Intent is inferred, not observed
Intent lives in the person's head; marketing sees only behaviour. Every intent signal is therefore a probability, not a fact: a pricing-page visit usually means a decision is near, and sometimes means a competitor is checking. The craft is in choosing signals that predict well, combining them so single accidents do not dominate, and checking the inference against outcomes. The stages guide describes the states; this page covers how to tell which state someone is in.
Intent levels
A workable scale with five levels, roughly matching the journey stages and the See-Think-Do-Care model:
| Level | The person… | Typical signals |
|---|---|---|
| None | has no active need | general browsing; no category behaviour |
| Problem-aware | knows something is wrong and is learning about it | problem and "how to" searches; reading educational content; watching explainers |
| Solution-aware | knows the kinds of solution and is comparing approaches | category searches; comparison and "best" queries; guide and review reading; multiple sources in a session |
| Product-aware | has a shortlist and is comparing products or brands | brand and product searches; "versus" queries; product and pricing pages; reviews of specific products; return visits |
| Ready | has decided and is checking logistics and risk | brand-plus-modifier searches ("…discount", "…delivery", "…near me"); cart, checkout, quote or demo pages; repeat visits within a day; contact actions |
The levels map onto message: education for problem-aware, comparison and proof for solution- and product-aware, reassurance and friction removal for ready. Mismatching is the most common waste in paid media — a discount offer to someone problem-aware, an explainer to someone ready.
Where the signals come from
Search queries. The richest intent source, because the person wrote down what they want. Query modifiers classify them: question words and "how to" (problem-aware); "best", "types of", "vs" (solution-aware); brand and product names, "review", "pricing" (product-aware); "buy", "near me", "discount", "delivery", model numbers (ready). The SERPThe page of results a search engine shows. itself is a check: the results Google shows for a query reflect the intent it has learned for it.
Pages and content. What a visitor read and in what order: problem content versus comparison content versus pricing and product pages. Sequences matter more than single pages; a path from guide to comparison to pricing is a journey, a single pricing visit may be anything.
Actions. Things that cost effort: downloading, saving, comparing, starting a trial, adding to cart, opening a chat, calling, requesting a quote. Effort is the strongest signal that intent is real.
Recency and frequency. How recently and how often the behaviour happened. Two visits in a day outweigh five in a year; a pricing visit last week outweighs one last quarter. Recency decays fast; most intent expires within days or weeks depending on the purchase. For people who have already bought, the same two signals plus spend are the basis of RFM segmentation.
Engagement depth. Time, scroll, video completion, email clicks — weaker signals on their own, useful combined with the others.
Context. Device, location, time of day, referral source. A "near me" search on a phone at lunchtime is a different moment from the same query on a desktop at midnight. Context qualifies other signals rather than standing alone.
Declared data. What people tell you in forms, surveys and preference centresA page where a subscriber chooses which messages, topics and channels they receive and updates their own details, instead of an all-or-nothing unsubscribe.: their situation, timeline, budget. Declared intent is the most reliable where it exists and should override inferred intent when the two conflict.
How channels use intent
- Search advertising targets intent directly through the query. Match types, negative keywords and bidding by query cluster are intent reading operationalised; the click-through calculator shows why intent-matched ads earn higher response.
- Organic search does the same through content: a page written for a problem-aware query will not rank for, or satisfy, a ready query.
- Retargeting (RemarketingAdvertising to previous visitors or engagers.) segments by observed behaviour: product-page visitors, cart abandoners, pricing-page visitors, with different messages and frequency for each level. One retargeting audience for "all visitors" throws the signals away. The remarketing expertise page shows how the author structures audience windows.
- Social and video platforms infer intent from on-platform behaviour and sell it as in-market or intent audiences, and extend it through lookalikes (Lookalike AudienceAudience similar to a seed list.) of people who converted — which are interest-plus-resemblance, not intent, and convert accordingly.
- Email and automation read intent from clicks and site behaviour to trigger the next message or to hand a lead to sales.
