RecipeBook by Shofo Review: Hidden data (60/100) — SiteList

Metric	Value
Domain	recipebook.shofo.ai
Category	Computer Vision Training Data
Pricing	Unknown
Pages Crawled	6
Crawl Date	2026-08-04

Shofo.ai is currently in a 'pre-visibility' state. While the underlying product (a massive video dataset) is a high-value linkable asset, the technical implementation of the Next.js frontend and a malformed robots.txt file are preventing search engines from discovering or ranking the site. The site currently lacks a keyword strategy, with metadata that is either too thin or duplicated across functional pages. Fixing these foundational issues is a prerequisite before any content or link-building investment can yield ROI.

Themes

• Indexation Roadblocks: The site is effectively 'dark' to crawlers due to a robots.txt file that returns a 404 error and the total absence of an XML sitemap.
• The Hydration Gap: The 'app-first' architecture hides the marketplace grid behind a JavaScript loading state, leading to a 'Thin Content' profile.
• Identity & Intent Mismatch: Page titles are generic and do not target high-intent keywords like 'Computer Vision Datasets'.
• AI Search Readiness Gap: The site lacks the structured data (Schema.org) required to be cited in AI-driven search results.
• Content Authority Deficit: With zero informational content, the site has no topical authority beyond its brand name.

RecipeBook positions itself as a high-utility video data marketplace for ML training, yet it lacks the trust signals required for enterprise adoption. While the claim of 25 million clips is specific and compelling, the homepage provides no proof adjacency, such as customer testimonials or research lab logos. The site relies entirely on its technical utility rather than established authority. This creates a friction point for procurement teams who require evidence of scale and reliability. To improve, the site should anchor its "by the hour" pricing with specific rates and introduce a "Trusted By" section to substantiate its market position.

The site uses technical jargon like "FPS" and "Train probe", assuming a high-sophistication audience. However, the question coverage for new users is poor. Critical information regarding hourly rates and data usage rights is missing from the primary crawl, creating high friction for high-intent buyers. While the self-orientation ratio is low—focusing correctly on the user's search task—the lack of "Voice of Customer" evidence remains a significant trust gap. Adding tooltips for technical features and a clear data licensing FAQ would reduce this initial hesitation.

RecipeBook functions well as a technical tool but fails as a self-contained marketplace. The search and filtering mechanics are highly relevant to researchers, allowing for immediate discovery without a login. However, the "business" side of the site is siloed on a separate domain. A user looking for support must click "Our website" in the footer, then navigate to a contact page on shofo.ai. This 3-click journey across domains is a significant friction point for lead generation. Additionally, technical buttons like "Train probe" lack tooltips, which may cause hesitation for first-time visitors unfamiliar with the ecosystem.

RecipeBook contains critical accessibility failures for screen reader users, specifically on the /report form. The legal reporting form contains eight inputs with zero accessible names, relying entirely on placeholders that disappear once a user begins typing. This violates WCAG 2.1 AA standards. Furthermore, the site lacks a "Skip to Content" link, forcing keyboard users to tab through the entire sidebar on every page load. The use of Tailwind's "focus:outline-none" also suppresses standard browser focus indicators, making navigation difficult for users with low vision who rely on clear visual cues to track their position.

RecipeBook uses a disciplined design system with only six distinct font sizes and 19 colors. However, this discipline does not extend to mobile ergonomics. Our audit found over 300 instances of interactive elements, such as the "+ Order" and menu buttons, that are significantly smaller than the recommended 44px touch target. Additionally, the primary brand purple (#9f1bfe) fails the 4.5:1 contrast threshold on the site's dark background, computing to 4.02:1. This affects the readability of links and small UI labels for users with visual impairments.

The site is functionally fast once loaded but suffers from structural delays during the initial visit phase. Because the main results grid relies on client-side data fetching, users and search crawlers initially encounter a "Searching..." skeleton state rather than the 25M+ clips promised. This pushes the Largest Contentful Paint (LCP) deep into the timeline. Furthermore, while the subdomain is relatively lean, the parent domain serves a massive 42.1MB payload, indicating a lack of asset-pipeline discipline. We also detected 18 different font faces, which adds unnecessary blocking requests. Transitioning to server-side rendering for the initial results would significantly improve perceived speed.

Editorial quality is split between a minimalist app interface and unnavigable legal documentation. The Enterprise Terms page presents over 2,800 words in a single paragraph block, making it nearly impossible to read on any device. Additionally, the site uses unsupported superlatives like "The World's Largest Video Library," which may alienate a technical audience of ML engineers who prefer verifiable metrics. Replacing these claims with the concrete "25M+ clips" figure found in the metadata would build immediate credibility. The site also suffers from duplicate meta descriptions across four functional pages, which hinders search orientation.

RecipeBook is a stable but vulnerable new domain with a strong 80/100 score in this dimension. Registered in February 2025, the site lacks legacy baggage but faces significant risk from AI-driven search erosion. Because the marketplace targets technical computer vision data, its core keywords are prime targets for Google's AI Overviews. Without Dataset schema, the site may lose click-through traffic to automated summaries. Establishing niche-relevant backlinks and implementing structured data are the primary paths to stabilizing this new authority.

The site's editorial quality is currently weak at 48/100, marked by inconsistent mechanical execution. A primary concern is the use of the unsupported superlative "World's Largest Video Library," which lacks citation or comparative data. Furthermore, 66% of crawled pages feature duplicate titles and meta descriptions, suggesting a templated approach that lacks manual review. The legal pages are particularly dense, with sentence lengths averaging 49 words, significantly exceeding the 20-word target for web clarity. Immediate fixes include differentiating page titles and adding citations for marketplace claims.

The site provides no public-facing documentation, resulting in a critical score of 12/100. Probes for common paths like /docs, /api, and /faq all returned 404 errors. This creates significant friction for ML engineers who require technical specifications, data schemas, and licensing details before committing to a purchase. The absence of an llms.txt file further prevents AI-driven research tools from accurately representing the marketplace's 25M+ clips to potential enterprise buyers. Establishing a public help center and a technical spec sheet is a high-priority requirement for self-serve conversion.

Technical SEO is fair at 66/100, but foundational errors hinder crawlability. The robots.txt file is malformed, returning a 404 HTML page that contains a noindex meta tag—a confusing signal for search bots. Additionally, there is no XML sitemap to guide discovery. A significant "negative rendering gap" exists: the raw HTML contains 557 words of data, but the rendered DOM shows only 64 words, suggesting that core value propositions are hidden behind JavaScript hydration. Deploying a plain-text robots.txt and an XML sitemap are the most urgent technical requirements.

RecipeBook by Shofo is a high-utility marketplace for AI training data that is currently undermined by its technical execution. The site’s strongest asset is its underlying data—25 million video clips—and a clean, modern design stack. However, it fails on basic SEO hygiene, documentation, and trust signals.

Fixable Weaknesses:

1. Technical indexation (robots.txt and sitemap).
2. Metadata optimization for high-intent ML keywords.
3. Public-facing documentation and licensing details.

Who it's for: ML Engineers and Computer Vision Researchers who already know the brand or are referred directly to the marketplace.

The review is based on a crawl of 6 pages conducted on 2026-08-04. Several dimensions were excluded: 06, 08, and 18 failed to meet minimum content thresholds, while 12, 13, 22, 28, and 31 were not applicable to this marketplace model. Measurement readiness was assessed via PostHog implementation. For a full explanation of our 13-dimension framework, visit our methodology page.