LLM Visibility Audit: Will ChatGPT, Claude and Gemini Cite Your Site?

Why LLM Visibility Matters for Your Website in 2026

The SEO paradigm has shifted. Until 2023, the question was “how do I rank on Google?” In 2026 the question is twofold: “how do I rank on Google?” and “how do I get cited by LLMs when someone asks about my industry?” The distinction matters because the traffic flows are no longer the same.

The person who used to type “best chatbot software for restaurants” into Google and click through five results now asks that exact question to Claude, ChatGPT or Perplexity, and gets a consolidated answer citing three to five sources. If your website is one of those sources, you win — the LLM just handed you authority and traffic. If it isn’t, you don’t even enter the conversation.

Traditional SEO isn’t enough for this second battle. There’s overlap — semantic HTML, authority, clear content — but LLMs evaluate additional signals that a Lighthouse report or Search Console simply doesn’t measure. How Google’s AI Mode competes with ChatGPT Search shows where this whole ecosystem is heading.

How to Test Your Website Against LLMs (3 Methods)

Before you optimize anything, measure. Here are three practical methods to find out what LLMs actually know about your website today.

Method 1: Ask About Your Brand Directly

In Claude, ChatGPT, Perplexity and Gemini, ask literally: “what is [your business name]?”, “tell me about [your brand]”, “what services does [your business] offer?” Note what each model gets right, what it gets wrong, and what it makes up (“hallucinates”). This gives you a qualitative baseline. If all four LLMs say different things — or worse, invent details — your authority as a source is out of control.

Method 2: Ask the Queries Your Sector Actually Gets Asked

Ask the natural questions your customers would ask: “AI consultant near me,” “best local SEO agency,” “difference between a chatbot and an AI agent for small business” — or, if you run a real estate agency, literally “which real estate agency would you recommend,” which is exactly what a visibility audit for real estate agencies reviews. Check whether your site gets cited, and in what position. If you never show up, there’s an authority or readability problem. If you show up but with wrong information, the LLM is reading you, just not understanding you correctly.

Method 3: Inspect Your Server Logs

In your access logs, look for user-agents that indicate LLM crawling: GPTBot (OpenAI), ClaudeBot (Anthropic), Google-Extended (Gemini), PerplexityBot, CCBot (Common Crawl, which feeds several models downstream). The frequency and depth with which these bots visit your site is a direct indicator of how seriously they’re considering you as a source. On this portfolio, the logs show visits from all four major bots at different frequencies — I’ll break down the real numbers below.

The 5 Things an LLM Evaluates When It Reads Your Site

Of these five, the last one is the least obvious and the one corporate websites get wrong most often. Plenty of companies in 2024-2025 blocked AI bots “to protect against scraping” without understanding they were opting out of appearing in generative answers entirely. Today that block translates directly into invisibility for users who now ask their questions to an LLM before they ever open Google.

Real Case: LLM Audit of This Portfolio

Applying all three methods to the very site you’re reading this on, here’s what came back:

• Method 1 (direct question): Claude answers with correct facts about the consultant (Calafell location, SEO + AI services, portfolio launch year). ChatGPT also gets it right but occasionally mixes it up with a different consultant with a similar surname. Perplexity gets it right and cites 2-3 concrete URLs. Gemini has partial information. Conclusion: Claude and Perplexity read this site best.
• Method 2 (sector queries): for “AI consultant Calafell” or “SEO consultant Tarragona,” the site gets cited by Perplexity and Claude among 2-5 sources. For broader queries (“best SEO consultants in Spain”), it doesn’t appear — not enough authority yet to compete against large agencies.
• Method 3 (logs): GPTBot ~120 visits/month, ClaudeBot ~85 visits/month, PerplexityBot ~40 visits/month, Google-Extended ~60 visits/month. That frequency has grown 3-4x since early 2025.

