A framework for measuring AI visibility

The question I get asked most in 2026 is: How do we measure this?

• How do we measure whether our brand is showing up in ChatGPT? 
• How do we measure whether Perplexity is recommending us? 
• How do we measure whether the work we did last quarter on grounding for AI Mode moved the needle?

Nobody has solved this.

Anyone selling you a clean dashboard for tracking presence in grounding, visibility in display, or action at won across search, assistive, and agent simultaneously is selling you a snapshot view that amounts to a bad best guess.

The standard advice is “track these queries that we think people might ask,” or “track these queries that are a best-guess adaptation of search keywords.” 

That advice is unhelpful because prebuilt keyword lists pick queries that are easy to track, map to existing marketing efforts, or would be ideal if the audience were predictable. 

The visibility question is right. The precise-number answer it expects is wrong.

The measurement question, as the industry currently frames it, uses the wrong reference discipline. Brands still hunting for the perfect AI-era visibility KPI are hunting for something that doesn’t exist and never will.

The right answer is a methodology that takes its discipline from how economists measure systems too complex and opaque to measure precisely. My methodology is the Funnel Query Pathway, and it does more than measurement. It’s one operational artifact that does three jobs simultaneously: strategy, measurement, and analysis.

Marketers want a number on a dashboard, tracking week over week, tied to a specific query on a specific engine for any user, the way search delivered for 20 years. Search could deliver that number because the surface was finite, the rankings were stable, the click was measurable, and the journey was observable. Assistive and agential surfaces deliver none of that.

We’re operating in a new environment now, and that environment forces us to ask different questions, measure different signals, and act on different proof.

Why AI visibility is a macro measurement problem

I studied economics and statistical analysis at Liverpool John Moores University, which is why the shape of this measurement problem looks familiar. The same shape shows up whenever a discipline that worked at one scale tries to operate at a scale where its instruments stop applying. 

Microeconomics versus macroeconomics is the canonical case. The corner shop measures inventory precisely, the central bank can’t measure inflation precisely, and both disciplines are correct at their scales. Neither discipline’s instruments work in the other’s environment. The discipline I’m proposing isn’t macroeconomics applied to brands. It’s the macro instinct applied to AI-era brand measurement.

AI surfaces are macro for the same three structural reasons macroeconomics had to develop its own discipline. 

The first is opacity. The system’s internal state isn’t observable, the way central banks can’t observe every transaction and modern LLMs can’t expose why they decided what they decided. 

I call this brand-user-algorithm (BUA) opacity. The user can’t see the alternatives the algorithm rejected, the brand can’t see the journey within the walled garden, and the algorithm can’t fully introspect on why it decided what it did.

The second reason is personalization, the AI-era equivalent of heterogeneous agents: Each user gets a different answer because the engine factors in different context.

The third is the explosion of possibilities, and the explosion isn’t just across the seven engines. The surfaces now include apps (Copilot in Word, ChatGPT inside Slack, Perplexity in Comet), operating systems (Copilot baked into Windows, Apple Intelligence in macOS and iOS), and hardware (Lenovo Copilot+ laptops with a dedicated Copilot key, Samsung Galaxy AI on the phone, and Meta Ray-Bans on your face). 

Ambient research becomes a major entry mode. The AI surfaces a recommendation unprompted because it understands the context. 

That’s where the funnel query pathway lives. Importantly, it isn’t an evolution of keyword mapping or a pimped-up intent-based methodology. Because it looks at the macro level, it’s a fundamentally different beast.

The unit of measurement is a cohort

Most practitioners running keyword campaigns think they’re grouping queries by intent, but more often than not, they’re grouping by category, which isn’t the same thing as intent. A typical Google Ads campaign would place every Phuket hotel query into one ad group, with the implicit logic that “Phuket hotels” is a logical intent group. It isn’t.

“Phuket hotels” defines the destination. The buyer behind “5-star hotels in Phuket” and the buyer behind “cheap hotels in Phuket” share a destination and have almost nothing else in common: different budgets, decision criteria, conversion paths, and downstream behavior. Grouping them produces an ad group whose performance averages across two cohorts that should never have been combined.

Categories group things. Cohorts group people.

Intent is about people, not things. Google engineers tell me this is the most common mistake they see in AI Max and Performance Max campaigns because the algorithm routing a prospect doesn’t ask, “What category is this query in?” It asks, “What cohort does this user belong to, with what intent?”

