Your marketing director shows you the dashboard: organic traffic holds, Google positions hold, the site is fast. And last week a buyer told you, in a meeting, that he had asked ChatGPT who makes what you make, and your company did not come up. Both things are true at the same time. What follows explains why, and the reason is not where your team is looking.
The dashboard that no longer measures the right question
For fifteen years there was one question: where do I rank when someone searches for my product? That question still has an answer, and your team answers it well. The problem is that a growing share of buyers stopped asking it that way.
Semrush tracked more than ten million keywords through 2025: Google AI Overviews appeared on 6.49% of searches in January, peaked at 24.61% in July and closed November at 15.69%. Google decides, search by search, when to answer itself instead of showing ten links. Source: Semrush, AI Overviews Study, December 15, 2025 update, checked October 25, 2026.
When Google answers, it cites two or three sources. When ChatGPT, Gemini or Perplexity answer, they cite two or three sources. If your site is not among them, the buyer reads someone else's name and never reaches the link where you still rank first. That is the symptom. The cause is less comfortable.
What an AI assistant needs in order to cite you, and almost no corporation provides
An AI assistant does not read your site the way a customer does. It crawls it with a robot that has a name: GPTBot, ClaudeBot, PerplexityBot, Google-Extended. Then it looks for a direct answer to the question it was asked, and only then does it decide whether your page deserves the citation. Three steps, and a typical corporate site fails the first one without anyone noticing.
- The robots.txt file shuts those robots out, sometimes through a rule written years ago to protect the content.
- The catalog lives in PDF: hundreds of technical sheets that search and AI see as blank pages.
- No structured data: the site tells no machine what the company is, what it sells or where it operates.
- No llms.txt, the file that summarizes a site for language models. Rankability measured it on September 18, 2026: only 93 of the 1,000 most visited domains in the world publish one, 9.3%. Source checked October 25, 2026.
You have seen the symptom and the closed doors. What is missing is the part that decides who wins, and it is the one that surprises executives most when they see it with data.
The competitor the AI is citing is not the one you think
On November 13, 2025 Conductor published its benchmark study covering 13,770 enterprise domains and 3.3 billion sessions. Two figures matter here. Referral traffic arriving from AI assistants is still 1.08% of the total, and 87.4% of it comes from ChatGPT. Source: Conductor release on Business Wire, checked October 25, 2026.
Read it carefully. Traffic from AI is small today. The citation is not: the buyer who asks an AI is at the moment of deciding, and the answer carries a name. In my readings by question and by market, the name that appears is rarely the category leader on Google. It is usually a mid-sized company that published buying guides, put its sheets in HTML and opened the door to the robots. Nobody at the corporation considers it competition. The AI does.
What gets measured before a single line of the site is touched
This is where everything above was heading. AI visibility is not fixed with a tool or a campaign: it is diagnosed. Question by question, market by market, language by language: whom does ChatGPT cite, whom does Gemini cite, whom does Perplexity cite, whom does the Google AI Overview cite, and why them? With that table in hand the plan stops being a list of best practices and becomes a sequence with an owner and a metric.
In the illustrative case of my sample deliverable, an industrial manufacturer selling in the United States appeared in none of the twelve buying questions measured, and a single competitor appeared in nine. That figure is a sample in the sample; in your case it comes from your data. It is the first number of the analysis and the one that defines everything else.
Does this apply if my site already has good SEO?
Yes. SEO decides who appears in the links; the AI citation decides who appears in the answer. They are two different readings, and the second has almost never been measured at a corporation.
How long until a change shows?
It depends on the number of markets and on how closed the site is today. What is immediate is knowing where you stand: the analysis delivers the table of citations by question and market.
Can I measure it myself?
You can start: type your site address into the request form on this site and within seconds you will see eight signals, among them whether AI crawlers can get in and whether an llms.txt exists.
