- Published
- Oct 29, 2026
- Reading time
- 11 min read
Prebid Local VAST Cache: Diagnose Missing AdX Demand
Audit a Prebid local VAST cache migration when video plays but AdX delivery falls. Trace player signals, GAM requests and controlled acceptance checks.

In this guide (10 sections)
If Prebid video starts working after a local-cache migration but AdX delivery falls, inspect who now makes the GAM request. Receiving playable VAST XML is one acceptance check. Preserving the player-added signals, auction path and measurement population is another. A repaired creative fetch does not establish that the migration preserved monetization behavior.
An older public publisher question about Prebid, local VAST caching and AdX describes that exact sequence: changing from a video-tag URL to getVastXml and supplying the response through IMA restored Prebid playback, while the publisher reported much less AdX delivery. The report establishes a real troubleshooting question, not a verified loss estimate or a universal outcome. The original poster later reported that a self-hosted cache made their setup work again. That is their reported resolution; the public thread supplies no controlled request comparison or general demand-recovery guarantee.
Separate the cache from the request path#
There are three decisions to document: where the winning creative is stored, which component requests GAM, and what the player receives. Calling all three “the cache” hides the migration's important changes.
The current Prebid video-with-GAM tutorial describes a local path that stores VAST in browser memory. getVastXml requests GAM and substitutes matching local creative content into the returned response. GAM is still involved; local caching does not inherently remove its auction.
The same tutorial warns that players cannot augment the GAM URL in this path as they do when receiving an ad-tag URL, and says lower AdX demand is likely. Treat this as a documented integration risk to investigate, rather than proof that it explains your campaign's entire change.
Write a before-and-after table for your implementation:
| Checkpoint | Before migration | After migration |
|---|---|---|
| Creative storage | Actual remote or bidder-managed location | Actual local or remote location |
| GAM request initiator | Player, SDK or another component | Observed component |
| Player input | Tag URL or another supported input | XML document or another supported input |
| Player-added request information | Captured permitted fields | Preserved, changed or unavailable |
| Auction arrangement | Documented competing demand | Verified arrangement |
| Playback and reporting | Defined events and reports | Same definitions or stated changes |
These are investigation categories, not a template that assumes every publisher previously used the same architecture.
Trace the request after the cache change.
Use the checklist to record who makes the GAM request and what the player receives.
Use the tracking QA checklistOpens a practical guide. No email required.
Confirm the exact implementation before choosing a repair#
Record the deployed Prebid.js build and modules, player, SDK version, cache configuration and relevant GAM line items. Include the change timestamp and affected inventory. Keep the exact released configuration, rather than relying on a developer's current local copy.
The getVastXml API reference requires the gamAdServerVideo module. It describes fetching and processing the GAM response, with local-blob substitution when cache.useLocal is enabled. Its arguments also allow publisher parameters to override values supplied by Prebid. Review those overrides when comparing requests.
Do not copy a tutorial's old player-library versions into a new release merely because its architecture is relevant. Confirm the supported behavior of the build actually deployed. The existence of an API method in current documentation does not establish that your particular player integration uses it correctly.
Also distinguish VAST creative storage from bid reuse. Prebid's configuration reference describes cache.useLocal for storing VAST blobs and useBidCache for reconsidering eligible earlier bids. Changing one is not a substitute for diagnosing the other.
Trace the player boundary and the GAM boundary separately#
Google's IMA HTML5 AdsRequest reference distinguishes an adTagUrl, which the SDK requests, from an adsResponse, which supplies a VAST response already available. It also exposes player-state methods and content properties used in ad requests. Passing XML is therefore a different boundary from asking the SDK to fetch a tag.
Start with an authorized test page and one reproducible opportunity. Capture the sequence from bid request through the GAM request, returned VAST, player load and defined playback events. Record which component initiated each network request. Preserve request timing so you can distinguish a changed initiator from a duplicate call.
Then compare the actual outgoing GAM requests. Check the ad unit, custom targeting, format and dimensions, page/content context, privacy state and player information applicable to that integration. Classify differences as intended, unexplained or unavailable for inspection. An absent field is a lead; confirm its documented purpose before calling it the cause.
Keep sensitive values in restricted evidence. Redact signed URLs, consent strings, identifiers and commercial configuration before sharing a support example. A sanitized comparison can show that a field changed without disclosing its original value.
Do not reconstruct protected SDK-generated signals by guessing a parameter name from a forum reply. Obtain the supported integration method from the player, SDK or platform owner. Copying a token from one request to another does not establish a valid or privacy-consistent request.
Identify the earliest stage that changed#
A lower AdX impression total can result from several different changes. Find the first observable difference before adjusting auction settings.
| Observation | Next evidence | What it does not prove |
|---|---|---|
| Fewer GAM requests per eligible opportunity | Request scheduling and initiator trace | Buyers rejected otherwise identical requests |
| Request count stable, relevant request information changed | Before/after request contract and platform diagnostics | One missing field explains every lost impression |
| GAM returns an ad, but playback fails | Returned VAST, player error and selected media | The auction had no demand |
| Playback evidence stable, report total changes | Report scope, metric and processing window | The cache damaged delivery |
| Demand share changes while total delivery holds | Demand-source mix and comparable rates | Publisher revenue necessarily improved |
For errors after a response exists, use the VAST error and timeout workflow. For reports counting different boundaries, use the ad-server impression discrepancy guide. Keeping these investigations separate prevents an auction diagnosis from becoming a generic tag rewrite.
Inspect the GAM decision using the evidence the account exposes. Confirm whether the intended Prebid line item could compete and which line item was selected. Request-level targeting is not itself a rendered-impression count, and a selected line item is not proof of successful playback.