The intersection of cohort and intent defines the node

A cohort is a group of people who’ll behave in a similar way given a specific stimulus. XL men, luxury travelers, and parents shopping for kids. Each is a cohort, defined by some durable identity that persists across time and context. The XL man is still an XL man when he’s buying winter coats in November, a vacation in July, and a wedding ring in March.

An intent is the situational vector that crosses through the cohort at a moment in time. Buying a shirt, booking a hotel for next month, and kitting out a child for summer. Each is an intent, and each one spans many cohorts. Buying a shirt pulls in XL men, S men, women, and parents shopping for kids, all walking different paths to different brands at different price points.

Every cohort carries many intents across a lifetime, and the same intent spans many cohorts across the market. The intersection of cohort and intent is what defines a node in the Funnel Query Pathway tree. XL men buying a shirt in winter is a node. Luxury travelers booking a hotel for next month is a node. Parents shopping for kids’ shorts for summer is a node.

Importantly, cohort alone doesn’t work because XL men buying pajamas behave differently from XL men buying office shirts or holidays. Intent alone won’t track because luxury travelers booking Bali behave differently from budget travelers booking Bali. The intersection is where behavioral coherence lives, and behavioral coherence is what makes the node trackable in the opaque AI surfaces we’re working with.

The query qualifies for tracking when both cohort and intent are legible in it

The test for whether a query belongs in a funnel query pathway tree is whether both cohort and intent are legible in the query itself. “Men’s red shirt from Uniqlo” surfaces a man shopping for clothes (the cohort) and buying a red shirt at the buying moment (the intent), with the brand named as the commercial destination. Both axes are legible.

“Hotels in Bali” surfaces an intent but hides the cohort (luxury, business, budget, honeymoon, family, backpacker), which is why it can’t function as a node. The people submitting it will behave nothing alike as they work their way down the funnel. Narrow it to “cheap hotels in Bali,” and the budget cohort emerges alongside the intent, and the query qualifies for the funnel query pathway.

The test is behavioral coherence, not specificity. If both axes are clear, it’s a node. If not, narrow it until they are, and you’ll discover the cohort and intent that together make sense to your business.

Build the funnel query pathway from the conversion moment upward

The funnel query pathway doesn’t track what users actually type. It tracks what the cohort would ask given the intent. Every query in the tree is a theoretical representative of cohort behavior at the buying moment, not an empirical record of individual users.

This is the macro discipline in practice. We don’t research search volume for these queries because they aren’t necessarily queries anyone has typed. We construct them by reasoning forward from cohort plus intent, building the ideal pathway a representative member of the cohort would walk.

The “would” carries the entire methodology, and the moment you slip into thinking about what users “actually” type, you’ve collapsed back into the micro instinct the methodology was designed to escape.

Once a query passes the test, it’s your starting point. The funnel query pathway (branching tree) builds upward from there. This mirrors the funnel flip at the query level. AI-era acquisition starts at the conversion moment and projects upward because the algorithm forward-calculates the conversion path from intent, not from awareness.

Start with the ideal branded BOFU query for one cohort with one intent, then project upward through the evaluation questions that cohort would ask, then upward again through the awareness questions that would come even earlier.

Example: Building one funnel query pathway tree from a single Uniqlo query

Take Uniqlo as the brand and “men shopping for clothes” as the cohort. The intent is the situational vector that defines the buying moment, and different intents inside the same cohort produce different trees: men buying a shirt, men buying winter outerwear, and men buying gym kit. Each is a node.

Start with one. For example, pick the intent of buying a red shirt, which I do often. The branded bottom-of-funnel query that fits the cohort-intent intersection is “men’s red shirt from Uniqlo.” That’s the conversion node.

Five to 10 variations of similarly shaped queries fit the same intersection and don’t need to be tracked individually: “men’s Uniqlo Oxford shirt,” “Uniqlo men’s smart shirt,” “men’s red dress shirt Uniqlo,” and “Uniqlo men’s casual red shirt.” Each is the same cohort with the same intent landing on the same brand. Pick the one that’s most useful for your business. Build upward.

Next, find the middle-of-funnel branches that would land at your ideal BOFU query. In our example, “men’s red shirt from Uniqlo,” we’re looking for the evaluation queries the same man would ask the engine before arriving at the branded buying moment. The cohort is still men shopping for clothes, the intent is still buying a red shirt, and the brand isn’t named yet because the cohort is still considering options:

• “Best red shirt for men”
• “Red shirt for office work”
• “Where to buy a quality red Oxford shirt”
• “Which red shirt looks best with chinos”
• “Affordable men’s red shirts that don’t fade”
• “Red shirts for men under €50”
• “Best affordable clothing brands for men”
• “Minimalist menswear brands with color ranges”
• “Where to buy quality basics for men online”
• “Best affordable men’s shirt brands”

Ten branches, all the same cohort, all the same intent, all logically routing to “men’s red shirt Uniqlo” as the ideal BOFU commercial query for the brand.

