What is intent density?

Why a new measure was needed

Display advertising prices inventory by who is seeing it. Audience segments, behavioural profiles and lookalikes are all attempts to answer one question: how valuable is this person?

On a conversational surface there is no durable identity to price, and there is something better available. The user has just said what they are trying to do, in their own words. Intent density is the attempt to quantify that: how close is this conversation to a decision, and how commercially relevant is it?

The shift is from valuing the person to valuing the moment. A completely unknown user asking a precise buying question is worth more than a richly-profiled one browsing idly.

What raises it

• Specificity. "CRM for a 12-person agency under £50 a seat" is dense; "what is a CRM" is not.
• Constraints stated. Budget, timeline, must-haves and dealbreakers all indicate someone actually choosing.
• Comparison framing. Asking about alternatives to a named product is close to a decision by definition.
• Conversation depth. Intent usually accumulates — the fifth turn is denser than the first.
• Transactional verbs. Buy, book, switch, sign up, cancel.

How it is used

On the sell side it is the basis of packaging and pricing. An AI SSP uses it to describe an opportunity to buyers without sending them the conversation itself — a topic, an intent score, a decision stage. Higher density supports higher floors.

On the buy side an AI DSP bids against it directly. Because supply is scarce, the practical strategy is usually to bid narrowly and high on genuinely dense moments rather than broadly and cheaply — the opposite of a display reach strategy.

It is also an eligibility signal. Very low density conversations are frequently ones that should carry no advertising at all, which makes the same measure a guardrail as well as a price.

What it is not

It is not a purchase probability, and treating it as one leads to overbidding. A dense conversation indicates someone deciding; it says nothing about whether they will decide in your favour or on your timescale.

It is also not comparable across platforms. Every publisher estimates it differently and there is no shared scale, so a score of 0.8 on one surface and 0.8 on another are not the same thing. Calibrate against your own outcomes per surface rather than assuming the numbers mean the same thing.