Interstitials: closing the ad without losing the route
Audit interstitial exits and lost journeys: two worked publisher pilots, complete costs, reversals and clear stop or expansion criteria.
An interstitial interrupts a journey with a promotional screen. A visitor may close that screen successfully and still fail to reach the page they originally wanted. This guide separates those two outcomes and gives a publisher a practical audit, a measurement sheet and two complete training pilots. The objective is to protect the original task while assessing accepted affiliate results. More ad clicks alone are not evidence of a better experience or a profitable route.
Define the screen and the decision
An inline card stays inside the page; an interstitial occupies an intervening step. A popunder opens another browsing context. Those formats need different tests. Here the subject is a web journey on a site you control, with an optional advertising interruption between useful content and the next requested resource. This is not an application SDK tutorial or a recipe for intercepting every click.
Write down the original task before drawing the advertisement: for example, reading a checklist and opening a comparison. Identify the page that must remain reachable after dismissal. An advertisement link and a continue link must have different meanings. Never make the close control lead to an advertiser, disguise advertising as an operating-system warning or make access depend on submitting contact details.
Our editorial default is the uninterrupted route. Consider an alternative only after checking the applicable source and offer conditions, your authority to modify the site, and the implementation's effects on visitors. A frequency limit does not prove that an interruption is useful. If you cannot verify a supplier's exit behaviour, keep that supplier out of the pilot.
Separate documented facts from your test plan
Sources checked on 22 October 2026: Google Search Central discourages intrusive content obstruction; Better Ads Standards distinguishes desktop, mobile-web and app experiences. Neither provides blanket approval for your campaign.
Google Ad Manager's web documentation describes an exit option and frequency controls for its own format. This article does not promise account eligibility, integration or available inventory.
For custom dialogs, W3C APG describes keyboard containment, Escape and logical focus restoration. MDN documents native dialog behaviour and form-validation interactions.
WCAG target guidance explains a 24-by-24 CSS-pixel minimum with exceptions; focus guidance addresses fully obscured focused controls. Assess the complete interface, not just one measurement.
Everything below is an editorial testing method, not a platform requirement. There is no universal permitted countdown, optimal display rate or guaranteed search penalty here. Country, product and data-processing requirements need their own assessment before live use.
Prepare a placement passport
Collect the source page, intended destination, advertiser destination, approved copy, placement owner and person able to disable the screen. Record the campaign identifier, language, device scope, test dates, attribution window and acceptance deadline. Specify what happens when the ad cannot load: the visitor should retain an ordinary path to the requested resource.
Use this reusable brief: “Original task: open the comparison. Advertisement: optional planning-software demonstration. Dismissal: continue to the comparison. Promotional action: visit the named supplier. Measurement: eligible journey, visible screen, deliberate action, restored route, accepted result. Owner: named editor. Rollback: restore the original page version.” Replace the example with verified details before use.
Prepare a plain-text alternative and a no-screen control. Keep artwork lightweight and readable on a narrow display. Separate the advertising label, supplier identity, benefit, action and exit. These are proposed materials for your own site, not names of controls in AffiliateCore or another advertising console.
Check closure before spending
Test a small phone in portrait and landscape, a desktop keyboard, enlarged text and slow loading. Start at the original content, activate the intended transition, dismiss the screen and verify the destination and reading context. Repeat after returning from the advertiser. Record actual browser versions and reproducible steps rather than writing “mobile passed”.
In a custom modal, inspect the active focus position, movement between controls, Escape and restoration after dismissal. Also test the visible close action independently. A form must not make an empty required field a condition for leaving an advertisement. Do not assume native dialog markup automatically fixes the complete user journey.
Check a failed image, a delayed third-party response, browser back navigation and an unavailable destination. If a screen is owned by an external provider, test supported behaviour without changing its controls or circumventing restrictions. A reproducible trap, accidental advertiser navigation or loss of the original destination is grounds to disable the pilot immediately.