Top-of-funnel branches that would land at each of those middle-of-funnel queries are the broader awareness questions the same man would ask even earlier, before narrowing to specific shirt types or brands.

For “best red shirt for men”:

• “Can men wear red shirts to work”
• “How to add color to a man’s wardrobe”
• “Shirt color rules for office wear”
• “How many shirts should a man own”
• “Which shirt colors suit men with what skin tone”
• “What color clothing would make me stand out in a crowd”

That’s one 60-query funnel query pathway. I could’ve included 120 or more. That’s a choice, as we’ll see. As a rule of thumb, 60 is a reasonable number from a budget-versus-insights perspective. The point of the macro approach is that it doesn’t need you to go granular to measure.

The important thing here is that the 60 queries all route to one branded buying moment for one cohort with one intent. Do it again with another intent inside the same cohort (men buying winter outerwear, men buying office trousers), then another cohort (women shopping for clothes, with the intent of buying pajamas, branded BOFU “women’s pajamas Uniqlo”).

The tracking surface is a forest of trees, accumulated as the methodology runs.

---

AI routing uses the same math as Google Ads bidding

I discovered this while running keynotes and workshops for Google Marketing Live in Asia Pacific this month, in conversations with senior Google engineers about how Gemini routes recommendations. 

The math Gemini runs to decide which answer to surface next is the same math Google Ads has been running to decide which ad to serve next: forward-calculate the probability that this cohort, with this intent, lands at a conversion, and pick the path most likely to get them there.

Every practitioner who’s bid on a campaign in the last 15 years has been working with that probability calculation. For me, this is the most useful framing the funnel query pathway can inherit, because it explains why the cohort-with-intent unit aligns with the engine’s internal logic. 

The engine isn’t tracking categories or queries in isolation. It’s running a funnel pathway probability calculation on cohort plus intent. Every node you populate teaches the engine which path is the fastest way to get this user to the best solution to their problem.

Ads includes profit margin. Organic doesn’t.

The operational formula in Ads is cohort x intent x conversion rate x profit margin. Google holds all four because the advertiser provides Google with the commercial information needed to optimize bidding. The auction maximizes expected profit because Google has the inputs to calculate it.

The operational formula in organic is cohort + intent + conversion rate. Profit margin drops out because the engine doesn’t have the commercial information. The engine doesn’t know your gross margin on a red shirt versus your gross margin on pajamas, and it doesn’t optimize for your bottom line. It optimizes for user satisfaction, which is its own proxy for engine-level commercial outcome, but not for yours.

The principle holds across both surfaces: cohort + intent + conversion rate is the unit AI algorithms work with best. What differs is the precision of the conversion estimate. In organic, the conversion is inferred from behavioral patterns. In Ads, it’s measured from data provided by the advertiser.

Interestingly, the macro discipline operates in organic where micro precision isn’t available. Micro precision operates in Ads where it is. Luckily, the funnel query pathway tree works on both. Populate it once, and use it for organic content, Ads campaign structure, and analytical insights across both.

Build the funnel query pathway from the conversion moment upward

One terminological clarification in the 15-gate model I’ve built. The AI engine pipeline runs 10 binary gates:

• Discovered, selected, crawled, rendered, and indexed (DSCRI), which are handled by the bot, invisible to the algorithm.
• Annotated, recruited, grounded, displayed, and won (ARGDW), which are handled by the algorithm, invisible to the bot.

Our framework extends another five gates after being won: onboarded, performed, integrated, devoted, and codified (OPIDC), which are handled by post-transaction operations that serve people, invisible to both bot and algorithm. 

Fifteen gates total, each a binary checkpoint where the brand either survives or doesn’t.

Nobody inside the system sees the whole chain. Only the brand does. Won itself has three flavors depending on surface: 

• The imperfect click in traditional search.
• The perfect click in assistive engines.
• The agentic click in assistive agents.

The funnel sits on the display gate. The user’s journey from question to purchase moves through three phases at display — awareness, consideration, and decision. Phases are continuous human positions. Gates are binary machine checkpoints. 

The funnel query pathway tracks the queries the user submits across those three phases, with the branded buying-moment query landing at the decision phase that triggers won. Gates and phases aren’t synonyms, and conflating them breaks the methodology. 

