Something shifted in 2024 and most marketers still haven't clocked it. Consumers stopped Googling first. They started asking. They typed "best CRM for a small law firm" into ChatGPT. They asked Claude "who makes the best running shoes for wide feet." They asked Perplexity "which accounting software do most independent consultants use."

And for the vast majority of brands, the answer came back — and their name wasn't in it.

This is the new dark funnel. It doesn't show up in your GA4. It doesn't show in your search console impressions. There are no click-throughs, no UTM parameters, no referral sessions. Just a customer who asked an AI and got someone else's name.

Why standard analytics won't help you here

When someone finds you via Google, they click a result. That click generates a session in your analytics. When someone finds you via ChatGPT, there is no click — at least not initially. The recommendation happens inside the model's response. The user reads it, forms an impression, and might search for you directly later. That shows up as direct traffic, which tells you nothing about what triggered it.

There's a second problem: AI responses aren't deterministic. Ask Claude "best project management tools for agencies" three times and you may get three slightly different answers with different brand inclusions. This means a single manual check can be misleading. You need systematic, repeatable measurement.

The five methods, ranked by signal quality

1. Manual prompt auditing — the right way

Free · High effort · Inconsistent at scale

Effective manual auditing means testing category queries the way a real customer would phrase them — with no brand name at all. These are the prompts where you either show up or you don't.

Run each prompt in ChatGPT (GPT-4o), Claude (Sonnet), and Perplexity. Record verbatim whether your brand appears, where in the list, and what the surrounding language is. Positive framing, negative framing, or neutral citation all matter.

The limitation: you can realistically test maybe 50 prompts a week manually. The actual universe of queries where your brand could be mentioned is thousands.

2. Blind probe testing — the most honest signal

Free · Medium effort · High diagnostic value

A branded query tests recognition ("does the AI know about us?"). A blind probe tests recommendation ("does the AI surface us without prompting?"). The second one is what actually matters for acquisition.

When you get your name back in a blind probe response, that's genuine Share of Prompt — the AI chose to surface you. When you only appear in branded queries, you have awareness but not recommendation authority.

3. Perplexity as a proxy tracker

Free · Low effort · Limited but fast

Perplexity is uniquely useful for brand monitoring because it shows you its sources. When you run a category query and your brand appears in the response, you can see which URLs Perplexity cited to construct that answer. This tells you what specific content is driving your AI visibility.

Limitation: Perplexity's citation behavior doesn't perfectly predict ChatGPT or Claude behavior. Those models use different training data and retrieval architectures.

4. Automated GEO monitoring platforms

Paid · Low effort · Scalable

At some point — and that point is earlier than most people think — manual tracking doesn't scale. The query universe is too large, the models update too frequently, and the variance in AI responses means you need statistical coverage, not spot checks.

PlatformModels CoveredBlind ProbesCitation TrackingCompetitor ASOVPrompt Taxonomy
ClarifyHQChatGPT, Claude, Perplexity, Gemini
ConductorPrimarily Perplexity + web AI
SE RankingChatGPT, Perplexity
BrightEdgePerplexity, Copilot
Semrush AI ToolkitChatGPT, Perplexity

The differentiator to look for in any platform is blind probe architecture. Most tools run queries that include your brand name in the system context — which isn't a fair test of organic recommendation.

5. Indirect signal triangulation

Free · Passive · Weak but additive

The prompt taxonomy you need before you start measuring

Before you build your tracking system, you need a prompt taxonomy that maps to real customer behavior across three intent layers:

Aim for at least 50 prompts across these three layers before you begin systematic tracking.

What to actually do when you find a gap

Not appearing at all? This is an entity establishment problem. Focus on FAQ-structured content, comparison pages, and definition content. Schema markup helps retrieval-augmented models find and surface them faster.

Appearing with inaccurate framing? This is a brand voice problem. Flood the context with better primary sources: updated website copy, authoritative long-form content, and aligned third-party mentions.

Competitors outranking you? This is a Share of Prompt problem. Study what competitors are being cited for, what content they have that you don't, and build better, more citable versions.

Find out where your brand stands in 60 seconds. Run your free AI visibility scan — ClarifyHQ tests against ChatGPT, Claude, and Perplexity with no brand name in the query, and tells you exactly where you're winning, losing, and missing.