Build a small event ledger
Use your own documented event names, such as eligible_journey, screen_visible, close_requested, route_restored and deliberate_offer_click. These are analytical labels, not built-in AffiliateCore events. An eligible journey is one predefined original navigation attempt; a visible screen is not merely an ad request. Deduplicate repeated event delivery within that attempt.
A closure request is not successful restoration. Count the destination actually reached after dismissal, where measurement is available. Missing telemetry belongs in an unknown bucket: it is not automatically abandonment, fraud or refusal. Decide the observation window in advance and keep the same definition across variants. Avoid adding screenshots, email addresses or personal identifiers to tracking parameters.
In AffiliateCore, an applicable route starts with your material, continues through your existing campaign link and reaches the advertiser's landing page. Use it only where the offer and source have been accepted. The local campaign implementation supports sub1, sub2 and sub3; use descriptive experiment labels, not personal data. Screen display and closure remain your site's measurement responsibility.
If this route fits your work, prepare a publisher profile in AffiliateCore with your source and placement description. Registration does not guarantee a matching offer or permission to use this format.
In both exercises, closing, the direct promotional click and other outcomes classify the first interaction with the screen, once per journey. Recovery is a separate subsequent observation. If a visitor later follows an offer link from the restored resource, record that branch separately and attribute the final result only once. Do not add closure counts to advertising clicks and call the sum unique visitors.
Design the comparison before launch
Compare concurrent groups where possible, keeping source, content, offer and observation windows comparable. Allocate journeys consistently so repeated visits do not arbitrarily switch treatment. Choose a small predefined budget and a maximum exposure count. Do not keep a failing variant running until a convenient positive total appears.
Track both commercial and route outcomes. Recovery among closers equals restored routes divided by closure requests. Retained accepted results equal initially accepted results minus later reversals. Net contribution equals retained commission minus all allocated costs. Compare contribution per eligible journey as well as the aggregate total, because unequal group sizes can mislead.
For the training pilots below, each group contains 1,000 eligible journeys. All amounts are independent hypothetical USD budgets, not prices, exchange-rate conversions or AffiliateCore tariffs. Each group uses a seven-day enrolment period and a further fourteen-day observation period agreed for the exercise; real offer deadlines replace these assumptions. No pending results remain in the final examples.
Training pilot one: a comparison guide loses value
A publisher owns a guide comparing appointment-planning tools. The original resource is a feature checklist. The proposed screen promotes an optional software demonstration. Materials include the guide, the checklist, a supplier-approved claim sheet, two page versions and a ledger linking anonymous experiment labels to final result counts. The offer pays an assumed USD 20 for a qualifying demonstration request retained after review.
Suggested screen copy: “Advertisement. Explore the supplier's planning-tool demonstration. See eligibility and conditions on the supplier's page.” The promotional action reads “View demonstration details”; dismissal reads “Continue to the checklist”. Both are editorial mockup labels. No countdown, false urgency, notification prompt or hidden navigation is part of the experiment.
The no-screen control produces 150 deliberate offer clicks and 30 submitted requests. Six are rejected during validation, leaving 24 initially accepted; two later cancellations leave 22 retained. The screen group records 1,000 visible screens, 650 closure requests, 150 deliberate promotional clicks and 200 other or unknown outcomes. Of the 650 closers, 520 reach the original destination within the defined window; 130 lack confirmed recovery.
The screen group's 150 clicks produce 25 requests. Eight are rejected, leaving 17 initially accepted; three later cancellations leave 14 retained. “Rejected” and “cancelled later” are separate stages, so neither is deducted twice. These counts illustrate an analytical ledger, not observed customers or a prediction for this advertising format.
Allocate USD 240 traffic acquisition, USD 60 content labour, USD 40 testing and implementation, and USD 20 measurement and review to each group. Total cost is USD 360 per group, USD 720 combined. The equal allocations deliberately include the control's own testing work. There are no additional fees in this hypothetical budget; any real taxes or supplier charges must be added before a live decision.
Control revenue is 22 × USD 20 = USD 440; contribution is USD 80 and cost per retained result is USD 16.36. Screen revenue is 14 × USD 20 = USD 280; contribution is minus USD 80 and cost per retained result is USD 25.71. Break-even requires 18 retained results per group. The screen reduces contribution by USD 160 at equal exposure.
