When we first ran an AI visibility audit on this site, Nutriish scored 36 out of 100. Not because the product was bad. Not because the design was broken. Because AI systems like ChatGPT, Claude, and Gemini had no reliable way to understand what the site was, who it served, or what it offered.
This is the story of how we changed that — in two phases, under 30 days, without touching the visual design — and what the data showed when we ran citation tracking across 4 AI engines afterward.
Full score history and live citation data: aiauditscan.com/case-study/nutriish
Why AI Visibility Is Different From SEO
Search engines rank pages. AI systems cite sources. These are fundamentally different tasks.
A search engine reads your HTML and scores you on links, keywords, and page speed. An LLM needs to understand your site — what entity you are, what you offer, who you serve, whether you can be trusted, and whether your content is structured in a way it can extract and summarize.
Most sites that perform well in traditional search are still completely invisible to AI. The fixes are not the same.
The Baseline: What 36/100 Looks Like
The initial scan on February 22, 2026 revealed a textbook case of AI invisibility:
- 47 words on the homepage — below the minimum threshold for LLM content extraction
- Zero structured data — no JSON-LD, no Schema markup, no Organization or Service context
- No robots.txt — AI crawlers had no explicit permission to index the site
- No canonical tags — duplicate content risk
- Zero H2 or H3 headings — no content structure for LLMs to navigate
A score of 36/100 does not mean the website was broken. It means AI systems could visit the page and extract almost nothing useful from it.
Phase 1: The Technical Foundation (One Session, +21 Points)
On February 23, 2026, five structural changes moved the score from 36 to 57/100 without changing anything the human visitor sees.
JSON-LD structured data — We added Organization, WebSite, and Service schemas in the <head>. This gives LLMs machine-readable entity data: what Nutriish is, what it does, who it serves, and how to contact it.
FAQ Schema — Four structured Q&A pairs added to the FAQ page. Question-based content is the format LLMs most reliably extract and cite. This single change disproportionately improves citation rates on conversational queries.
Word count — Homepage content expanded from 47 to 342 words. This crosses the practical extraction threshold for most LLMs. Below ~200 words, models treat the page as a stub and skip it.
robots.txt — Created with explicit permissions for all major AI crawlers: GPTBot, ClaudeBot, anthropic-ai, Google-Extended, PerplexityBot, and 12 others. Crawlers that cannot find explicit permission increasingly default to skipping.
Canonical tags — Added site-wide. Prevents LLMs from citing a paginated or filtered version of a page instead of the canonical source.
Total time: one working session. Total cost: zero.
Phase 2: Authority and Content Signals (February — March 2026)
The second wave targeted the two lowest-scoring pillars: Publisher Signals and Entity Clarity.
Article schema on all news posts — Each article now carries structured Article metadata including datePublished, author, publisher, and mainEntityOfPage. This makes each post individually citable by LLMs, not just the homepage.
Automated content pipeline — A Node.js script fetches nutrition RSS feeds from ScienceDaily and other sources, generates listicle-format articles via Gemini, and deploys to Eleventy automatically. Fresh, structured, topically relevant content is the single strongest signal for content freshness — one of the highest-weighted AEO pillars.
llms.txt — A plain-text file at nutriish.me/llms.txt that describes the site in structured prose for AI systems. The emerging standard (similar to robots.txt but for LLM context). Includes entity disambiguation: Nutriish the nutrition concierge is explicitly distinguished from Rachael Ray's Nutrish pet food brand — a critical fix given Perplexity's confusion between similar brand names.
Honest failure — Wikidata — We created a Wikidata entry for Nutriish to establish machine-readable entity disambiguation. It was deleted by moderators citing insufficient third-party coverage. This is a known constraint for early-stage brands. The fix is PR-first: earning coverage in independent publications before re-attempting the entry.
The Results: 58/100 and 85% Citation Rate Across 4 AI Engines
The March 2026 scan returned a global AEO Index of 58/100 — 26 points above the nutrition industry average of 32/100, and above the general web average of 38/100 across 5,600+ sites audited by AI Audit Scan.
More importantly, the citation tracking run told the real story:
| Engine | Citation rate | Key finding |
|---|---|---|
| ChatGPT | 100% (5/5) | Cites Nutriish consistently against MyFitnessPal |
| Claude | 100% (5/5) | Cites reliably, flags limited third-party verification |
| Gemini | 80% (4/5) | Strong on comparison queries, one entity gap |
| Perplexity | 60% (3/5) | Confuses Nutriish with Nourish — entity gap |
ChatGPT and Claude now cite Nutriish on digital nutrition queries. Reliably. Perplexity remains the blocker — its RAG engine returned "I believe you are asking about Nourish, an online dietitian platform" on one of five queries. This is the entity disambiguation problem in action, and it cannot be fixed with on-site changes alone. It requires off-site authority: coverage in independent nutrition publications, and eventually a Wikidata entry with verifiable references.
What Comes Next
The path from 58 to the top-quartile threshold of 65/100 runs through three actions: earning coverage in at least two independent wellness publications, publishing long-form content with question-based H2 headings to close the Content Structure gap, and activating a LinkedIn company page to add the final Publisher Signal.
All the on-site technical work is done. The remaining gap is off-site authority — which takes time but starts with knowing where you stand.
What This Means If You Run a Website
Nutriish is not an outlier. The average AI readiness score across 5,600+ sites is 38/100. Most sites that rank well in Google are still invisible to the AI systems that an increasing share of users now consult first.
The technical fixes are not radical. A properly structured site can move from invisible to citable in days. The harder work — earning the off-site authority that source-based engines require — takes longer. But it starts with a score.
This audit, and everything documented here, was run using AI Audit Scan — a tool that measures brand visibility across AI answer engines and gives you a score with specific, actionable recommendations. Free audit takes 30 seconds.
Nutriish is a live AEO optimization case study maintained by AI Audit Scan. Live score and citation tracking: aiauditscan.com/case-study/nutriish.