The elements that were actively worked on to earn this LLM readability:

• Strict semantic HTML: every post follows the h1 → h2 → h3 hierarchy with no skipped levels. This makes it trivial for an LLM to extract structure.
• Rigorous Schema.org: Person on the homepage, ProfessionalService + Article on hub pages, FAQPage + BreadcrumbList on every post. LLMs use this as confirmation of identity and content type.
• A dense network of internal links between articles on related topics. This builds a relationship graph the LLM can lean on to build context.
• robots.txt explicitly allows GPTBot, ClaudeBot, PerplexityBot, Google-Extended. It doesn’t block or throttle any legitimate bot.
• Dense language with no filler. Concrete sentences with data, numbers, examples. Every post includes a section on mistakes made along the way, which adds experiential authority.

How Claude, ChatGPT, Perplexity and Gemini Differ When Reading Websites

The practical takeaway: optimizing for Perplexity and Claude has the best return in 2026 — they cite URLs in responses more than the others, which translates into measurable traffic. ChatGPT and Gemini are more opaque about it, but they still carry weight in how they consider you as a source.

6 Mistakes That Make Your Website Invisible to LLMs

• Blocking AI bots in robots.txt: the most serious and most common mistake. Companies that added a Disallow rule for GPTBot/ClaudeBot in 2024 “to protect their content” now don’t appear in generative answers at all. If you want to be cited, you can’t deny access.
• Content hidden behind heavy JavaScript: many LLMs either don’t execute JS or do so with real limitations. If your actual copy only appears after a heavy SPA renders client-side, the LLM never reads it. Server-side rendering or at least minimal pre-rendering is essential.
• Blocking cookie walls: if the bot hits a wall before anything else, it never reaches your content. The banner needs to be non-blocking — content should stay accessible even while the banner is showing.
• Non-semantic HTML: sites built entirely out of <div> instead of <h2>, <article>, <section>. The LLM can’t extract hierarchy and treats the content as flat text. The difference in citation rate is enormous.
• Missing Schema.org: without structured data, the LLM has to infer entity types from context alone. With Schema, it infers them directly. Still optional, but adoption is climbing fast.
• Generic content copied from templates: LLMs detect duplicate or near-duplicate content across sites and discard it as a source. Genuine editorial differentiation remains the only sustainable way to earn authority.

How to Fix It (In Priority Order)

1. Check robots.txt: make sure it does NOT block GPTBot, ClaudeBot, PerplexityBot, Google-Extended. Five minutes of work for the most obvious win available.
2. Reinforce semantic HTML: a hierarchical h1-h2-h3 structure, real lists instead of <br> tags, tables for comparisons. This is retroactively applicable, post by post.
3. Implement rigorous Schema.org: a complete guide on which Schema types to use for each case makes this straightforward to execute.
4. Create an llms.txt at your domain root (see next section). Optional, but cheap.
5. Densify internal links: every article on a given topic should link to 3-5 related pieces. The LLM uses that network to understand topical authority.
6. Rewrite generic copy: if your homepage or services pages read like template text, no LLM will prioritize them. Rewriting with real cases, data and concrete examples is what actually moves the needle.

llms.txt: The Emerging New Standard

Inspired by robots.txt, the llms.txt file lives at your domain root (yourdomain.com/llms.txt) and describes your site in markdown, a format LLMs can parse easily. Its purpose: give models a structured map of the site without forcing them to crawl the entire HTML hierarchy to figure out what matters.

Typical structure (the format proposed by the emerging spec):

# Cristian Corrales

> AI + SEO consultant in Calafell, Tarragona. AI implementation for SMBs and local SEO on Spain's Costa Daurada.

## Core services

- [Local SEO consulting](https://cristiancorrales.com/servicios/seo-local.html): technical and local SEO across the Camp de Tarragona region.
- [Professional web design](https://cristiancorrales.com/servicios/diseno-web.html): vanilla JS sites optimized for Web Vitals.