- Lead scoring in B2B combines signals into a number that decides when sales should call.
Scoring intent
A simple scoring model makes the inference explicit and testable:
- List the signals available in your data, by source.
- Assign each a weight reflecting how strongly it predicts conversion — effortful actions high, page views low, declared data highest — and a decay by recency.
- Sum per person or per session, and set thresholds for the levels.
- Calibrate against outcomes. Compare scores with what people subsequently did; adjust weights where high scores did not convert and low scores did. Conversion rate (CVRShare of visits or clicks that convert.) by score band is the check.
- Act differently by band: content and light-touch nurture for low bands, retargeting with proof for middle bands, reassurance, offers and human contact for the top band.
The model does not need to be sophisticated to be useful; it needs to be written down, checked and revised. The measurement plan records which signals are captured and named (consistent UTMURL tags identifying traffic source. tagging and event naming are what make the signals readable at all).
Reading intent for a business-insurance broker
Signals, in rising order of weight: reading an article on what cover a trade needs (problem-aware); comparing two cover types on a guide page (solution-aware); visiting the quote page and the reviews page in one session (product-aware); starting the quote form, then returning the next day and completing the first step (ready). Each level gets its own treatment: more guides for the first, a comparison email for those who left an address at the second, retargeting with a claims-handling testimonial for the third, a callback offer within the hour for the fourth. The scoring is checked quarterly against which bands actually bought; the reviews-page visit turns out to predict purchase more strongly than the quote-page visit, and its weight is raised.
Over-reading intent
The failure in the other direction is treating weak or accidental signals as decisions:
- One page view is not intent. Especially a pricing page, which curious people, competitors and recruiters all visit.
- Platform intent audiences are inferences about groups. The platform saw behaviour it associates with buying; it has not seen this person decide.
- Intent expires. Retargeting someone months after a visit reads stale intent as current and annoys the person.
- Interest is not intent. Content engagement shows a topic matters to someone, not that they will buy.
- Correlation drifts. A signal that predicted conversion last year may not now; recalibrate.
Intent in B2B and in long sales cycles
Business purchases stretch the signals over months and across people. A single visitor's behaviour reveals little; the pattern across an account does more: several people from the same organisation reading solution content in the same fortnight is a stronger signal than any one of them visiting the pricing page. Account-level intent combines website behaviour (matched by company where possible), engagement with emails and events, declared data from forms, and third-party signals about the organisation — hiring for a relevant role, a technology change, a published plan. Scoring is done per account, recency decays more slowly than in consumer categories, and the action at the top band is usually a conversation rather than an offer. The risk of over-reading is correspondingly larger, because a sales call prompted by a false signal costs more than a wasted ad impression; thresholds are set conservatively and calibrated against which accounts actually opened opportunities. The same logic applies to any long-cycle consumer purchase where research spans weeks: look for patterns over time, not single actions. The B2B and automation clusters will cover the lead-scoring mechanics; this page covers the reading. Whatever the category, the model is only as good as its last calibration against real outcomes. When a signal arrives on a record a salesperson already owns, the guide to sales alerts from marketing activity decides which single events justify interrupting them and which belong in a digest or the score.
For products bought once in many years, such as kitchens or roofs, the signals cluster around trigger events like moving home or a breakdown, and the guide to marketing products people buy rarely covers reading them alongside staying remembered between purchases.
Growth in low-intent visits is one of the causes covered in website traffic up but sales flat, alongside bots, broken purchase paths and sales that happen offline.
Common mistakes
- One message for all levels. Usually a ready-stage offer to everyone.
- Retargeting all visitors as one audience. The signals were there and were discarded.
- Scoring without calibration. Weights set by opinion and never checked against outcomes.
- Ignoring declared data. A form that says "not for six months" overridden by a page-view score.
- No recency decay. Last year's cart abandoner treated like yesterday's.
The touchpoints guide covers where these signals are captured; the lifecycle marketing guide applies the same reading to customers after purchase, where the signals are usage and silence.