Step 1: Start at the bottom of the funnel

Identify the queries your ideal customer profile (ICP) would ideally submit using your brand name at the moment they’re ready to buy. The emphasis is on “ideally.” 

Keyword research asks what people actually type. The funnel query pathway asks what the cohort with this intent would ideally ask the engine just before they purchase from you, with your brand name in the query. Branded, bottom-of-funnel, intent-confirmed, cohort-coherent.

Calibrate the specificity to the cohort definition. “Men’s red shirt from Uniqlo” fits the broad cohort of men shopping for clothes. “Men’s extra-large red shirt from Uniqlo” fits a sizing sub-cohort that behaves differently because size availability constrains the consideration set. Either is fine. Pick the cohort level where you want to operate, then operate consistently upward within the branches of your tree.

Generic keyword research won’t surface these queries because keyword tools optimize for volume, and cohort-with-intent queries are usually low volume by design. You have to know your cohort well enough to write them down yourself. If you can’t write five, your ICP work needs more depth before this methodology will produce results that are actually useful to your business.

Step 2: Project the pathway upwards

Each bottom-of-funnel query branches into multiple middle-of-funnel queries (the evaluation questions the same cohort would ask before arriving at the buying moment), each of which branches into multiple top-of-funnel queries (the awareness questions that would come even earlier). 

Build out gradually, one bottom-of-funnel query at a time. The funnel flip operates at the query level: Generation starts at the conversion query and projects upward, rather than starting at top-of-funnel awareness and hoping the buyer arrives at conversion.

Granularity is cohorts x intents. Tracking is a budget call.

The question of how many trees to build has one answer: as many as the team can populate. The question of how many trees to track has one answer: as many as give you statistically meaningful data.

The starting unit is one cohort with one intent. Men shopping for clothes, with the intent of buying a red shirt. That’s one tree, around 60 queries.

Add intents inside the same cohort (XL men buying winter outerwear, office trousers, and gym kit). Add cohorts (XL women, parents). Cohorts times intents gives the tree count. The numbers scale with the budget:

Cohorts	Intents per cohort	Trees	Approx. queries
1	1	1	60
3	5	15	900
5	10	50	3,000
10	10	100	6,000

What changes with resolution is the precision of the diagnosis. Track three trees, and you have a low-resolution read on three cohort-with-intent intersections. Track 100, and you have a high-resolution read on most of your buying landscape. Both are defensible macro reads because macro is about defining your methodology and scope to reliably read direction and rate of change, rather than specific values.

This methodology means you can start small and build out. Start tracking three Funnel Query Pathways for your most profitable ICP this month, then add another next month. Group them, and you can compare like with like starting today using a macro approach that scales and survives over time.

Populate the tree, and you teach the engine the conversion path

The shaping mechanism is what makes the funnel query pathway more than a measurement methodology. The engine routes recommendations by predicting what comes next for the cohort with the intent. 

When the brand feeds the AI with content that builds logically structured funnel query pathways and answers each node, the engine learns the chain: 

• Which awareness questions belong to this cohort.
• Which evaluation questions follow them.
• Which branded buying-moment query is the conversion answer.

For obvious pathways (red shirts), the algorithms already have the pathways ingrained, but for less popular pathways, the engine has no opinion, and you have every opportunity to shape its perception. 

Since the engine is an active participant in the funnel alongside the user, it can form a predictive map, and the path it surfaces for any prospect in the cohort is the path the brand trained.

Shaping isn’t a side effect. It’s the compounding mechanism, and it means the brand stops competing for individual query rankings and starts engineering the inference paths the engine forward-calculates from. The competitor optimizing query by query is optimizing against a model the engine has already moved past.

The deeper move: Mapping the funnel query pathway into every webpage

The methodology can sit beside the website as a tracking document, and that works, but the deeper move is mapping the funnel query pathway into your strategy, both on-site and off-site.

Every node in every tree corresponds to a query the engine surfaces for the cohort. Every query needs a passage that answers it. Every page names the cohort it’s serving. Every passage names the intent that might bring the cohort there and clearly outlines the next step in the cohort’s conversion path. 

• Top-of-funnel pages route toward the evaluation pages. 
• Middle-of-funnel pages route toward the branded buying-moment pages. 
• Bottom-of-funnel pages close the conversion.