## Featured posts — AI

- [AI agents for SMBs](https://cristiancorrales.com/blog/agentes-ia-pymes-que-son.html)
- [MCP servers explained](https://cristiancorrales.com/blog/mcp-servers-explicados-sin-tecnicismos.html)
- [ROI of an AI project](https://cristiancorrales.com/blog/roi-proyecto-ia-pyme-2026.html)

## Featured posts — SEO

- [Schema.org in 2026](https://cristiancorrales.com/blog/seo-tecnico-schema-org-2026.html)
- [Web Vitals 2026](https://cristiancorrales.com/blog/web-vitals-2026-que-cambia-como-medir.html)
- [Technical SEO audit step by step](https://cristiancorrales.com/blog/auditoria-seo-tecnica-paso-a-paso.html)

Perplexity already announced official support in 2025. Anthropic and OpenAI have mentioned it as good practice without a firm commitment yet. Implementing it takes 1-2 hours, and while its immediate impact is modest, it positions your site for when the standard matures.

Mistakes I Made on This Very Site

• In 2024, I left a Disallow: / rule in robots.txt for GPTBot for four months — following defensive advice that seemed sensible at the time. When I removed it and the logs showed bots starting to crawl frequently, I realized the real cost of that decision. Lesson: defending against legitimate bots equals invisibility in the generative ecosystem.
• I underestimated Schema.org. I thought it was a small-scale local SEO thing. But LLMs use it aggressively as a signal of authority. The difference in citations after implementing rigorous Schema was noticeable on Claude and Perplexity within 4-6 weeks.
• My internal linking between related articles was thin. Every post was an island. Once I started adding 5-8 natural inline links per post pointing to sibling articles on the same topic, LLMs started citing multiple articles from the site in the same answer — the relationship graph became legible to them.
• I didn’t have an llms.txt. It’s still not an established standard even today, but implementing it costs about an hour. It’s a cheap win that isn’t worth skipping.

Conclusion: Traditional SEO Isn’t Going Away, But It’s No Longer Enough

Optimizing for LLMs doesn’t replace traditional SEO — it complements it. The good news is that 70% of the practices overlap: semantic HTML, Schema, authority, well-written content, decent Web Vitals. The remaining 30% is specific to this new channel (not blocking AI bots, llms.txt, dense internal linking, dense language) and it’s where most websites still have room to differentiate themselves.

For a local SMB anywhere outside the biggest metro markets, the opportunity right now is enormous: most competitors still aren’t thinking about LLMs at all. A technical SEO audit that includes LLM readability, paired with fixing the six mistakes listed above, can move citations in generative answers within a few weeks. And as a bonus, whatever you fix for LLMs almost always improves Web Vitals and classic Google rankings too.

Is Your Website Readable by Claude, ChatGPT and Perplexity?

Free LLM audit: I test your website against the 4 major LLMs, review your robots.txt, Schema, and internal linking, and hand you an improvement plan prioritized by cost and impact.

Request an LLM audit

Or write to me directly at hola@cristiancorrales.com.

Frequently Asked Questions About LLM Audits

What is an LLM audit and how is it different from traditional SEO?

An LLM audit evaluates how well generative models read, understand, and use your website as a source. Traditional SEO optimizes for Google to rank the URL; SEO for LLMs aims for your brand to get cited when an LLM answers a question in your industry.

Is it worth optimizing for LLMs if Google is still dominant?

In 2026, yes. Traffic from LLMs already exceeds 5-15% of total traffic in many B2B and technical verticals. And the techniques that optimize for LLMs almost always improve traditional SEO too, so the opportunity cost is low.

Are SEO and GEO (Generative Engine Optimization) the same thing?

They share 70% of the same practices but diverge on the remaining 30%. Traditional SEO aims to rank URLs; GEO aims for your brand or content to be cited by LLMs as an authoritative source. They’re complementary, not mutually exclusive.

How do I know if LLMs are already sending traffic to my website?

Check the referrer in Google Analytics. Traffic from chatgpt.com, perplexity.ai, claude.ai (citations), or google.com/?udm=50 (AI Mode). If it’s still marginal, that doesn’t mean your site isn’t being cited — it can mean the LLM’s answer already fully covers the query without the user needing to click through.

Do I need to create an llms.txt file?

It’s good emerging practice, not a requirement. It’s a plain text file at your domain root that describes your site in markdown for LLMs to parse. Some LLMs already read it as an aid to indexing you. Creating it takes 1-2 hours and is a cheap win.