The head-to-head is the only AI visibility metric that matters

Strip the static away for a minute. Revenue targets, brand awareness, traffic, engagement, share of voice: every one of those is a proxy for something underneath it. Boil business success all the way down and one definition is left standing: more customers than your competition. And the moment a business generates revenue, it generates competitors, so that definition never retires. Success is relative, permanently.

I say this as the frame for everything else I write about AI visibility, because the frame decides where you focus. The question about AI is never “does AI mention us.” It is “when a buyer asks AI who to choose, do we win.”

That is the statement this whole series unpacks. A recommendation is a delivered decision, and the head-to-head is the metric that tracks it. Content is Code is my methodology for changing it, built on one idea: content is the material AI systems read when they decide what to recommend, so treat it like code. Query the models with real buying questions. Decode why the winning answer wins: the themes, evidence, and sources it leans on. Engineer that evidence on your own domain, which means concrete pages that state what you do, for whom, with what proof, in the language the winning answers already use. Then verify by rerunning the same questions. Everything else in AI visibility is instrumentation in service of that lens.

The recommendation is a delivered decision

When a buyer asks an AI platform for the best option in a category, the answer is a shortlist and usually a pick, with reasons attached. That is a different event from a search results page. The evidence on behavior points one direction: in the Pew Research Center's study of US Google users, visits that surfaced an AI summary clicked a traditional result 8% of the time against 15% without one, and links inside the summaries drew clicks on just 1% of those visits. One platform, one country, March 2025, and the direction matters more than the digits: influence is moving upstream of the click, into the answer itself. A recommendation you lose there is a customer you never got to pitch.

Mentions are presence. Recommendations are decisions.

Most AI visibility reporting counts presence: were you mentioned, were you cited, what score did the dashboard print. Presence matters, and it is the wrong finish line. One 2026 preprint tracking 100,000+ prompt responses across 100+ brands found the framing of a brand flipped about 6.7 times more often than whether the brand was mentioned at all: a vendor panel and a preprint, held loosely, but it matches what I see in practice. Presence is relatively stable; the framing moves. And in my own head-to-head series, the recommendation itself moves with it. That decision layer is where the action lives, and it is the layer attached to revenue.

That is why I hold programs to a head-to-head win rate against named competitors rather than a mention count. The five-layer measurement model I use puts head-to-head choice and recommendation as the top layers for a reason. A win rate is a business-shaped number: a board understands “we win the comparison four times in ten and our competitor wins six” without a glossary.

And the needle moves. Soapbox Bulletin, a client I can name, went from a 6.2% to a 36.2% AI recommendation win rate in eleven weeks: directional evidence rather than a controlled study, which is exactly how I report it. The mechanism was the method above, run as an operating loop: measure the losing comparisons on locked queries, publish the evidence the winning answers were citing from elsewhere, retest the same questions, repeat. Win rate here means the share of locked head-to-head questions where the platforms recommended the client. A financial services firm I keep anonymous took four out of four head-to-head wins in a strict URL-grounded series against named competitors. The full write-ups, limits included, live on the case studies page.

What the lens changes

Take the head-to-head as the core metric and your priorities reorder themselves.

First, your competitor list stops being an assumption. The competitors that matter are the ones the models actually name when they answer buyers, and that list routinely surprises people. Finding it is its own discipline, and it is the subject of the next piece in this series.

Second, your domain becomes an evidence base, aimed at comparisons. Google is explicit that its AI features work from the same crawled content as classic search, and there is no special file that shortcuts the work. The pages that win recommendations are the ones that carry the evidence the winning answers cite: what you do, for whom, with what proof, stated plainly enough to lift.

Third, the message consolidates. If the website says one thing, the social channels another, and the machine layer a third (the metadata, markup, and machine-facing files AI systems parse), the comparison engine averages your story while your competitor's stays sharp. One message, everywhere a machine or a buyer reads, is not a branding nicety in this lens. It is how you brief your advocate.

And fourth, every piece of content gets a harder question than “is this good”: does this help win a comparison a real buyer asks for? Content that cannot answer that question is a hobby.

What this does not say

Honesty about the lens's edges. AI mediates a growing share of recommendations, and nowhere near all of them; referrals, sales teams, and old-fashioned search still close customers. Success definitions can legitimately weigh margin, retention, or mission; more customers than the competition is the reduction I find under all of them, and it is a reduction. And a head-to-head result is volatile from question to question, which is why the metric only means something as a locked, repeated series, never as a screenshot. The lens tells you where to look. The instrument discipline makes what you see trustworthy.

Start with one question

You can test the lens today. Ask each major platform, ChatGPT, Claude, Perplexity, Gemini, and Grok, one real recommendation question in your category, phrased the way a buyer would phrase it. Write down who wins, who else gets named, and the reasons given. That is thirty minutes, and it answers the question most dashboards never ask: when the recommendation happens, are we in it, and do we win it?

If the answer stings, that is the starting line. First measure who is winning. Then change it: query, decode, engineer, verify. That is what the rest of this series and my practice are built on.

Questions worth asking next

Sources

1. Pew Research Center, "Google users are less likely to click on links when an AI summary appears in the results," July 22, 2025. https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/
2. Pratyush Kumar (Ranqo), "Generative Engine Optimization at Scale," arXiv preprint 2606.20065, June 2026. https://arxiv.org/abs/2606.20065
3. Google Search Central, "AI features and your website" guidance. https://developers.google.com/search/docs/fundamentals/ai-optimization-guide