Blog Reader Retargeting on Meta Ads: Content-Based Audiences

A blog visit indicates contact with a topic, not verified reading, profession or purchase intent. Retargeting can test whether a relevant next step helps an eligible reader. Define a permitted signal, match it to a truthful offer and measure qualified or paid outcomes before calling the audience valuable.

Match a topic to a useful next step

These are hypotheses. A comparison article can attract researchers or competitors; reading an advanced guide does not prove the reader is a media buyer. NovaStorm’s platform access is free under its current offer; a signup is not a paid subscription. Generation-credit purchases are a separate commercial outcome.

An educational active-time and scroll implementation

The example below emits an adapter event only after sixty seconds of consented foreground activity and at least 75% article depth. Both thresholds are chosen heuristics, not proof of reading or qualification. Activity expires after fifteen seconds without input. The shared in-memory seen set suppresses duplicate emissions for an article during the current page lifecycle.

The Page Visibility API provides the visibility signal; browsers may throttle background timers. Wire hasConsent to the current permitted-purpose state and emit to an approved measurement adapter. This code does not install Meta Pixel, create an audience or persist a cross-device reader identity.

function startBlogRead({ article, articleId, category, hasConsent, emit, seen }) { let activeMs = 0, maxScroll = 0, last = performance.now(); let lastInput = -Infinity, allowedBefore = false, disposed = false; const permitted = () => hasConsent() && document.visibilityState === 'visible'; const sample = (now) => { if (disposed) return; const allowed = permitted(); if (!hasConsent()) { activeMs = 0; maxScroll = 0; lastInput = -Infinity; } if (allowed && allowedBefore) { activeMs += Math.max(0, Math.min(now, lastInput + 15000) - last); const rect = article.getBoundingClientRect(); const depth = Math.max(0, Math.min(1, (innerHeight - rect.top) / rect.height)); maxScroll = Math.max(maxScroll, depth); if (activeMs >= 60000 && maxScroll >= 0.75 && !seen.has(articleId)) { seen.add(articleId); emit('BlogRead', { article_id: articleId, category, active_seconds: Math.floor(activeMs / 1000) }); } } last = now; allowedBefore = allowed; }; const activity = () => { const now = performance.now(); sample(now); if (!disposed && permitted()) lastInput = now; }; const tick = () => sample(performance.now()); const resetClock = () => { last = performance.now(); allowedBefore = false; }; for (const type of ['pointerdown', 'keydown', 'scroll']) window.addEventListener(type, activity, { passive: true }); document.addEventListener('visibilitychange', resetClock); const timer = setInterval(tick, 1000); return () => { disposed = true; clearInterval(timer); for (const type of ['pointerdown', 'keydown', 'scroll']) window.removeEventListener(type, activity); document.removeEventListener('visibilitychange', resetClock); }; }

Create one tracker per article view with a shared new Set() for that page lifecycle; call its returned disposer on navigation/unmount. Supply a positive-height article element and controlled article/category identifiers. Permission withdrawal resets accumulated time/depth and stops new emissions; it cannot retract an event already sent. Reinitializing an entire page creates a new lifecycle, so collection-side event identity/deduplication needs its own design.

Acceptance checks before collection

• Consent denied: no event; pre-consent activity does not count.
• Hide the tab, wait, then return: hidden time must not increase active time.
• Stop interacting: after fifteen seconds, additional idle time must not count.
• Meet both thresholds: one event; repeated ticks or the same article key do not produce additional events in the shared seen set.
• Withdraw consent: no new emission; dispose on navigation so old listeners stop observing a new page.

This educational implementation passed local simulated checks for denial, hidden-time exclusion, threshold emission, duplicate suppression, withdrawal and disposal. That is a code-level result, not a production analytics deployment or a measured reader-quality improvement.

Validate the actual audience rule

Meta’s audience schema supports website sources and rules. The presence of an event parameter is not proof that BlogRead/category rules are available in your setup. Save the accepted source, rule, retention and exclusions; mark unavailable criteria explicitly. An eligible URL-based audience can be a fallback if its collection and configuration are supported.

A count of three articles requires distinct article identities and a permitted persistent/session model. Reloads do not establish three articles, and a time threshold does not establish intent. Do not assign an occupation or a universal high-intent score from those signals.

Keep the paid-outcome loop small and honest

Hypothetical campaign: a budget-guide audience receives a factual budget-review offer. Cap media at $120, use a consistent destination and wait through the declared outcome window. If four mature paid service orders occur, media cost is $30 per order; if only two occur, $60. These assumed counts are noisy and do not prove incrementality or a stable audience advantage.

Record eligible population, actual spend, unique qualified/paid outcomes, returns and contribution. Stop at the cash cap, investigate failures and allow an inconclusive result. Use the small-budget evidence guide rather than spreading a tiny audience across many categories.

Prepare a paused NovaStorm campaign draft around the chosen content-to-offer hypothesis. BlogRead instrumentation, consent wiring, accepted audience rules and outcome reconciliation require their own implementation; this article does not claim NovaStorm already collects those events.