What it costs your brand when AI doesn’t name it

ChatGPT has 900 million weekly users (OpenAI, February 2026). Google’s AI Mode has passed 1 billion monthly users (Alphabet, Q2 2026), and it has worked in Spanish since September 23, 2025 in Hispanic America and since October 8, 2025 in Spain. Meta AI answers inside WhatsApp, Instagram and Facebook.

What changed isn’t just the channel; it is the format of the answer. A traditional search engine returns ten links and the person compares them. An AI engine returns a paragraph, “For a small business in Mexico, the most recommended options are…”, with a few names and a reason for each. The machine has already done the comparing. If your brand isn’t in that paragraph, you didn’t come in tenth: you don’t exist for that decision.

And the questions coming in are buying questions: “which health plan is best for a family in Córdoba?”, “which car insurance do you recommend in Spain?”, “which air conditioner brand uses the least power in Mexico?” They are category, comparison and purchase-intent questions, the same ones any salesperson would love to hear.

All three examples are fictional: they show the mechanism and don’t describe any client or any real brand.

Each case is measured differently. The first is a visibility problem; the second, a problem of share of voice and position against competitors; the third, a problem of perception: which adjectives, facts and traits AI associates with your brand. A report that only says whether you were mentioned sees the first case and misses the other two.

This cost leaves no trace. An AI answer that doesn’t name you generates no visit, and often neither does one that does: the person reads the recommendation and then searches for the brand by name, or goes straight to the store. There is no click to record and no drop to explain.

• Your analytics sees clicks, not answers. Even if Google Analytics 4 groups traffic from AI assistants into its own channel, that number only counts the people who clicked. It doesn’t count the people who read an answer you weren’t in, and those are exactly the ones you lost.
• Your Google ranking doesn’t predict it. Chat engines answer with what the model learned, not with your ranking.
• Your site is a small part of what AI cites. In Refractia’s data, a brand’s own site accounts for between 0.5% and 2.8% of what the engines cite when they talk about it. The rest is media and other third-party sites: what AI says about you is built mostly outside your site.
• One test is not a measurement. The same question can get different answers depending on the country and the engine, and ChatGPT doesn’t answer the same way Gemini or Meta AI does. Asking once from your own computer shows you one answer, not your situation.

Answer each one with “I know” or “I don’t know.” Doing the exercise takes no tools; answering it with data does.

1. Do you know what ChatGPT, Claude, Gemini, Google and Meta AI answer when someone asks for the best option in your category in your country?
2. Do you know which competitor they name first, and in how many of those answers it appears compared with you?
3. Does the way AI describes your brand (price, quality, coverage, service) match the one you are aiming for, and is it up to date?
4. Do you know which sites the engines cite when they talk about your category, and whether any of them name you or link to you?
5. Do you know whether AI crawlers can read your site? A CDN or firewall can block them even when your robots.txt allows them.

Every “I don’t know” is a question your customers already ask AI, and today you don’t know the answer. The first two are the starting point of the free diagnostic. The platform answers the other three: AI Perception the third, Influence Sources the fourth, and the Technical Workshop, which detects when a CDN or firewall blocks AI crawlers, the fifth.

1. 01

   Measure before you change anything

   Write down the questions a customer would ask, in their language and naming the country, and ask them to all five engines. Save the answers with a date. Without that baseline, you won’t know whether what you do next worked.

2. 02

   Separate chat from AI search

   ChatGPT, Claude or Gemini in chat mode answer with what the model knows; Google AI Overviews summarize what the web ranks. A brand can be in one and missing from the other, and what to do about it differs in each case.

3. 03

   Check that AI can read your site

   Confirm that your CDN or firewall doesn’t block AI crawlers. Don’t start with special files: Google says no llms.txt or special markup is needed to appear in its AI features.

4. 04

   Work where AI reads

   If your site is a small minority of what gets cited, look at which media and third-party sites show up in the answers for your category, and whether they name you. That is where much of what AI repeats about you takes shape.

5. 05

   Correct what is described wrong

   If AI repeats an old price or a trait that is no longer yours, update it on your site and in the sources the engines cite, and measure again with the same questions.