“What Is AI Saying About My Business?” Is the Wrong Question

TL;DR: Every AI platform rewrites your question and personalizes the answer to the person asking. “What is AI saying about my business?” changes by user, model, and day. Track the specific questions your buyers ask, daily, across platforms. One snapshot from your own account is not a measurement.

Last updated: July 2026

This piece came out of Episode 2 of To the Power of X, Xponent21’s podcast on AI and marketing. Will Melton, Chuck McCarthy, and I spent most of that conversation on why this question really shouldn’t have a blanket answer, and what a real answer requires.

Why AI Gives Different Answers to the Same Question

Two people can ask the same AI the same question on the same day and get different answers. That is not a malfunction – it’s actually how the systems are built.

ChatGPT search does not send your words to the web as you typed them. According to OpenAI’s documentation, it typically rewrites your query into one or more targeted searches. When Memory is turned on, it also folds in whatever it has stored about you, along with details like your general location.

Google runs a more elaborate version of the same process. Search Central describes a query fan-out technique behind AI Overviews and AI Mode. The system issues multiple related searches across subtopics and data sources before it assembles a single response.

Will Melton has been watching this mechanism through the early research of the Cognitive AI Ranking Lab (CARL Lab). Describing how AI constructs a response, he said (11:03), “one of the very first things that happens is it stops and says, ‘All right, this question is asked by a non-expert. So it’s not the best question to answer. So let’s come up with a shadow prompt that we really should answer.'” The shadow prompt, as Will calls it, is what the platforms document as query rewriting.

How ChatGPT, Gemini, Perplexity, and Claude Personalize Their Answers

Rewriting is the first layer. Personalization is the second, and each major platform reads a different slice of who you are.

• ChatGPT personalizes on your general location and, if Memory is enabled, whatever it contains (OpenAI Help Center).
• Gemini can draw on what Google calls Personal Intelligence, including Gmail, Photos, Search history, YouTube, and your past chats (Google, January 2026).
• Perplexity retrieves a memory store of interests and past conversations, along with an AI Profile that can include your occupation and format preferences (Perplexity, November 2025).
• Claude carries cross-conversation memory, scoped by project, with incognito chats excluded (Anthropic).

So the variation runs in two directions: Two people asking one model get different answers because the model knows different things about each of them. One person asking four models gets four different answers because each platform reads a different part of the same user.

Will saw the first kind play out (5:31) with a prospect in advanced manufacturing. He assumed the prospect had asked ChatGPT about agencies that do AI search engine optimization (AI SEO) for that industry. That wasn’t actually the case – the lead had asked which agencies do AI SEO, and because the model knew he worked in manufacturing, it told him Xponent21 works with manufacturers. The prospect never typed the word that shaped his answer.

What AI Visibility Tools Actually Capture

The tools in this market mostly work the same way. Send a prompt, record what comes back, report the result as your AI visibility. That capture shows what one model said to one synthetic user at one specific moment. It has the same evidentiary weight as Googling yourself once in incognito mode and calling the result SEO data.

These tools can fool many users, but top marketers are noticing the discrepancies. In May 2026, Digiday reported growing skepticism toward expensive AI visibility tools. Heather Physioc, chief discoverability officer at VML, described many of them as delivering point-in-time results rather than ongoing measurement. Paul Dyer, chief executive of /prompt, said that if you use three different tools and give them the same prompts, you get three different answers. 

The tools are not lying. They report exactly what one model told one synthetic user, once. That is real data, but it is not a measurement of what AI is saying about your business, because personalization guarantees a different user gets a different answer.

The Better Questions to Ask About Your AI Visibility

The fix for this large, vague question is to get more narrow. 4 ways of thinking can actually help you get to an answer. 

What Is AI Saying to the Buyer You Need to Reach?

Personalization means that your brand, the optimal answer to a user’s query, is going to be highly varied. A check from your own account measures your account, not your market. Consider the variety of the market you’re in and think about the content that would impact the people who actually fit your buyer personas. 

