Your dashboard went blind: measuring AI presence (the dark funnel)

Alright, let me start with a number that should keep anyone in charge of marketing up at night. In Google AI Mode (that AI-generated answer mode), 93% of searches end with no click at all. The user asks, reads the ready-made answer right there, and leaves. And it's not just Google: adding up the whole search landscape, today almost 6 out of every 10 searches already generate no click whatsoever. Think with me about what this does to your work. Your entire life your scoreboard was traffic, ranking, clicks. All three lived off one thing: the user leaving search and going to your site. When they stop leaving, your whole scoreboard goes blind, even if every metric on it stays "green." This is the lesson that fixes your dashboard.

01The dark funnel: why the dashboard went blind

Let's call the problem what it is. Your SEO dashboard was built for a world where search was a bridge: the user searched, saw a list of links, and crossed over to your site. Everything you measured (ranking, impression, click, time on page) happened after they crossed the bridge. It was a lit path, from start to finish, and your Analytics stood still at the exit, noting who arrived.

AI knocked down the bridge. Now the user asks and gets the whole answer right there, without needing to cross anywhere. The decision to buy, to trust, to eliminate a competitor, all of that happens inside a private conversation with AI, one you have no access to. It's as if the negotiation moved into a closed room you can't enter and from which not even the audio comes out. This invisible piece of the journey is what the market calls the dark funnel: the part of the customer's decision that lives inside AI and never touches your CRM or your Analytics.

And here lies the trap that confuses so many good people. Your ranking can keep looking great while your business bleeds. Ranking well on Google no longer guarantees you show up in AI's answer, they're two different games. So the dashboard shows the stable position (old game, still measured) and hides that you've disappeared from the answers (new game, not measured). It's an honest instrument pointed at the wrong place, like a perfect thermometer stuck in your pocket while you want to know the room's temperature.

Notice this isn't the end of marketing, it's a change in what you measure. The money didn't disappear, the competition didn't end. It just changed rooms. And the first thing you need, before any tactic, is a new dashboard that can see the new room. Fair?

02The new dashboard: the KPIs that replace ranking

Here's the instrument swap. Five indicators that measure AI presence, and what each one answers. Don't memorize the acronym, understand the question each one asks, because that's the question that goes into your report.

• Share of Model (SoM). This is the KPI the market points to as the direct replacement for the old "share of voice." It measures: of all the times AI answers a question in your category, in how many does it cite or recommend you as the primary solution? It's your slice of the model's "mind." If AI thinks about your category, does it remember you?
• AI Share of Voice (AI SoV). Your slice of mentions among competitors, within AI's answers. Simple math: if in 100 relevant answers your brand appears in 28, your AI SoV is 28%. It answers: of the category's mention pie, how much is yours?
• Share of Answer (SoA). This one is finer and a lot of people confuse it with the ones above. It doesn't measure WHETHER you showed up, it measures HOW MUCH of the answer is about you. Are you the star of the answer or just a footnote cited in passing? It answers: do you own the answer or are you a bit player in it?
• Citation rate. How often you're actually included or cited in the answer. In 2026, "being in the answer" became the primary visibility KPI, because ranking well on Google stopped guaranteeing you show up in AI. It answers: are you in, or are you left out?
• Sentiment. Showing up isn't enough, HOW you show up matters. Does AI talk about you as the leader, as the expensive option, as the cheap alternative, as the one who had a problem? It answers: when they cite you, does the tone work for you or against you?

Think of these five as a ladder that climbs in demand. First you simply want to show up (citation). Then you want to show up more than the competitor (AI SoV). Then you want to be the primary recommendation (Share of Model). Then you want to dominate the answer, not just appear in it (Share of Answer). And, at the top, you want to show up well spoken of (sentiment). You can't skip rungs: it doesn't matter how great your sentiment is if your citation rate is zero, because nobody's talking about you to begin with.

The big question here is to change your report's question. Out goes "what position do I rank at?" and in comes "how often, and how, does AI talk about me when it answers my category?". Every AI presence tactic you saw in this module's earlier lessons (writing in extractable blocks, becoming an entity, earning third-party mentions) only becomes real management once you can measure the result. Without measuring, it's faith. With these five, it's a scoreboard.

03The instrument: the prompt library (your marketing's eval)

Alright, you have the new dashboard. But how do you actually measure this, if the customer's conversation with AI is private and you have no access to it? The answer is direct and might surprise you with its simplicity: you become the customer. You build a fixed set of questions that mimic your real buyer and run those questions, recurrently, across the various AIs (ChatGPT, Claude, Gemini, Perplexity), noting how each one answered. This has a name in the marketing world: marketing evals, a prompt library.

