTFB, FCP, LCP, CLS, TBT, plus PageSpeed Insights and real-user Chrome data when available.
Method: Chromium · PageSpeed Insights
Semantics
- What the page is about to a crawler: topic by zones — title, h1, URL, JSON-LD, headings.
Method: Deterministic rules: every issue has a place in the code
breadth-first crawlup to 1,000 pages
Crawl a whole site — and see it in 3D
Breadth-first from the start page, like Googlebot: up to 1,000 pages from raw HTML or 300 with rendering. Broken pages and who links to them, redirects, duplicate titles, orphans, sitemap gaps, keyword cannibalization — on a 3D map of click depth.
example.com · 134 pages
raw HTML crawl · 41 s
- indexable · 93
- not indexable · 16
- ▲redirect · 12
- ●4xx/5xx error · 7
- ○orphan · 6
rings — clicks from the homepage
MCPStreamable HTTP
Your coding agent fixes the SEO bugs
Connect SearcherVision to Claude Code, Cursor or any MCP client. The agent checks a page and gets every problem with its position in the code — then fixes it.
Server tools · 7
- analyze_page
- checks a page: Google and Yandex verdicts, top problems with their place in the HTML, a link to the 3D report
- get_report
- every problem in a report, filtered by category and severity
- get_html
- the source around a line and column — to find and fix it
- get_structured_data
- the structured data found on the page
- crawl_site
- crawls a site, up to 1,000 pages: broken pages, duplicates, orphans, cannibalization, sitemap, depth
- get_crawl
- crawl problems with every affected URL
- list_recent
- recent checks
Server address: https://api.searchervision.com/mcp
Claude Code · setup
claude mcp add --transport http searcher-vision https://api.searchervision.com/mcpCheck the homepage and fix the markup errors
searcher-vision · analyze_page(https://example.com/)
Google: indexable · Yandex: indexable
Problems: 3 errors, 5 warnings, 12 notes
Errors:
- [HTML (W3C)] Bad value "dialog" for attribute "role"
after JS 107:33498 <aside> cid=412
- [Structured data] Product › offers: "price" is missing
raw HTML 41:5get_html(cid=412) → editing components/Sidebar.vue
Pricing
Free during beta. Pro is coming.
Use SearcherVision free while it’s in beta. Pro and Agency are in development — join the list for early access.
Free
$0
Works today, no sign-up.
- 40 page checks per hour
- 6 site crawls per day
- MCP for your agent
Pro
in development$19/mo
For people who check sites every day.
- Higher limits
- Search Console integration
- AI crawler checks
- Client-ready reports
Agency
from $99/mo
For teams that run client sites.
- Team seats
- White-label reports
- API
Prices are a hypothesis — join the list and tell us what you’d pay for.
FAQPagemarkup matches the text
FAQ
The answers on this page and in its structured data match word for word — exactly what SearcherVision checks for FAQ markup.
How is this different from Google Search Console?
Search Console shows only your own verified sites and only Google. SearcherVision checks any URL, shows Google and AI crawlers side by side, and points to the exact element in the code.
Do I need access to the site?
No. Any public URL works.
Why might results differ from Google?
We emulate crawlers from our server. Geo rules and bot protection see our server, not Google. Search Console integration (coming in Pro) adds Google’s own data.
Do AI crawlers really skip JavaScript?
GPTBot, ClaudeBot and PerplexityBot fetch raw HTML and don’t execute JavaScript, according to Vercel’s analysis of their traffic. Google-Extended is the exception: it uses Google’s rendering.
What are the limits?
40 checks per hour and 6 crawls per day per IP during beta.
Why is the interface in Russian?
We started in Russia and CIS. The English interface is next — join the list and we’ll tell you when it’s ready.