Test one architectural change at a time#
Treat a cache migration as an acceptance test with more than “the video played” as its exit criterion. Establish permitted traffic, an owner and a rollback point before the test.
First verify the request route and a playable response with a small controlled sample. Then compare delivery and demand mix over comparable populations. Where an experiment is appropriate, keep browser/device mix, inventory, player behavior, privacy state, floors, timeouts and GAM configuration consistent except for the migration being tested. Record remaining differences explicitly.
Use eligible video opportunities as one denominator and GAM requests as another. Count completed playback separately if it matters to the test. Specify whether a player replay, retry, second ad in a pod or separate content session creates a new opportunity. Otherwise an apparent fill improvement can be a change in what the counter includes.
A test should cover the environments you actually serve. A successful desktop test does not establish behavior on a mobile browser, embedded webview, native app or SSAI path. This guide concerns the documented Prebid.js/client-player arrangement; those other integrations require their own supported request and cache contracts.
Do not relax consent controls, falsify player state or manufacture requests to recover a preferred demand mix. Preserve the real environment while finding a supported way to transmit its information.
A hypothetical comparison: playback recovered, demand mix changed#
Suppose two hypothetical, comparable test groups each contain 10,000 eligible single-ad video opportunities in an agreed UTC window. Each opportunity generates exactly one GAM request. Assume the demand-source impression groups below are disjoint and use the same counting definition.
| Observation | Reference path | Local-cache candidate |
|---|---|---|
| Eligible opportunities | 10,000 | 10,000 |
| GAM requests | 10,000 | 10,000 |
| Counted Prebid delivery | 4,000 | 6,500 |
| Counted AdX delivery | 4,500 | 1,500 |
| Other counted delivery | 500 | 500 |
| Opportunities without counted delivery | 1,000 | 1,500 |
The reference path has 9,000 counted impressions; the candidate has 8,500. Defined delivery per opportunity is therefore 90% and 85%. AdX delivery per opportunity changes from 45% to 15%, while Prebid delivery increases from 40% to 65%.
Those figures establish differences in this example's measured populations. They do not establish revenue, a percentage loss of bidder demand, or the cause. Counted AdX delivery also depends on competition and downstream delivery; it is not a direct count of every buyer's willingness to bid.
The next check is the request and decision evidence: did the two groups actually carry comparable context and maintain the intended auction setup? Separately check whether the 500 additional opportunities without counted delivery had no returned ad, failed during playback, or remain unresolved. Do not assign every unmatched opportunity to the cache without a trace.
Choose a supported migration path, not a comforting workaround#
A compatible remote cache, bidder-managed creative storage where supported, and a local-cache/player arrangement have different operational contracts. Assess availability, access controls, capacity, browser behavior, creative retrieval and the request path for the specific option being considered. A remote endpoint is not interchangeable merely because it can return XML; the Prebid configuration reference requires a compatible cache interface.
A sequential fallback also changes the question. If one demand path runs first and another is requested only after it fails, they are no longer being compared in the same simultaneous auction arrangement. That may be an intentional design, but do not describe it as preserving the previous competition without checking the actual decision sequence.
Likewise, enabling raw-VAST support is not evidence that GAM, the player and AdX retain the same information. A setting can solve the creative-input requirement while leaving another handoff unresolved.
Accept a candidate only when its observed request path, permitted signaling, demand arrangement and playback meet the documented requirement. A rollback is useful evidence too: preserve the results and explain which boundary changed, rather than claiming that one cache type is universally superior.
Prepare a support packet that names the missing handoff#
Include the deployment version and timestamp, affected inventory, expected versus observed request initiator, redacted sample traces, GAM decision evidence, final player input and the reports used for comparison. State the exact outcome that needs confirming: for example, whether the supported player path can preserve the required request information while using the proposed creative storage.
Keep commercial questions separate. To evaluate revenue after the technical repair, use comparable monetization records and the publisher revenue investigation guide. More Prebid impressions or fewer network fetches do not independently establish higher earnings.
For AdProof-measured campaigns, export the applicable campaign delivery record with explicit dates, timezone and supported filters. Place it beside the publisher's player and auction evidence. Our anonymized streaming-publisher verification case study illustrates scoped delivery review; it does not establish a native Prebid cache debugger or direct AdX integration. Request a review with the deployment window and the delivery question you need to reconcile.
Frequently asked questions#
Does getVastXml bypass Google Ad Manager?
Not in the documented local-cache flow described here. It requests GAM and processes the returned VAST, substituting matching locally cached creative content where applicable. Verify your integration's actual network sequence rather than assuming that XML input means no GAM auction occurred.
Does a working Prebid video prove the migration is complete?
No. It verifies part of the creative and playback path. Also check the GAM request initiator, supported player information, intended auction competition and comparable delivery reports before accepting the migration.
Is useBidCache the same as cache.useLocal?
No. Bid reuse and browser-local VAST creative storage address different tasks. Read the configuration documentation for your deployed build and inspect the setting responsible for the changed behavior.
Can I fix missing AdX delivery by copying request parameters manually?
Do not assume so. Identify the missing information and obtain a supported implementation from the relevant platform or player owner. SDK-generated identifiers, player state and privacy signals must not be fabricated or copied blindly. Verify the resulting request and delivery with a controlled test.
Research and scope#
Reviewed 29 October 2026. The linked original is an older troubleshooting question, not proof of a new outage, current search volume or a typical demand-loss percentage. Primary references were opened from Prebid and Google. The diagnostic sequence is a recommendation; the comparison is hypothetical. No customer results or guaranteed demand recovery are claimed.
Trace the request after the cache change.
Use the checklist to record who makes the GAM request and what the player receives.
Use the tracking QA checklistOpens a practical guide. No email required.
