24% of Search Budgets Now Chase an AI-Search Score the Vendors Can't Pin Down.

Marketers have already redirected 24% of their search and content budgets toward AI visibility. Most of them, 81%, still call the work “SEO.” Fractl reported both figures on July 21, and together they say something uncomfortable. The money is moving faster than the language for it, and faster still than anyone’s ability to measure it. The gap between the spend and the scorecard isn’t an accident someone is racing to close.

Every marketing team is asking the same question about AI search: how do we show up in the answer? There’s a quieter question underneath it that decides whether the budget was well spent. Who owns the number that would tell you it worked, and what do they sell?

Start with what we already know is broken, because we’ve spent a month documenting it. Google’s AI Mode carries ads on 29.45% of commercial queries, but the advertiser’s own domain shows up among the answer’s cited sources 11.53% of the time, so buying the slot beside the answer tells you nothing about whether you’re in it. On the organic side, 62% of the pages Google’s AI cites are not the pages that rank, which means the dashboard tracking your rankings is watching a different race than the one that decides citations. And the click that used to prove any of it happened is gone: two in three Google searches now end without one, and the referral data that fed every marketing dashboard stopped counting the traffic that matters.

So the instruments and the ground truth have come apart. The reasonable next move is to buy a better instrument, and a whole industry has appeared to sell you one.

Here’s that industry’s strongest case, stated fairly. AI visibility is becoming a real measurement discipline, the way search analytics did a decade ago. As of mid-July there are at least eight platforms selling it, by MarketScale’s count: Profound, AthenaHQ, Otterly.AI, Peec AI, Scrunch AI, Rankscale AI, Semrush, and CiteLens. They bring statistical rigor the old spot-checking never had. CiteLens wraps every visibility score in a 95% confidence interval to account for the fact that an AI answer changes from one run to the next. This is what maturing looks like: guesswork turning into instrumented, repeatable numbers a CFO can put in a plan.

It’s a good case. It holds right up until you check what those eight platforms are counting.

They count the same thing every time: brand citations, how often you get mentioned in AI answers. Not one of them measures the distinction that would make the spend accountable, which is whether your presence in an answer was paid or earned. None of them measures the position that moves revenue, which is whether the model recommends you by default or just lists you among options. They measure the mentions because mentions are the part they can see from the outside. The rest lives inside the platform, and the platform isn’t sharing.

That confidence interval is worth sitting with, because it’s an honest admission dressed as a feature. A vendor wrapping every score in a statistical band is telling you the underlying signal is unstable enough that a point number would be a lie. They’re doing the responsible thing with data they don’t own and can’t reconcile against the source. Rigor applied to a proxy is still a proxy.

This is the part the “measurement is arriving” story gets wrong, and the history it leans on is exactly where it breaks.

The ad industry has been here before, and it did fix it. In November 2012 the Media Rating Council told the market not to trade on viewable impressions. Nobody could agree on what counted as an ad a human could actually see. Eighteen months later, on March 31, 2014, the MRC lifted the advisory and approved a standard: 50% of an ad’s pixels in view for at least one continuous second. A messy, unmeasurable market got a number buyers trusted, and it happened fast.

But look at what made those 18 months work, because that’s the whole lesson. The standard came from the MRC, a neutral body that sold no ad inventory and no measurement product, so its number carried no conflict. And it was forced by organized buyers, the advertisers and trade bodies who refused to keep paying for inventory they couldn’t verify. A disinterested referee, and a buy side with the clout to demand one. Two preconditions.

AI search has neither. The party that owns the ground truth, the answer itself, is Google, which also sells the ad running beside it and has every reason not to publish a number that separates the two. There’s no MRC for AI citations, no neutral body with authority over a dozen model providers who share no common surface to standardize. And the closest thing to a measurement industry is eight companies selling scores, each with a commercial interest in the metric it happens to be best at producing. The mechanism that fixed viewability is simply not present in AI search.

That’s the pattern to hold onto. The measurement gap in AI search stays open not because the problem is hard, but because every party positioned to close it does better with it open. The platform sells more ads when you can’t tell paid presence from earned. The vendors sell more seats when “AI visibility” stays a proprietary score rather than a shared standard. Nobody in the room is the neutral referee, and nobody’s the advertiser coalition banging the table, because the advertisers are 24% of the way into the budget already and still calling it SEO.

Here’s the forecast, and it’s a forecast, not a report. Grounding it in the only rate-of-change we have: viewability took 18 months to standardize with a referee and an organized buy side. AI search has been commercial for more than a year with neither, and the tooling that’s appeared measures the sellable proxy instead of the accountable number. So the call is that no major AI platform, Google, Microsoft, or OpenAI, will expose inside its own analytics whether your presence in an answer was paid or earned before the end of 2027. The one number that would make the spend accountable is the one number the house will never print.

What would change my mind is specific, and it’s the same thing that changed it last time. If advertisers organize a neutral measurement body with real muscle, or if regulation forces the disclosure the way AI content-labeling rules are already forcing other admissions, the standard could arrive faster than the incentives suggest. Watch for a buy-side coalition, not a vendor launch. The vendor launches are the noise.

Until one of those happens, run AI search like a channel you can’t fully see, because you can’t. Budget paid placement and earned citation as two separate lines with two separate owners, since no vendor will split them for you. Trust the numbers you control, branded-search lift and direct traffic, over any score that ships with a confidence interval attached. And treat every “AI visibility platform” pitch as what it is: a useful proxy sold by someone who can’t see inside the box either, priced as if they can.