Table of Contents
- AI Visibility Diagnostic
more clicks from search · Benarrivati
- Request diagnostic
Avenue Cookery School moved from invisible to GBP 2,262 in ChatGPT-attributed revenue in 30 days after we fixed four extractability failures.
- ChatGPT picks sources by trust, relevance, and extractability — not by Google rank.
- Only 10% of Google’s top results overlap with ChatGPT citations (Rank Math, 2026).
- Four failures explain most invisibility: entity drift, retrieval-shape mismatch, crawler routing collapse, attribution gaps.
- Most fixes ship in 30 to 60 days when the foundation work is done first.
Why ChatGPT ignores most pages
ChatGPT does not rank pages the way Google does. It evaluates whether a page can be trusted, parsed, and extracted into a clean answer.
A 2026 Rank Math study found only around 10% of Google’s top-ranking results overlap with ChatGPT citations. The retrieval rules are different. The signals are different. The judgments are different.
If your site has not been chosen, the problem is almost never raw quality. It is structural.
We have audited hundreds of sites for AI visibility. The same four failures appear on most of them.
Failure one: entity drift
ChatGPT and other AI engines build internal representations of entities. The model needs to know what your company is, what it sells, who it serves, and how it differs.
Most sites send conflicting signals. The homepage says one thing. The Schema says another. The About page describes a third version of the business. Customer mentions on Reddit and review sites describe a fourth.
The model cannot resolve the contradiction. So it skips the page.
Entity drift is the most common failure we find. It is also the cheapest to fix. The diagnostic isolates the inconsistencies in the first week.
Failure two: retrieval-shape mismatch
AI systems extract answers in specific shapes. A direct question gets a direct answer in one to three sentences. A comparison gets a parallel structure. A how-to gets a numbered sequence.
Most sites bury the answer in paragraph form. The first 30% of a page carries 44.2% of all LLM citations, according to a Wellows study in 2026. If your answer is not in the first third, in the right shape, it does not get cited.
This is not a writing problem. It is an information architecture problem.
Section headers must carry the question. The first sentence under each header must carry the answer. Supporting evidence comes next.
Pages built this way show up. Pages that meander do not.
Failure three: crawler routing collapse
OpenAI runs three crawlers that matter for ChatGPT visibility. GPTBot collects training data. OAI-SearchBot indexes content for ChatGPT search. ChatGPT-User fetches pages live during user conversations.
Most sites either block all three by accident or pass conflicting signals through robots.txt, llms.txt, and the meta robots directives.
If you block GPTBot, you remove yourself from the training corpus. If you block OAI-SearchBot, you remove yourself from the retrieval index. If you block ChatGPT-User, you cannot be fetched during a live answer.
These are not the same crawler. They serve different purposes. We see at least one of the three blocked on most diagnostic audits.
llms.txt is a separate question. As of 2026, no AI provider has confirmed it is read. Our position is documented in the diagnostic: useful as a manifest, not a substitute for crawler access.
Failure four: attribution gaps
You cannot fix what you cannot see. Most sites cannot tell the difference between a ChatGPT-driven visit and a direct visit.
ChatGPT often strips the referrer header. Mobile and Brave browser sessions arrive looking like direct traffic. Semrush, Ahrefs, and GA4 each report different numbers for the same site.
The fix is a custom channel group in GA4 with a regex filter for chatgpt.com, openai.com, perplexity.ai, and the other AI source domains. The channel must sit above the standard Referral channel in your channel order, or AI visits get reclassified as organic search.
Without this, the diagnostic cannot measure recovery. Every recommendation we make starts here.
What the Intelitune diagnostic checks
Our diagnostic runs across ChatGPT, Perplexity, Gemini, Claude, and Copilot. It checks every page for the four failures above.
It returns a written report in three business days. It does not require a sales call. It does not produce a 150-issue dump.
Intelitune’s senior team leads strategy on every engagement. Specialists execute. The senior team reviews the work and stays in every decision that matters.
The output is a priority list. The first decisions are usually pruning decisions.
Anchor case: Avenue Cookery School
Avenue Cookery School came to us invisible across ChatGPT. The Schema was partial. The page structure buried answers under prose. The crawler routing blocked OAI-SearchBot through a stale robots.txt.
We ran the diagnostic. Four failures, all present. The team fixed them in sequence over 30 days.
By day 30, Avenue ranked first in the AI visibility benchmark for cooking schools. ChatGPT visibility hit 96.6%. Active users grew 81.3%. Key events grew 173.9%. Purchases grew 62.9%.
The pipeline outcome was GBP 2,262 in revenue attributed directly to ChatGPT in a single month.
Visibility isn’t the win. Behaviour is. Pipeline is the proof.
Where to start
If you cannot see yourself in ChatGPT, the question is not whether you should publish more. It is which of the four failures is blocking retrieval.
The diagnostic finds out. The fix follows.
Saqib Junaid
Co-Founder & CTO, AI SEO
Saqib Junaid is co-founder and CTO of Intelitune, the AI SEO agency. He leads technical strategy across AEO, GEO, and LLM SEO — schema architecture, crawler routing, retrieval engineering. Saqib has audited AI visibility across luxury travel, culinary education, B2B research, and professional services. He works with founders at $2M to $50M premium service brands and scaling B2B SaaS who need senior interpretation, not a 150-issue tool dump.
- Previous
- Next
