elsaspeak.com — AI Site Grade

Analysis

I have enough data now. Let me compile the audit.

The site has no structured data at all — zero JSON-LD schemas on any page examined, despite being an AI company that should be the most obvious candidate for Organization, FAQPage, Product, SoftwareApplication, and Course markup.

Crawler Access

All major AI crawlers — GPTBot, ClaudeBot, PerplexityBot, Google-Extended, OAI-SearchBot, ChatGPT-User, Applebot-Extended, Bytespider, anthropic-ai — receive a full 200 response with identical byte-size content (167,541 bytes) as a browser on https://elsaspeak.com/en. No UA-based blocking exists. The site runs on Cloudflare with HSTS, CSP, and X-Frame-Options: DENY. The robots.txt contains a single Disallow: /api/ rule for User-agent: * and no AI-bot-specific directives at all. /llms.txt returns a 404 (WordPress 404 page). The sitemap at sitemap_index.xml is valid with 5 sub-sitemaps covering ~335+ URLs across multiple locales (en, ja, vi, zh, th, id, pt, es).

Schema Posture

Zero JSON-LD schemas were found on any of the 10+ pages inspected: homepage, about-us, efficacy, accolades, blog, FAQs, enterprise/business, subscription, and individual FAQ articles. No Organization, FAQPage, Product, SoftwareApplication, BreadcrumbList, Course, or WebSite schema exists. The FAQ section has 30+ individual Q&A pages with clear question-answer pairs but no FAQPage markup. The efficacy page describes research-backed methodology, CEFR alignment, and a scoring engine with 93.88% expert-rater agreement — all natural candidates for ScholarlyArticle or ClaimReview schema — but none is present.

Cold-Knowledge Gap

The LLM cold-knowledge prior describes ELSA Speak as a "50M+ download" pronunciation app founded in 2015, based in San Francisco and Ho Chi Minh City, with a $23M Series C in 2023. The actual site claims 92M+ downloads and 90M+ learners — a gap of ~40M users that the model does not reflect. The site also prominently features a World Economic Forum Technology Pioneer 2024 award, a new CTO (Akshaya Aradhya, ex-GitHub/Netflix), and an expanded product suite (ELSA Business, ELSA Schools, ELSA API, Speech Analyzer, AI Role-play, multi-accent training) that the cold model does not mention. The model's knowledge is frozen at roughly the 2023-early 2024 state.

Content & Structure

The homepage is a rich, ~1,338-word marketing page with clear H1/H2 hierarchy, FAQ signals, and comparison language. However, the FAQ pages are thin: individual FAQ articles return only 34-65 words of visible text, and the FAQ category pages are essentially link lists. The blog has only 6 visible articles, with dates ranging from October 2025 to January 2026 (future-dated content). The news-events page is nearly empty (27 words, one entry dated May 2026). The site uses WordPress (Rank Math SEO, wp-sentry, wp-custom-css) with a custom theme, and pages are server-rendered (no JS-shell risk — all bots get full HTML).

External Signals

The accolades page links to external press coverage (TechCrunch, Forbes, CNBC, Digitimes, Entrepreneur) but these are from 2016-2021. The most recent external press link is from April 2025 (techpoint.africa). No Reddit threads, independent review sites, or third-party app store review pages were found in search results. The site has 11 Google Search Console verification TXT records — an unusually high number suggesting many past ownership changes or agency handoffs.

Findings

1. Zero JSON-LD schemas on any page High

   No structured data (Organization, FAQPage, Product, SoftwareApplication, Course, etc.) was found on any of the 10+ pages inspected, including the homepage, about-us, efficacy, accolades, blog, FAQs, enterprise, and subscription pages.

   What to change: Add JSON-LD schemas for Organization, FAQPage, Product, SoftwareApplication, Course, BreadcrumbList, and WebSite to relevant pages.

2. Cold LLM knowledge undercounts users by ~40M Medium

   The LLM cold-knowledge prior states 50M+ downloads, but the site claims 92M+ downloads and 90M+ learners. The model also lacks awareness of the World Economic Forum Technology Pioneer 2024 award, new CTO, and expanded product suite.

   What to change: Publish an /llms.txt file and update the site with structured data to help LLMs stay current with the latest metrics and awards.

3. Missing /llms.txt file Medium

   The /llms.txt endpoint returns a 404 (WordPress 404 page). This file would help LLMs discover key pages and facts about the company.

   What to change: Create an /llms.txt file listing important URLs and a brief company summary.

4. FAQ pages have very thin content Medium

   Individual FAQ articles contain only 34-65 words of visible text, and category pages are essentially link lists. This provides little value to users or crawlers.

   What to change: Expand FAQ answers to at least 150-200 words each and add FAQPage schema markup.

5. Blog posts have future dates Medium

   The blog shows only 6 articles with dates ranging from October 2025 to January 2026, which are in the future relative to the audit date. This may confuse crawlers and users.

   What to change: Correct the dates on blog posts to reflect actual publication dates.

6. News and events page is nearly empty Low

   The news-events page contains only 27 words and one entry dated May 2026, providing no useful content.

   What to change: Populate the news-events page with actual recent news and events, or remove it.

7. External press links are mostly from 2016-2021 Low

   The accolades page links to press coverage from TechCrunch, Forbes, CNBC, etc., but most are from 2016-2021. Only one link from April 2025 is recent.

   What to change: Update the accolades page with recent press mentions and awards.

8. No AI-bot-specific directives in robots.txt Low

   The robots.txt only has a Disallow: /api/ for User-agent: * and does not mention any AI crawlers by name. While not blocking them, this misses an opportunity to guide AI crawlers.

   What to change: Consider adding specific directives for AI crawlers to allow or disallow certain paths.

9. Unusually high number of Google Search Console verification TXT records Low

   The DNS has 11 Google Search Console verification TXT records, suggesting many past ownership changes or agency handoffs, which could indicate inconsistent SEO management.

   What to change: Clean up old verification records to retain only the current ones.

What's working

• All major AI crawlers receive full content — All 11 tested AI crawlers (GPTBot, ClaudeBot, etc.) receive a 200 response with identical HTML content as a browser, with no UA-based blocking.
• Pages are server-rendered with full HTML — All pages inspected return full HTML content to bots, with no JavaScript shell issues. Bots receive the same content as browsers.
• Valid sitemap with multiple sub-sitemaps — The sitemap_index.xml is valid and contains 5 sub-sitemaps covering ~335+ URLs across multiple locales, aiding crawler discovery.
• Homepage has rich marketing content with clear hierarchy — The homepage is a ~1,338-word page with clear H1/H2 headings, FAQ signals, and comparison language, providing good context for crawlers.
• Efficacy page details research methodology and metrics — The efficacy page describes CEFR alignment, a scoring engine with 93.88% expert-rater agreement, and research-backed methodology, which are strong candidates for structured data.
• Enterprise page details product suite for businesses — The enterprise page (1,267 words) describes ELSA Business, ELSA Schools, ELSA API, Speech Analyzer, and AI Role-play, providing comprehensive information for B2B crawlers.
• Accolades page lists awards and press mentions — The accolades page (354 words) lists awards and links to external press coverage, providing social proof for crawlers.
• Blog contains relevant articles about new features — The blog has articles like 'Discover the New ELSA Speak Experience' (649 words) that provide useful content for users and crawlers.