Chapter 15 — What to Watch

Forecasting here has a poor record, so what follows is framed as things to watch with thresholds attached, rather than predictions. The useful question isn’t “what will happen.” It’s “what would I do differently, and what would tell me it was happening.”

Will specialists really win?

The theory says yes: because retrieval rewards proximity to a specific question rather than aggregate reputation, small specialists gain an advantage. Small brands double down on niches, large brands respond by segmenting into sub-brands, and a new layer of specialists emerges between manufacturers and consumers.

If that holds, something significant follows: brand equity becomes local to a market rather than a portable asset that lets a big company enter new categories cheaply. That would invert one of the oldest assumptions in marketing strategy.

The counter-evidence is substantial. The largest observational study of brand visibility found a steep incumbency gradient — household names appearing in roughly 73 percent of relevant answers, mid-market brands 44 percent, small and niche brands 11 percent. Whatever theoretical advantage specialists hold, incumbents are winning today.

Watch for: specialists appearing in your constrained-recommendation questions while absent from broad category questions; the incumbency gradient flattening in successive studies; sub-brand launches from large competitors built around narrow use cases.

Either way: the specificity work is the hedge. It costs the same whether you’re the incumbent or the challenger.

Will anyone be able to buy the recommendation?

Probably not the organic one, for the reasons in Chapter 12 — though the separation is thinner than it looks. Ads already appear in about a quarter of ChatGPT responses. Google is piloting programs that let retailers surface exclusive discounts inside AI Mode.

The distinction that matters isn’t “ads or no ads.” That’s settled. It’s whether the organic recommendation can be influenced by spend.

Watch for: unlabeled placement changes correlated with spend; platforms selling “recommendation visibility” as a product; a measurable correlation between ad spend and organic mention rate that survives controlling for brand size. That last one is a study you could actually run.

Do now: measure organic mention rate separately from any paid placement, starting immediately. If the surfaces blur later, you’ll want the clean baseline.

Will agents actually buy things?

The bull case is the trillion-dollar forecasts. The bear case is what happened in March 2026, when OpenAI retreated from native checkout after roughly a dozen merchants adopted it, because people research in the assistant and buy elsewhere.

The likely middle is agent-initiated, human-confirmed: the agent assembles the cart or fills the checkout, and a person confirms. That’s functionally the same as one-click purchase on your own site, it preserves user authority, and it avoids the accidental-purchase problem, where consumers may remain liable for transactions they never explicitly approved. Full autonomy creates regulatory problems nobody has solved.

Watch for: checkout share of agentic traffic rising above single digits; UCP checkout expanding beyond the initial retailer set; a major platform shipping autonomous purchase with a published liability framework.

Do now: the Chapter 9 list and nothing beyond it. A feed, the eight attributes, a manifest, a quarterly test. That work pays off in discovery today regardless of whether transaction volume ever arrives.

Will measurement standardize?

Regulatory pressure says partly. UK competition authorities have required impressions, click-throughs and click-through rate; Google currently supplies only impressions, with further obligations scheduled.

Architecture says no. Personalization makes a universal ranking incoherent. Non-determinism makes single measurement meaningless. No regulator can mandate a number that doesn’t exist.

The likely outcome is partial: better first-party reporting, no standardized share-of-voice metric, continued vendor disagreement. Which means the discipline in Chapter 11 stays a competitive advantage rather than becoming table stakes.

Watch for: Google shipping clicks and CTR in Search Console; an industry body publishing a sampling standard; vendors starting to publish confidence intervals.

Six numbers, quarterly

Retrieval rate — the share of prompts triggering a web search. Currently 18 to 35 percent depending on measurement, and falling. If it keeps falling, the entity lever grows and the content lever shrinks.

Citation breadth — unique domains receiving referrals. Off its late-2025 peak but well above a year earlier. If it resumes narrowing, the specialist thesis is losing.

Ad density inside answers — around a quarter of ChatGPT responses.

Checkout share of agentic traffic — around 3 percent.

Platform share volatility — a twenty-point swing in the last twelve months.

First-party reporting — whether clicks and CTR ever arrive.

Each has a threshold at which your allocation should change. Write those thresholds down now, while nobody’s under pressure, and revisit them every quarter.