Purchase intent sounds like one signal. In practice it is two very different systems sharing a label. One is a stated answer on a survey. The other is an inferred score built from searches, site visits, receipts and platform behaviour. Media buyers and brand teams keep mixing them, and the gap shows up in wasted budget.
The two purchase intents
Stated intent appears in brand lift studies. A five-point “definitely will not buy” to “definitely will buy” scale is the classic format; researchers often report the top box or top two boxes because they track real purchases better than an average. An older probability alternative, the 11-point Juster scale from 1966, still holds up in long-term validation.
Inferred intent powers audience targeting. Google’s in-market segments, Amazon DSP audiences and B2B providers like Bombora and Intentsify all promise to find people or accounts currently shopping. No one is asked a question; a model decides.
What’s changed
The recent moves are about predicting further out and tying intent to sales. Sam’s Club Connect launched segments in September 2026 that forecast which households will buy in a category up to 12 months ahead. Harvey Ma, VP and GM of Sam’s Club Connect, called the model “an indication of what might come 12 months into a member’s life cycle.”
Meanwhile Intentsify says it processes about 1.1 trillion intent signals a month across more than 33,000 topics, and Cint began pairing brand lift with sales lift for Magnite clients in September 2026. Google’s DV360 documentation already gives a practical detection threshold: lift above 4% becomes detectable at about 1,200 responses, while a line item showing nothing after 16,800 responses may never show an effect.
Where intent data goes wrong
Stated intent has a say-do gap. People overstate purchases. One study put the correlation between stated intention and purchase at 0.74 for cars and 0.54 for personal computers; cheaper items and longer horizons weaken it further. Worse, simply asking about intent can change behaviour—the mere-measurement effect found in 1993 research.
Inferred intent has an opacity problem. A 2019 field study of third-party data broker segments found gender was identified no better than chance. Google doesn’t publish the signals or recency windows that make someone “in-market.” Correlation studies, like HubSpot’s finding that AI-assisted search users were 36% more likely to buy, show accompaniment, not cause.
How to use both without fooling yourself
Separate the measurement question from the targeting question. If you need brand movement, treat survey lift as one input. If you need conversions, treat intent segments as a hypothesis to test before scaling.
- Treat survey lift as a brand-health metric, not a sales forecast.
- Ask vendors which signal, recency window or survey question sits behind “intent.”
- Validate inferred segments with your own holdout test or conversion lift study before scaling.
- Watch AI-generated shortlists: if an assistant shapes consideration, the intent you’re paying for may belong to the model’s inference, not the shopper’s.
Purchase intent remains a bridge between upper-funnel spend and sales. Use it as a directional input, not a proof point.
Source: PPC Land
