GTM Tool Website Scraper

Build a GTM (go-to-market) tool directory at scale. Give this actor a list of tool website URLs and it returns, for each URL, a clean structured record:

• tool_name — the company/tool name
• summary — a plain-text description of what the tool does (deterministically kept under 500 words by default)
• demo_video_url — the first demo/product video found on the site
• demo_video_source — where that video lives (youtube, vimeo, wistia, loom, html5, link, or none)

It renders every site with a real headless Chromium browser (Playwright), so modern client-side-rendered React / Next.js / Webflow marketing sites work just as well as server-rendered ones.

---

Why this actor

Generic website-content crawlers hand you a wall of raw page text. This actor gives you exactly the three fields a GTM tool directory needs — name, a concise "what it does" summary, and the demo video — as clean structured columns, one row per URL. No LLM cost, fully deterministic and reproducible.

Who it's for: GTM / sales-ops teams building tool directories, product researchers, VCs mapping software categories, and content teams writing tool round-ups.

Legal: it only visits the tool's own public marketing website — no login, no gated data.

---

Input

Field	Type	Required	Default	Description
startUrls	array of { url }	✅	—	The GTM tool website URLs to scrape. One output item per URL.
summaryMaxWords	integer	—	500	Hard cap on summary length (min 20, max 2000). The summary is truncated on a word boundary to stay strictly under this number.
maxRequestsPerCrawl	integer	—	100	Safety cap on total requests to prevent runaway crawls.

Example input

{
  "startUrls": [
    { "url": "https://www.clay.com/" }
  ],
  "summaryMaxWords": 500,
  "maxRequestsPerCrawl": 100
}

---

Output

One dataset item per input URL. Every value is a string.

Field	Description
url	The URL that was scraped (resolved loadedUrl).
tool_name	Company/tool name.
summary	Plain-text "what it does" summary, under summaryMaxWords words.
demo_video_url	First demo/product video URL, or "" if none.
demo_video_source	youtube | vimeo | wistia | loom | html5 | link | none.
scraped_at	ISO-8601 UTC timestamp.

Example output

{
  "url": "https://www.clay.com/",
  "tool_name": "Clay",
  "summary": "Infrastructure to get any data, run agentic workflows, and launch GTM plays. Build systems to grow revenue ...",
  "demo_video_url": "",
  "demo_video_source": "none",
  "scraped_at": "2026-07-16T12:00:00.000Z"
}

On clay.com the hero video is a decorative background loop (autoplay loop muted), which is intentionally excluded from demo detection, and /demo is a gated "book a demo" form with no embedded player — so demo_video_source is correctly none.

---

How extraction works

Tool name (first non-empty wins)

1. og:site_name
2. og:title (split on |, -, —, : — keep the brand segment)
3. <title> (same cleaning)
4. header/nav logo img[alt]
5. capitalized registrable domain (e.g. clay.com → Clay)

Summary (deterministic extract-and-truncate — no LLM)

Candidate text is assembled in priority order — meta description → og:description → hero <h1> → hero subhead → first substantive <main> paragraphs → feature heading/description blocks — then whitespace-normalized, deduped, and truncated on a word boundary to strictly under summaryMaxWords. Nav, footer, cookie/consent and legal boilerplate are filtered out.

Demo video (first-match priority)

1. YouTube embeds / youtu.be / lite-youtube facades → youtube
2. Vimeo player iframes → vimeo
3. Wistia embeds → wistia
4. Loom embed/share links → loom
5. Genuine <video> mp4 — decorative background loops excluded (autoplay + loop/muted, or video-bg/hero-video/bg-video classes) → html5
6. Anchor last resort (/demo, /watch, /video, /product-tour, or "watch demo" / "see it in action" / "product tour" link text) → link
7. Nothing → "" + none

All video URLs are resolved to absolute URLs.

---

Edge cases handled

• Client-side SPAs (rendered with Playwright before extraction)
• Decorative background <video> loops excluded from demo detection
• Demos gated behind a "book a demo" form → none
• Missing meta tags → priority-chain fallbacks, domain fallback for the name
• Title padded with taglines/separators → brand segment kept
• Cookie banners / nav / footer legal text filtered out of the summary
• Very long pages → word-boundary truncation under the cap
• Redirects (http→https, www, trailing slash) → original input URL recorded, resolved page scraped
• Timeouts / 4xx-5xx → retried, then a best-effort row is still emitted (never lose a URL)
• Relative / protocol-relative video hrefs → resolved to absolute
• Duplicate input URLs → deduped to one item each

---

Notes

• Uses Apify Proxy automatically to avoid IP blocks and geo-redirects across many domains.
• Memory: ~2 GB recommended (headless Chromium).
• Anti-detection/stealth is not needed — these are public marketing sites with no bot-hostile gating.