Can You Actually Measure AI Visibility? Probably Not Yet.

May 2026, and the question still hangs: can you track whether an AI model mentions your brand? The honest answer is roughly, sometimes, with serious caveats. SparkToro dedicated an entire Office Hours session to this problem in March, and the conclusion was less than satisfying. Current tools infer visibility from prompt-response sampling, not from any deterministic index. You are working with inference, not measurement.

The Decision: Should You Shift Budget from SEO to AI Visibility?

No. Not as a swap. The temptation is understandable. Google's AI Overviews now surface more links than they did twelve months ago. That +47% growth in citations makes it feel like a new channel is crystallizing. Meanwhile, Practical Ecommerce's latest analysis urges brands not to abandon traditional search tactics that still drive revenue. The calibrated move is to layer new tactics on top of proven ones. Treat AI visibility as an additive experiment, not a replacement for what already converts.

Why Measurement Remains the Core Problem

Consider what tracking AI visibility actually requires. You need to know which prompts users type into ChatGPT, Claude, Gemini, and Perplexity. You need to know how often your brand appears in responses. You need to know whether those mentions drive any downstream action. Today, none of these data points are reliably available at scale. Models hallucinate brand names, rotate sources based on token cost and context window, and produce different outputs for nearly identical prompts. The latency between a user seeing your brand in an AI response and visiting your site is invisible to most analytics stacks. SparkToro's session made this plain: the tools that claim to track AI mentions are running sampled queries and extrapolating. That is not the same as measurement.

The Right Decision: Build for Discoverability Without Demanding Attribution

This sounds uncomfortable. It should. But it is probably correct. The brands that will benefit most from AI-driven discovery in 2027 are the ones structuring their content and data now. Structured data markup. Comprehensive FAQ coverage. Topical depth that gives models unambiguous context about what your brand sells and why it matters. These are not speculative investments. They also improve traditional search performance, rich snippet eligibility, and programmatic ad relevance. The new ecommerce tools launching this month lean hard into predictive AI and autonomous marketing features. Most of them still require clean, well-structured product and content data to function. You are not building for one channel. You are building a foundation that reduces vendor lock-in across channels.

Implementation: Four Moves This Quarter

First, audit your structured data coverage. Check every product page, category page, and help article for schema markup. If coverage is below 80%, fix that before anything else. Second, identify your top 50 purchase-intent queries and evaluate whether your content answers them in a format an AI model could parse cleanly. Short, declarative paragraphs outperform long narrative blocks in model retrieval. Third, set up a manual eval cadence. Once per month, run your top queries through ChatGPT, Claude, Gemini, and Perplexity. Log whether your brand appears. This is crude. It is also the most honest measurement available right now. Fourth, do not cut your existing SEO or paid search spend to fund AI visibility experiments. Layer the work. The audience research principles SparkToro advocates apply here: understand where your buyers actually spend attention before reallocating budget based on hype cycles.

What I Am Not Sure About

I remain uncertain whether AI Overviews will consolidate link clicks or fragment them further. The current data suggests more links appearing, but click-through behavior on those links is poorly documented. If Google or a major model provider opens a verified analytics API for brand mentions, my stance shifts. That would turn inference into measurement and justify dedicated budget. Until then, this is a hedging game.

Three Questions to Pressure-Test Your Position

1. What percentage of your product pages currently carry complete structured data markup, and when was the last time someone verified it? 2. If you ran your top 20 purchase-intent queries through four AI models today, how many times would your brand appear versus your closest competitor? 3. How much of your current SEO budget is producing measurable revenue, and would you be comfortable defending that number to your board before moving any of it to an unmeasurable channel?