Does AI Recommend You for Your Most Valuable Queries?

At Xponent21 we have a name for the questions your best customer asks right before they buy. We call them Most Valuable Queries, or MVQs. An MVQ is the full natural-language question a real person types into a large language model (LLM) at the moment of decision. Tracking your brand against generic prompts tells you little. Tracking it against your MVQs tells you whether you show up when it counts.

How Is the AI Answer About Your Brand Changing Over Time?

Models update and sources shift. Competitors keep publishing. An answer that included you in March can drop you by July, and a one-time check in on that visibility will likely miss that occurrence. 

What Sources Is AI Using to Describe Your Business?

This is where visibility turns into strategy. My colleague Garry Callis distinguishes between a mention, a citation, and a recommendation. The sources feeding the model are what move a brand up those tiers. As I said in the podcast, (5:02), “the key part there is to be giving it good things to say.” Measurement tells you whether the good things are landing.

What Accurate AI Visibility Measurement Requires

Answering those four questions takes a different type of instrument than many SEO pros are used to. These tools require prompt-level tracking, so the data maps to real questions. It requires daily capture, so change becomes visible. It requires multiple models, because your buyers are spread across platforms that each personalize differently. And it requires a consistent methodology, so a difference in the data reflects a difference in the models rather than a difference in the check.

In the episode, Will described the premise behind building that instrument (2:30). “Every single time we prompt large language models, we get a response. And every single one of those responses can be data.”

That premise became the CARL Lab’s research program, and the research became a product. CARL Intelligence tracks prompts with a daily capture, anchored to Coordinated Universal Time (UTC), across ChatGPT, Claude, Gemini, and Perplexity. The design exists because a single-prompt snapshot is not a measurement. In our experience, the brands that get real value from AI visibility data are the ones watching specific prompts move over weeks, not the ones holding onto a screenshot of a personalized occurrence.

If you want to see what that looks like against your own questions, CARL Intelligence offers a free 90-day trial at intel.xponent21.com. It includes 50 tracked prompts across ChatGPT, Claude, Gemini, and Perplexity, and no credit card is required. Start with the question your best customer asks right before they buy. If you would rather have someone interpret the data with you, book a discovery call.

Frequently Asked Questions About AI Visibility

What is AI saying about my business?

There is no single answer, and that is by design. AI platforms rewrite your question before answering it, personalize the response based on what they know about the asker, and change as their models and sources update. The answer varies by person, platform, and day. The useful version of the question names a specific audience and a specific prompt, then tracks the answer over time.

Why does ChatGPT give different answers to the same question?

According to OpenAI’s documentation, ChatGPT search typically rewrites a query into one or more targeted searches before answering, and it folds in context like the user’s general location and stored Memory. Two users typing identical words are effectively sending different queries with these factors considered, so they receive different answers.

Why do I get different AI answers than my colleague?

Each platform personalizes its own signals. ChatGPT uses general location and Memory. Gemini can draw on Gmail, Photos, Search history, and YouTube through Personal Intelligence. Perplexity retrieves stored preferences and an AI Profile. Claude carries project-scoped cross-conversation memory. The model is answering you specifically, not the question in the abstract.

How do I track my AI visibility accurately?

Track specific prompts rather than general impressions. Choose the questions your real buyers ask, capture the answers daily across multiple models, and hold the methodology constant so a change in the data reflects a change in the models. Evidence points to longitudinal, multi-model tracking as the only way to separate a real shift from ordinary variation.

What is AI visibility measurement?

AI visibility measurement is the practice of tracking how AI systems such as ChatGPT, Claude, Gemini, and Perplexity describe, cite, and recommend a brand in their answers. Done well, it is prompt-level and longitudinal. It distinguishes between being mentioned, being cited as a source, and being recommended as the answer. Those three outcomes carry very different business value.

For a full breakdown on this subject, watch To The Power of X, Episode 2.