If you can align the content across your brand’s digital footprint to the forward-calculation logic the engine is already running — cohort, intent, awareness layer, evaluation layer, conversion layer — then when the engine forward-calculates the next step for any user in the cohort, the brand’s site is one of the few places that has the complete chain laid out, and the probability calculation tilts in your favor.

Build all the funnel query pathways for your ICP, and you’re teaching the machine exactly what the path looks like for every cohort-intent intersection you serve, while encouraging it to bring the subset of its users who are your ideal audience right to your door.

One framework for strategy, measurement, and analysis

The funnel query pathway does three jobs simultaneously: strategy, measurement, and analysis. 

• Strategy: You populate every node of the tree with content that proves the answer at that phase of the buying journey: awareness content at the top, evaluation content in the middle, and the branded conversion moment at the bottom. Stop running content generation as a calendar against a keyword list, and start engineering paths that represent your ICP’s buying journey.
• Measurement: You run the same funnel query pathways across the three modes (search, assistive, and agent) and the engines (Google, ChatGPT, Perplexity, Claude, Copilot, Siri, Alexa, etc.). You can’t track every surface those engines appear on (Copilot in Word, ChatGPT in Slack, Apple Intelligence in iOS, and Copilot+ on a Lenovo laptop are all closed contexts that don’t let you rank-track). But every surface runs the same underlying engine, so your tracking extrapolates to every surface each engine sits inside.
• Analysis: You can use the pattern of where the brand surfaces and where it doesn’t across the funnel query pathway, by mode and by engine, as the macro view you can rely on for a like-for-like comparison over time.

What you actually get from the funnel query pathway

Here’s what you actually get from running the funnel query pathway: a quarter-after-quarter read of whether AI is recommending your brand to the right people at the right moment. 

You see direction, momentum, and a record of what’s working. You build, you measure, you analyze, and you adjust. Then you do it again next quarter. The brands that start this discipline now will be the ones AI knows by name in three years.

Pick one cohort, the most strategically important if you have several. Pick one intent inside that cohort. Write five to 10 branded bottom-of-funnel queries that cohort-with-intent would ideally submit at the buying moment (“men’s red shirt from Uniqlo” in our example). 

Pick one and map upward: five to 15 middle-of-funnel queries that would land at it, then three to 10 top-of-funnel queries that would land at each of those. You now have one tree, somewhere between 50 and 200 queries.

Run strategy, measurement, and analysis on the funnel query pathway branches.

• Strategy: Do you have pages and passages that address each of the nodes? Fill the gaps.
• Measurement: Run the tree across engines and document where the brand surfaces.
• Analysis: Where are the gaps clustered, which node is weakest, and which engines are recruiting most consistently?

Build out the content that fills the gaps in your ICP funnel query pathways, and track that set of queries monthly. You’ll see results, and you’ll be able to measure them.

AI-era optimization is about defining your methodology, picking your ICP and tracking, and building and strategizing with a macro mindset, which is the subject of the next article in this series.

---

This is the 14th piece in my AI authority series. 

• Part 1, “Rand Fishkin proved AI recommendations are inconsistent – here’s why and how to fix it,” introduced cascading confidence. 
• Part 2, “AAO: Why assistive agent optimization is the next evolution of SEO,” named the discipline. 
• Part 3, “The AI engine pipeline: 10 gates that decide whether you win the recommendation,” mapped the full pipeline. 
• Part 4, “The five infrastructure gates behind crawl, render, and index,” walked through the infrastructure phase.
• Part 5, “5 competitive gates hidden inside ‘rank and display’,” covered the competitive phase.
• Part 6, “The entity home: The page that shapes how search, AI, and users see your brand,” mapped the raw material.
• Part 7, “The push layer returns: Why ‘publish and wait’ is half a strategy,” extended the entry model. 
• Part 8, “How AI decides what your content means and why it gets you wrong,” covered annotation — the last gate where you’re alone with the machine. 
• Part 9, “Why topical authority isn’t enough for AI search,” opened the competitive phase proper with topical ownership.
• Part 10, “The funnel flip: Why AI forces a bottom-up acquisition strategy,” named the process.
• Part 11, “The framing gap: Why AI can’t position your brand” exposed the gap between evidence and recommendation.
• Part 12, “The 10-gate AI search pipeline: Find where your content fails,” showed you how to find (and repair) your F grades in the AI engine pipeline.
• Part 13, “The delegation boundary: How AI decides which brands win,” mapped how delegation moves between user and engine across search, assistive, and agent modes.
• Up next: The micro-macro shift, the paradigm framework that names the structural change in measurement, analysis, and strategy that the AI era requires.