What gets measured as marketing success when AI can hit any proxy metric but brand trust is still unmeasured?

The shift

Producing content and optimizing toward a measurable proxy — clicks, CTR, engagement, open rates, ranking — goes from scarce (spend, headcount, testing cycles) to abundant and near-free. A proxy metric only ever worked as a stand-in for value because gaming it used to cost real money and effort; when saturating any defined number is cheap, the proxy stops correlating with the thing it was proxying for.

The axioms

• A proxy metric (clicks, CTR, engagement) is a reliable stand-in for marketing value, because moving it required real spend and effort that mediocre work couldn't fake.
• Hitting the number is evidence the work was good — the metric and the underlying value moved together because faking the metric was expensive.
• Teams and agencies should be rewarded, budgeted, and promoted on the measurable outputs they can move, because those outputs tracked business value closely enough.
• Signal separates itself from noise in-market, because producing noise at volume was costly, so the feed was self-limiting and a metric earned in it meant something.
• Brand equity is real but too slow and unattributable to manage directly, so you manage the leading proxies instead and trust they roll up to it.
• Knowing what's worth saying is downstream of knowing what performs — let the metrics tell you, since testing was the scarce, authoritative input.

Invalid axioms

1. A proxy metric is a reliable stand-in for value because moving it costs real effort. The cost of saturating a defined number collapsed. AI can generate the content, the variants, the SEO surface, and the engagement-bait to hit almost any measurable target at near-zero marginal cost — and where the metric can be moved by bots or synthetic engagement, the flooring is near-zero too. Habit-trap: dashboards still treat CTR/engagement as the scoreboard, so a team looks like it's winning precisely when it has learned to Goodhart the metric fastest.
2. Hitting the number is evidence the work was good. The metric and the value came apart the moment the metric became trivially hittable. A rising engagement line is now equally consistent with real interest and with optimized noise, and the number itself can't tell you which. Habit-trap: teams still read "we hit target" as "we did good work" and report it upward as such.
3. Reward teams and agencies on the measurable outputs they can move. Any incentive pinned to a gameable proxy now selects for whoever games it best, not whoever builds most value. Habit-trap: comp, agency SLAs, and OKRs still key on leads/clicks/MQLs — paying most for the metric AI made cheapest to fake, and training the org to optimize the proxy over the business.
4. Let the metrics tell you what's worth saying. When producing and testing thousands of variants is free, "what performed" is abundant and no longer authoritative — the test can be won by content that erodes the brand. Habit-trap: strategy is still ceded to whatever the performance data ranks highest, delegating taste to a proxy that AI can now win without saying anything worth saying.

Unchanged axioms

1. Brand trust is the thing actually being bought, and it's still unmeasured and hard to fake. Trust rests on accumulated real experience, verification, and standing — none of which got cheaper. AI can flood every measurable proxy, but it can't manufacture a customer's earned confidence, and no clean metric captures it. The scarcity didn't move; it just became the only thing left that a proxy couldn't counterfeit.
2. Someone must be accountable for whether marketing built durable equity or gamed a number. A quarter of Goodharted metrics and a quarter of real brand-building look identical on a dashboard and diverge only later, in retention, pricing power, and pipeline that doesn't evaporate. A named person still owns that outcome; a model can hit the proxy but can't be answerable for the equity.
3. Judgment on what's worth saying is a call under uncertainty, not a generation task. Deciding the positioning, the claim, the bet that builds trust rather than churns attention is still a human call — and AI is structurally pulled toward the median of what already performed, which is exactly the noise the proxy now rewards.
4. Taste is what separates signal from optimized noise once the feed is saturated. When everyone can hit the number, the discriminating act is recognizing which content builds something and which merely scores — a verification-and-judgment problem, and those stayed scarce.

New axioms

1. When any proxy is gameable at zero cost, the scarce problem becomes measuring the unmeasured thing — trust itself. Marketing has always managed proxies precisely because equity was too slow to measure directly. That workaround breaks: the proxies are now noise, and the org needs a read on durable trust it never built the instrumentation for. This is the load-bearing open problem and it has no accepted metric yet.
2. Distinguishing real brand-building from metric-gaming AI output, when both produce identical dashboards. Verification capacity — telling equity-building work from proxy-saturation — didn't scale with generation capacity. Whether AI can be turned on its own output to audit this, or whether it just launders the ambiguity, is unresolved.
3. What replaces the proxy as the operating signal, given a team can't run on a lagging trust metric alone. Kill the gameable proxy and you remove the daily instrument teams steer by; nothing real-time and honest has replaced it, so teams either fly blind or keep steering by a number they know is now fake.
4. Who owns "measurement integrity" once hitting the target is trivial. Governing what the org rewards — resisting the pull to bank easy proxy wins — has no owner. The bottleneck that used to enforce metric honesty (moving a number was hard) is gone, and no equivalent gate replaced it.
5. When the audience for optimized content is increasingly another AI, the proxy detaches from human trust entirely. If ranking and engagement are mediated or generated by agents, the metric can be won with content no person ever trusted or even saw — early, but moving fast enough to flag.

Where it breaks

"Reward teams on the measurable outputs they can move" (invalid) collides with "the scarce problem is now measuring trust, which has no metric" (new): the org keeps paying for the exact numbers AI made free to fake, which actively funds proxy-gaming over trust-building — and because trust is unmeasured, nothing on the dashboard flags the trade until equity has already eroded. The incentive system is optimizing hardest for the thing that stopped meaning anything.

"Let the metrics tell you what's worth saying" (invalid) collides with "distinguishing real brand-building from metric-gaming output" (new): the performance data teams still defer to is now won by content engineered to score, so ceding strategy to it systematically selects the noise a human would have caught — the org has automated away the taste that was the only remaining defense, and calls it being data-driven.

ai-axioms.vercel.app/axioms/marketing/marketing-cmo-metrics-ai-axioms