Purchase intent is the likelihood that a person will buy a particular product, brand or service within a given period. The term covers two related things. In research, it is a stated measure: a survey answer in which someone rates how probable a purchase is. In advertising technology, it is an inferred signal: a judgement, made by a platform from searches, page visits, shopping behaviour or transactions, that a person or company is currently in the market for something. Both exist because advertisers want to reach people about to buy, and to know whether advertising moved anyone closer.

The stated version still anchors brand measurement; the inferred version powers audience targeting sold by Google, Amazon, data brokers and business-to-business (B2B) intent providers. Both travel under the same label, a frequent source of confusion.

How stated purchase intent is measured

The most common survey format is a five-point scale running from "definitely will not buy" to "definitely will buy". Researchers frequently report only the "top box" (the share choosing the highest point) or "top two box" (the two highest points), because those shares track later purchases more closely than an average across all five.

A second format asks for a probability rather than an attitude. The economist F. Thomas Juster set out the case for it in Consumer Buying Intentions and Purchase Probability: An Experiment in Survey Design, published by the National Bureau of Economic Research (NBER) in 1966. The resulting 11-point Juster scale asks respondents to place the chance of buying on a 0 to 10 scale, with verbal anchors from "no chance" to "certain". According to Rene Befurt and Alvin J. Silk, writing for the American Bar Association's Landslide publication in 2019, a meta-analysis of 40 independent studies supports the scale's long-term track record and predictive validity.

Raw answers are rarely taken at face value; forecasters discount them, and the discount varies by category. Manohar Kalwani and Alvin Silk showed in Marketing Science in 1982 that intentions predict buying for both durable and packaged goods, but in different shapes: roughly linear for generic durables, and with a threshold effect for branded packaged goods.

Purchase intent inside brand lift studies

On advertising platforms, stated purchase intent appears most often as one question in a brand lift study. An audience is split at random into an exposed group that sees the campaign and a control group that does not. Both receive the same short survey, and the difference in answers is the lift. Google's YouTube studies, for example, send a single-question survey about a day after exposure and list purchase intention of the product among the outcomes they can measure.

Display & Video 360 (DV360) documentation lists purchase intent alongside association, awareness, consideration, favourability and video ad recall, with a cap of three metrics per study. The advertiser picks an "intended action" and an "object type", enters its brand and up to three competitors, and the platform generates a question asking which of those brands the respondent is most likely to buy from next. According to the DV360 Help Center, lift above 4% becomes detectable at about 1,200 responses, and a line item that shows no lift after 16,800 responses may never show one.

Purchase intent sits at the bottom of the survey funnel. Awareness and recall usually move first and further; intent moves last and least, which is why vendors have started pairing survey lift with transaction data. In September 2026, Cint began offering Magnite clients studies that report brand lift and sales lift together while campaigns are still running.

Inferred purchase intent and audience targeting

The targeting version works without asking anyone anything. Google introduced "in-market buyers" on its Display Network in November 2013, built from browsing patterns; Search Engine Land reported that frequent visitors to a site were excluded, on the logic that they probably already owned the product. Google's own help pages now describe in-market segments as a way to "reach users based on their recent purchase intent", determined by activity on Google products and third-party websites.

The definitions expanded from there. Google announced in May 2017 that in-market audiences would come to Search campaigns, according to Search Engine Land. In November 2017 it began rolling out custom intent audiences on the Display Network, letting advertisers describe the topics and URLs likely buyers would research. In October 2019, Google said it would bring in-market and affinity audiences to Search campaigns. In September 2020, it merged custom intent and custom affinity into a single custom audiences option, relabelling keyword-based lists as "people with any of these interests or purchase intentions". A 2024 guide recommended 10 to 15 keywords for Display and at least 50 search terms for YouTube.

Retailers sell a different kind of intent: one built from receipts rather than browsing. Amazon's demand-side platform (DSP) extended its in-market audiences to 32 countries with unified definitions on November 4, 2025, before launching natural-language AI targeting that draws on shopping, streaming and browsing signals. Sam's Club Connect went further in September 2026, launching segments that forecast which households will buy in a category up to 12 months ahead. Harvey Ma, vice president and general manager of Sam's Club Connect, described the model as "an indication of what might come 12 months into a member's life cycle."

In B2B marketing, intent is usually measured at the company level. Bombora's Company Surge pools content consumption from a data co-op of publishers and brand sites to flag businesses researching a topic; by December 2025 it offered more than 19,000 intent-based audiences, then available on Reddit. Intentsify, a rival, says it processes about 1.1 trillion intent signals a month across more than 33,000 topics. The scores typically feed lead scoring models.

Why it matters

Purchase intent bridges upper-funnel spending and sales. A brand that cannot run a sales experiment can still show a rise in the share of people saying they would buy, while media buyers use intent segments to concentrate budgets near the point of transaction. Google's own reporting reflects that distinction: DV360 began separating in-market audiences as a distinct category in audience performance reports at the end of 2025.

InMarket, a location and commerce data company, framed April 2026 consumer packaged goods results around real-time shopper intent, reporting incremental return on ad spend (iROAS) as high as $11.43. Those figures are the company's own and were not independently verified.

