Query Priming

Definition

Query priming is what happens when an AI search measurement prompt contains the answer you're looking for. Ask "What do you know about Notion's MCP server?" and the model has been told the answer is "Notion's MCP server" — it will almost always return something relevant. The query primed the response.

Priming is the default failure mode of naive AI visibility measurement. Most brand-monitoring queries are primed: "How good is [Brand] for [use case]?", "Compare [Brand] and [Competitor]", "[Brand] reviews." Each one tells the model the brand is the subject. The model obliges. The resulting "visibility" number is inflated relative to what an actual buyer — who doesn't yet know your brand — would experience.

The counterpart discipline is the anti-prime query: a prompt that describes the buyer's job-to-be-done without naming the brand or category, then measures whether the brand surfaces organically.

Why it matters

The gap between primed and unprimed (anti-prime) surface rate is the single most important signal in AI search measurement.

In the Citare Notion audit, Notion's MCP integration surfaced on 100% of cells where the query named "MCP" explicitly — and only 15% of cells where the query described the job without naming MCP. The primed number said "we're fully visible." The unprimed number said "buyers describing the problem we solve almost never find us." The second number is the one that reflects reality.

If your monitoring tool only runs primed queries, you will systematically overestimate your AI visibility — sometimes by 5-6x. You'll think you're winning when you're invisible to the buyer language that actually matters.

Primed vs anti-prime — a spectrum, not a binary

Queries sit on a priming spectrum:

• Heavily primed: "Tell me about Notion's MCP server" (names brand + feature)
• Brand-primed: "Is Notion good for engineering docs?" (names brand)
• Category-primed: "Best knowledge base software" (names the category label)
• Lightly primed: "Where should my team keep its documentation?" (names the job, neutral)
• Anti-prime: "My five-person dev team keeps losing track of decisions in Slack threads — what should we do?" (describes the pain, no category label)

The further toward anti-prime, the closer the measurement gets to real buyer behavior — and the harder it is to surface. Both ends of the spectrum carry information: primed measures "how well-described are we when someone already knows us," anti-prime measures "will a new buyer discover us."

How to avoid accidental priming

1. Strip brand names from any query meant to measure organic discovery.
2. Strip category labels where buyers wouldn't use them. Real buyers say "sticky-note app," not "knowledge management platform." A query using your industry's internal jargon is partially primed toward category incumbents.
3. Describe the pain, not the product. "How do I stop losing customer emails" is anti-prime; "best shared inbox software" is category-primed.
4. Run both, compare the gap. The delta between your primed and anti-prime surface rate is your actual marketing-distribution problem, isolated from product quality.

Common pitfalls

• Reporting only primed numbers. This is the most common way AI visibility gets oversold — by the brand to itself, or by a vendor to a client. Primed-only measurement always flatters.
• Assuming anti-prime is the "real" number and primed is worthless. Both matter. Primed tells you how well-described you are once discovered; anti-prime tells you whether you'll be discovered. You want strong scores on both.
• Using the same query set for months. Buyer language drifts. A phrasing that was anti-prime last year may have absorbed a category label that primes it today.