What do buyers ask AI to optimize for?

A requirement-anchored recommendation is one brand statement in an AI answer tied to one extracted buyer requirement: the requirement the answer says the brand serves. Each requirement is classified into a category: feature requirement, workflow, pricing and contract terms, integration requirement, audience fit, business size, location, migration, or general. Of 920,930 recommendations anchored to a specific (non-general) requirement, 38,214, or 4.1%, anchored on pricing and contract terms. That is about 1 in 24.

These are the requirements AI anchors its recommendations to in answers to a tracked panel of buyer prompts, not organic user telemetry. Within that frame, the result is consistent: when an AI answer explains why it recommends a brand, the reason is almost never price.

Feature requirements supplied 487,764 of 920,930 recommendations tied to specific requirements, or 53.0%. Workflows (the buyer's process or use case) supplied 328,083, or 35.6%. Together that is 88.6%. Pricing and contract terms reached 4.1%, integration requirements 3.7%, audience fit 2.0%, business size 1.0%, location 0.44%, and migration 0.09%.

Feature requirements outnumbered pricing 12.8 to 1, and workflows outnumbered pricing 8.6 to 1. A further 144,030 recommendations carried only a general requirement with no specific category; they sit outside this denominator.

The headline rests on a model-extracted classification, so we checked it two more ways. First, a plain token match: 6.3% of extracted requirements contained a price word (price, cost, budget, affordable, cheap, free, fee, or discount): 65,310 of 1,038,672. Second, counting answers: 10.1% of answers contained at least one price-anchored recommendation (20,626 of 204,857), against 72.9% for feature requirements and 45.8% for workflows.

All three measures put price at the bottom of the hierarchy. Even when counting any price-related recommendation anywhere in an answer, feature requirements appear 7 times as often.

The recommendations we could assign to markets fall into 31 industries with at least 5,000 specific-requirement recommendations each. The table shows each industry's share of feature, workflow, pricing, and integration requirements. Every industry splits differently, and the spread is wide: feature share runs from 33.6% (Collaboration) to 78.9% (Clothing and Apparel), and price share from 1.3% (Artificial Intelligence) to 10.0% (Consumer Electronics).

Read your own row before writing content. The overall mix hides which requirement type your buyers' questions actually anchor on.

Clothing and Apparel had the highest feature share at 78.9%, followed by Consumer Goods at 74.3%, Gaming at 71.7%, Consumer Electronics at 70.8%, and Sports at 69.7%. In these categories, the requirement behind a recommendation is usually a product attribute: a material, a capability, a fit.

For a consumer brand, the practical unit of AI visibility is the attribute. The answer recommends the product that matches the stated spec, so pages that state specs plainly match more requirements.

Collaboration is the only industry where workflows lead: 56.8% of recommendations tied to a requirement, against 33.6% for feature requirements. Professional Services follows at 51.3% workflow, then Education at 47.0%, Sales and Marketing at 44.5%, and Administrative Services at 42.8%.

In these categories the buyer requirement is a process: running payroll, scoring leads, managing a case load. The recommendation goes to the brand the answer connects to that process, not to a spec sheet.

Consumer Electronics had the highest price share at 10.0%, with Clothing and Apparel at 10.0%, Consumer Goods at 8.1%, Commerce and Shopping at 7.9%, and Health Care at 7.2%. At the other end, Artificial Intelligence recommendations anchored on price 1.3% of the time, Collaboration 1.6%, and Gaming 2.1%.

No industry crosses 11%. Price-driven verticals exist, but even there, nine of ten recommendations tied to a requirement concern something other than price.

The most repeated price-anchored requirements are not "cheap option". They are structured cost requirements: low-cost domain registration (114 recommendations), transparent payroll pricing (63), affordable prescription eyewear (62), bare metal server pricing (54), and a real estate CRM without per-user fees (50).

The pattern is a pricing model, not a discount: no per-user fees, transparent pricing, insurance-covered. A pricing page that names its model in plain words can match these requirements; a generic "affordable" claim matches none of them.

In every large industry, the single most repeated requirement is a narrow one, not a broad category. Financial Services led with high-yield savings account (377 recommendations of 66,319 requirement-anchored recommendations in the industry). Professional Services led with slip and fall legal representation (500 of 28,999), and Commerce and Shopping with custom engagement ring design (237 of 37,679).

Even the top requirement in an industry carries under 2% of that industry's recommendations tied to a requirement. These leaders describe our tracked prompt panel as much as the market, but the shape is the point: the recommendations refer to thousands of specific requirements.

The 897,918 specific-requirement recommendations with a requirement string resolve to 482,521 distinct requirements. Of those, 358,900, or 74.4%, appeared exactly once in the whole corpus. The 10 most common requirements together cover 3,009 recommendations, or 0.34% of the total.

There is no head of the requirement distribution to chase. Coverage of many narrow requirements beats depth on any single one.

Restricting to the current engine pair (June 2 through August 28, 2026), Google AI Mode anchored 70.0% of its specific requirements on feature requirements, 17.3% on workflows, and 2.8% on price, across 64,869 recommendations. ChatGPT Search came in at 69.0%, 19.5%, and 2.3% across 64,730.

The engines landed within 2.3 points of each other on every requirement category, despite their differences in brands, sources, and recommendation order.

This is measured in the Parse index over the stated window. Requirements are extracted by a model from AI answers to Parse's tracked panel of buyer prompts, so they describe what AI answers optimize for on those prompts, not organic user telemetry, and not what buyers privately want. The panel over-represents some markets, which shapes the named requirement examples.

The sample spans two engine eras: 929,360 recommendations from ChatGPT and Google AI Overviews (October 3, 2025 through April 25, 2026) and 135,600 from ChatGPT Search and Google AI Mode (June 2 through August 28, 2026). We make no time-trend claims across eras. Industry comparisons cover only recommendations we could assign to a market, and industries under 5,000 recommendations are excluded. Requirement classification is model-extracted; checks for price words and separate counts by answer provide checks on classification error.

AI buying answers anchor recommendations to requirements and use cases: 53.0% feature requirements, 35.6% workflows, 4.1% price. The requirement distribution is almost all tail, and nearly every recommendation is conditional.

Write pages that state, in the buyer's words, the specific requirement your product serves: one requirement per section, plainly worded, including the pricing model where price is the requirement. Check which requirements AI currently attaches your brand to, then close the gap between the requirements you serve and the requirements your pages state. Rerun the same study next quarter on a same-length window.