Limitations and disputes

The say-do gap is the central problem with stated intent. People overstate what they will buy. Yacheng Sun and Vicki Morwitz noted in a 2010 model of stated intentions that the estimated correlation between true intentions and purchases was 0.742 for cars, a mature category, but 0.536 for personal computers, then a fast-growing one. They also cite an observed intention-purchase correlation of 0.76 for durable goods. New products, cheaper items and longer time horizons tend to weaken the link.

Asking can also change the answer. Morwitz, Eric Johnson and David Schmittlein reported in the Journal of Consumer Research in 1993 that simply asking people about their intention to buy a car or computer raised purchase rates, a finding now called the mere-measurement effect. Brand lift respondents are therefore not perfectly neutral observers.

Inferred intent has its own accuracy problem. Nico Neumann, Catherine Tucker and Timothy Whitfield tested third-party data broker segments in field studies published in Marketing Science in November 2019. According to the INFORMS release, gender was identified no better than chance, and Tucker said "advertisers have little chance of assessing how accurate the profiles they are buying are." Platform segments are opaque too: Google does not publish the signals or recency windows that qualify someone as "in-market".

Correlation is another trap. HubSpot's June 2026 survey of more than 3,000 CRM buyers found that those who used AI search during evaluation were 36% more likely to purchase, but the data was self-reported and drawn from HubSpot's own customer base. Such research shows what accompanies intent, not what causes it.

Not the same as

Consideration measures whether a brand is on the shortlist, not whether a purchase is likely. A person can consider five cars and intend to buy one.

Purchase frequency describes how often people actually buy over a period. Purchase frequency is observed behaviour after the fact; purchase intent is a forward-looking estimate.

Search intent classifies what a query is trying to achieve (informational, navigational, commercial or transactional). Only the transactional slice overlaps with purchase intent.

Conversion lift measures actual conversions caused by advertising, using exposed and holdout groups. It replaces the intent proxy with the outcome itself.

Recent developments

Generative AI is changing both sides of the term. Researchers from PyMC Labs and Colgate-Palmolive reported in October 2025 that language models scoring text responses against Likert anchors reached 90% of the test-retest reliability of human purchase intent panels across 57 personal care surveys and 9,300 human responses. The headline figure is a ratio to a human reliability ceiling, not a simple accuracy rate.

On the targeting side, intent providers argue that AI agents need their data. In a May 2026 white paper, Intentsify president Charlie Allieri warned that an agent working from weak signals is "confidently wrong". Meanwhile, purchase decisions themselves are moving into AI assistants. A Koddi survey published in June 2026 found that 75% of US consumers were comfortable with AI helping them choose purchases, but only 20% with letting it act alone, and that 84% of commerce media leaders would invest in visibility inside AI-generated shortlists. If an agent forms the shortlist, whose intent is being measured becomes harder to say.

Timeline

  • 1966 - F. Thomas Juster publishes Consumer Buying Intentions and Purchase Probability through NBER, setting out the 11-point probability scale.
  • 1982 - Kalwani and Silk publish their study of the reliability and predictive validity of purchase intention measures in Marketing Science.
  • 1993 - Morwitz, Johnson and Schmittlein report that measuring intent changes purchase behaviour.
  • November 2013 - Google introduces "in-market buyers" targeting on the Display Network.
  • May 2017 - Google announces in-market audiences for Search campaigns.
  • November 2017 - Google begins rolling out custom intent audiences on the Display Network.
  • October 2019 - Google announces in-market and affinity audiences for Search campaigns.
  • November 2019 - Neumann, Tucker and Whitfield publish field evidence on third-party profiling accuracy.
  • September 2020 - Google merges custom intent and custom affinity into custom audiences.
  • October 27, 2025 - PyMC Labs and Colgate-Palmolive publish research on LLM-simulated purchase intent.
  • November 4, 2025 - Amazon DSP expands in-market audiences to 32 countries.
  • December 3, 2025 - Bombora and Reddit announce B2B intent audiences.
  • June 12, 2026 - HubSpot publishes data on AI search and CRM purchase decisions.
  • August 20, 2026 - Clay and Intentsify announce an intent data partnership.
  • September 17, 2026 - Sam's Club Connect launches predictive purchase targeting.
  • September 28, 2026 - Cint brand lift and sales lift studies become available to Magnite clients.

Summary

Who: Market researchers, brand lift vendors such as Google and Cint, audience platforms including Google, Amazon and Sam's Club Connect, and B2B intent providers such as Bombora and Intentsify measure or sell purchase intent. Advertisers and agencies use it.

What: A measure of how likely a person or company is to buy, either stated in a survey (five-point or Juster probability scales) or inferred from searches, browsing and transactions to build targeting segments.

When: Survey methods date to at least F. Thomas Juster's 1966 NBER work. Google introduced in-market targeting in November 2013, custom intent audiences in November 2017, and merged custom options in September 2020. Retail and B2B intent products expanded through 2025 and 2026.

Where: In brand lift surveys on YouTube, DV360 and third-party panels; in targeting menus in Google Ads, DV360 and Amazon DSP; and in B2B data co-ops feeding CRM and account-based advertising.

Why: Advertisers want to reach people close to buying and to show that campaigns move them closer. The measure is contested because stated intentions overstate behaviour, asking can alter it, and inferred segments are opaque and sometimes inaccurate.