And here I want you to have a click of recognition. This isn't a new technique you need to learn from scratch. It's the exact same evals discipline you already saw in lesson 5.3 and G.6 in the Guardian track. There, the idea was: to trust an AI output, you create a fixed set of test cases and always run the same ones, to catch when quality drops. Here it's the exact same move, except the "test case" is a buyer's question and the "output" you evaluate is whether (and how) your brand showed up. You already know how to do this. You're just pointing the method at a different target.

There's an honest distinction that separates whoever measures correctly from whoever fools themselves, and the market named it lab data versus field data. Lab data is exactly your prompt library: you control the question, run it whenever you want, compare month over month. It's cheap, it's yours, and it's great for catching trends. But it's your scenario, not the real world's. Field data is the actual behavior of millions of users, captured by tools that measure what real people actually ask. The market's mature recommendation is to calibrate both: start with the lab (which you can build today, for free) and, as the channel grows, validate against field data. Don't confuse your controlled test with the whole of reality.

The discipline that makes this work is recurrence. A single measurement is a photo, and a photo doesn't show movement. You want the movie: run the same library every month, with the same questions, and watch the variation. When did your Share of Model go up? When did sentiment turn sour? When did the competitor overtake you in AI SoV? It's the recurrence that turns five loose metrics into an alert system that warns you before the damage shows up in revenue.

And here's the stitch with the rest of the course, without reopening anything. When you're building and reading this library, three precautions you already learned are worth gold. One: sycophantic AI exists (the very same thing you saw in synthetic research), so don't ask "is my brand good?", ask the way the customer would, neutral, and read the raw answer. Two: remember the prompt injection lesson (G.2) and whoever tries to plant an instruction to manipulate AI's answer? When you measure sentiment, you might run into your brand described in a strange or unfair way, and part of that could be noise or third-party manipulation, not the truth about you. Three: the audit discipline (G.3) applies to your own scoreboard, double-check the reading before taking it to your boss, because AI can even summarize its own answer incorrectly. You're not learning a new method, you're reusing what the track already gave you.

04The report you bring to the table

You don't need an expensive tool to get started, you need method and recurrence. There are market platforms that automate this (AI visibility indexes that track mentions for you), and they help as the channel grows. But the real move, the one that changes your power in the meeting, is simpler and is in your hands today.

Your marketing report gains a new page, next to the old ones (which you don't throw away, classic SEO still matters where there's still a click). This new page answers, month by month, five questions: in how many of my category's answers did I show up (citation), what's my slice against competitors (AI SoV), in how many was I the primary recommendation (Share of Model), how much of the answer was about me (Share of Answer), and with what tone (sentiment). Five lines. One column per month. The variation tells the story.

Notice what this does for you at the decision table. Before, when traffic dropped and the ranking was good, you had no answer, you just shrugged and blamed "the algorithm." Now you walk in and say: "traffic dropped because our Share of Model plunged from 31% to 12% in three months, AI stopped recommending us and started pointing to the competitor, and here's the exact month it flipped." That's not just one more report, it's the difference between being the one who explains what's happening and being the one caught off guard. The blind dashboard turns you into a hostage; the new dashboard turns you into someone who sees in the dark.

To take away: your customer's journey moved inside AI (the dark funnel) and the SEO dashboard went blind, because good ranking no longer guarantees showing up in the answer. The new dashboard has five KPIs in a ladder: citation rate (do you show up?), AI Share of Voice (more than the competitor?), Share of Model (are you the primary recommendation?), Share of Answer (do you dominate the answer?), and sentiment (do you show up well spoken of?). And the instrument to measure all of it is a prompt library you run every month, the exact same evals method from 5.3 and G.6, now pointed at your marketing. Fair? Next.

Do it now

Practice

Fair? Let's close out this lesson's message together. Your marketing dashboard didn't get worse, it went blind: the customer's journey moved into conversations with AI (the dark funnel), and good ranking stopped guaranteeing you show up in the answer, which is where the decision happens now. The fix has two parts. First, swap the dashboard: out goes the obsession with position and clicks, in come the five KPIs in a ladder (citation rate, AI Share of Voice, Share of Model, Share of Answer, and sentiment), which answer whether you show up, whether you show up more than the competitor, whether you're the primary recommendation, whether you dominate the answer, and whether you're well spoken of. Second, build the instrument: the prompt library you run every month, with the same buyer questions, the exact same evals discipline the Guardian track (5.3 and G.6) already taught you, now pointed at your marketing. Whoever does this walks into the meeting knowing the exact month AI stopped recommending the brand, while the competitor next door still shrugs, looking at a stable ranking, swearing everything's fine. In the dark, whoever brought a flashlight wins.

For the board