Recovery is 520 ÷ 650 = 80%; unconfirmed recovery is 20% among closers, not 20% of all visitors. Stop the screen, preserve the control and inspect the missing-route sample before discussing another test. Even if the 130 missing observations turn out partly to be measurement loss, the completed economic cohort does not justify scaling. A repaired variant needs a new budget and new comparison.
Training pilot two: repair first, then assess a limited expansion
A second publisher runs a home-office guide and an optional demonstration offer for document-workflow software. The chosen next resource is a downloadable checklist that remains available without advertising engagement. The material pack includes the checklist, accessible screen copy, a destination map, a failure log, approval criteria and a record of the corrected page version.
During private testing, closing a custom screen removes the overlay but leaves an invisible layer above the page. A tester can no longer activate the checklist link. Disable that version before paid exposure. Ask the site owner to correct the layer cleanup and restore a usable destination; repeat keyboard, phone and failed-load tests. Do not count this private defect discovery as a completed commercial pilot.
Mockup copy for the repaired candidate: “Advertisement. Compare document-workflow features with the supplier. The checklist remains available.” Actions: “Read supplier details” and “Continue to the checklist”. The comparison starts only after both actions work as described. Keep the same advertiser, source and acceptance rules across groups; a different offer would confound the screen evaluation.
The control's 1,000 journeys produce 140 deliberate clicks and 24 requests: four rejected, 20 initially accepted, two later cancellations, 18 retained. The repaired screen's 1,000 displays produce 700 closures, 180 deliberate clicks and 120 other or unknown outcomes. Among closers, 665 recover the original route and 35 lack confirmed recovery. Its 32 requests yield six rejections, 26 initial acceptances, two later cancellations and 24 retained results.
The assumed commission is USD 25. Each group costs USD 300: USD 180 traffic, USD 50 content labour, USD 50 implementation and QA, USD 20 measurement and review. The implementation allocation includes the repair effort; the combined experiment costs USD 600. Control revenue is USD 450 and contribution USD 150. Candidate revenue is USD 600 and contribution USD 300. Cost per retained result is USD 16.67 versus USD 12.50; break-even is 12 retained results.
Recovery among closers is 665 ÷ 700 = 95%, with 5% unconfirmed. Contribution improves by USD 150 per 1,000 eligible journeys. For this exercise, the predeclared continuation gate is recovery of at least 95%, no reproducible traps, a fully reviewed cohort and nonnegative incremental contribution. This is a management choice, not an industry benchmark or statistical significance test.
The candidate meets that limited gate but remains a small sample. Authorise only another comparable batch capped at USD 300 for the candidate, retaining a separately budgeted control. Do not extrapolate daily income. If four additional candidate results reverse, revenue becomes USD 500 and contribution USD 200; eight additional reversals leave USD 400 revenue and reduce contribution to USD 100, below the original control's USD 150. Reconcile both groups again at the same later cutoff.
Diagnose discrepancies and hand over the decision
High closure volume can mean visitors deliberately prefer the original resource. Do not “fix” that by hiding the exit. Low recovery may reflect navigation defects, slow destinations or incomplete instrumentation. Review those explanations separately. A sudden conversion increase alongside repeated accidental taps is a reason to investigate quality, not to increase exposure.
Your handover sheet should contain the version identifier, source cohort, event definitions, unknown counts, expense allocation, rejected and reversed results, and the agreed final observation date. Attach a short recording of reproducible defects without visitor data. Keep the rollback owner reachable while the experiment is active. A direct partner programme can use the same ledger when no suitable network offer exists.
Existing partners can sign in to AffiliateCore to check their campaign link and available conversion records against the agreed terms. The cabinet does not replace your closure audit or prove that an external screen is accessible.
For the next working session, choose one original journey, write both action labels, test every exit, and calculate the result after reversals. Keep the no-screen route if the evidence remains incomplete. Related reading: in-page advertising and deliberate clicks. Its inline-card audit complements this interruption-specific review without replacing the close-and-return tests.