Can a new Google Ads campaign learn from offline conversions uploaded before their action became Primary?

Preserve the valid conversions, but do not base the new campaign's launch on assumed reuse of their Secondary-era history. Google supports historical imported data helping Smart Bidding in general. An older practitioner's report addresses promotion more directly and says that previously stored Secondary data would not be used afterward. The evidence does not confirm how that exact transition works today.

The useful decision is firmer: preserve truthful events, make sure the new campaign pursues the intended conversion action, and keep the import stream reliable. Retracting and recreating valid events is not a documented way to improve their historical usefulness.

Follow the same 22 conversions

Suppose you import 22 completed sales on Monday into an action called Completed sale, which is Secondary. The upload results show accepted rows, and you can reconcile the sales with the action's reported conversions. On Tuesday you make that same action Primary. On Wednesday you prepare a new campaign, but the older sales still appear only in All conversions.

There are four questions inside that one observation:

Google describes Secondary actions as observation only under standard goals. Primary actions become bidding objectives when the campaign uses their standard goal. A selected custom goal is an exception: included Secondary actions can also be used for bidding. These are the documented action and goal rules.

The report of Monday's sales therefore has a limited message. It shows where the sales are counted. It does not, by itself, prove that Tuesday's promotion has made their history usable—or permanently unusable.

What supports historical learning, and what remains unresolved

Google's Changing Biddable Conversion Actions deck, page 17, asks whether Smart Bidding will use historical data uploaded when transitioning to an imported conversion action. Its answer begins: “Smart Bidding will use this data.” It favors consistent, frequent uploads because they help the system understand conversion delay. The deck's publication date is not established.

That is substantive affirmative evidence for historical imports helping. It is stronger than a promise that future conversions will work.

The same deck ties training to the Conversions or conversion-value columns on page 12. Its passages do not settle our full sequence: accepted as Secondary, promoted in place, then used by a new campaign. Nor do they promise retrospective movement between report columns.

Frederik Boysen's public account addresses the closer question: whether stored Secondary data becomes usable after promotion. He reports asking an unnamed global Smart Bidding product lead at a Dublin roundtable and receiving a negative answer, with a change still being developed. When checked on September 19, 2026, the page displayed an approximate age of two years. No transcript or confirmation that the contemplated capability had shipped was available in that source check.

That report gives a specific reason to doubt automatic reuse, beyond the gap in the general backfill guidance. It is more directly applicable to this sequence, but it is an older account of a conversation rather than a current formal specification. It cannot establish a definitive current no. Equally, the general affirmative import passage does not cancel its warning about data collected while Secondary.

Google's campaign-transition guidance recommends checking that new actions report correctly before making them Primary. Collecting observations before promotion is therefore a supported preparation step.

A campaign's bidding objective is also different from information that can help its predictions. Google's Primary/Secondary guidance says that Primary actions not used for optimization may help enhance predictions.

For Wednesday's launch, do not count the 22 sales as a confirmed source of bidding history. None of the evidence above confirms retrospective use of those accepted Secondary events after promotion, and counting visible rows cannot settle it.

Make the campaign pursue the right sale

Return to Wednesday. Suppose the new campaign still selects only Submit lead form. Making Completed sale Primary has not selected its goal for this campaign. Re-uploading the sales would not fix that choice.

Open the new campaign's Settings → Conversion goals. Check whether it uses account-default goals or campaign-specific goals, and inspect the actual actions inside the selected goals. For the ordinary standard-goal route, the intended sale action must be Primary and its goal selected. Campaign-specific settings can override account defaults. Google documents the campaign goal controls here.

If the campaign already selects the correct goal and action, that configuration problem is solved. An unchanged historical column is not a reason to keep changing goals.

A custom goal is useful when you deliberately need a particular combination of actions that standard goals do not express. It can include a Secondary action, but it does not document a way to unlock earlier rows. Adding form submissions to a custom goal alongside sales would also tell the campaign to optimize for both. Choose that combination only if both outcomes belong in its objective—not to make the history look fuller. Google's custom-goal guidance.

Re-uploading changes a different thing

An ordinary repeat upload does not make an existing sale fresh. For click-based offline imports, Google documents duplicate handling using the identifier, conversion name and conversion time. Sending the same event again is not a second opportunity to classify it for bidding. Offline import guidelines.

Retraction goes further: it withdraws the recorded conversion. Google says subsequent adjustments to that retracted event are ignored. Its current guidance does describe replacement uploads: a changed timestamp, plus a new unique order ID when an order ID was used. Those are new events; they do not reopen the original. About conversion adjustments.

Google's transition deck discourages deleting and replacing history to introduce new values because it may misteach conversion delay. This concerns a related operation; it does not test these 22 sales or establish a benefit from recreating them. Deck, page 18.

This is why preservation holds under either answer to the unresolved question. If the earlier sales can help, their intact records preserve that possibility. If they cannot help through this transition, a new identifier or timestamp does not demonstrate that the limitation has been safely repaired. You would be changing known, truthful records to pursue an unverified benefit.

Now change one fact: one of the 22 sales was actually a test transaction incorrectly imported as a real sale. That gives you a reason to correct the event, independent of its Primary/Secondary setting. If the sale was real but its value was wrong, a value restatement may be the appropriate correction. Confirm the original record and the adjustment support for your import method before acting. If you cannot identify the original event reliably, retrieve the upload details first; do not guess an ID or manufacture a replacement batch.

Verify the records and the next uploads

If “22 accepted sales” is still an assumption, first identify the action's conversion type and import method. The name Completed sale does not make it Google's specialized Store sales type. Google excludes Store visit and Store sales conversions from its by-conversion-time columns. Conversion reporting definitions.

For an ordinary offline import from ad clicks that is covered by those columns, add All conv. (by conv. time) to the Campaigns report and segment by conversion action. Match the date range to the actual conversion dates and compare with the upload results file, including rejected rows. This is Google's recommended import reconciliation. Ordinary conversion columns generally use the ad-interaction date, so an upload-date comparison can send you looking in the wrong period.

If your action is excluded, or you cannot establish that those columns support it, inspect its actual data source and any import receipts and diagnostics that source provides. Keep truthful records and leave the occurrence-date comparison unresolved. A missing total in this situation does not establish zero conversions, a rejected upload or whether historical events trained bidding.

For imported records, if you lack the results file or equivalent import receipt, ask the importer or the person who manages the integration to recover the import details. Until those are available, keep the existing events and treat acceptance as unverified. A missing receipt is a reason to investigate the feed, not retract everything visible in the report.

For subsequent real sales, verify acceptance and reporting under the intended action and campaign goal. Continue regular uploads; Google's import FAQ recommends daily uploads when possible, otherwise a consistent interval. If new accepted conversions still appear only in All conversions, revisit the selected goal, action setting and reporting scope before changing historical events.

Twenty-two events alone establish neither campaign readiness nor likely performance. Conversion cycles include the delay between a click, the outcome and its reporting to Google; assess campaign results with that delay in mind. Google's bidding-learning guidance explains how to use the bid strategy report where delay information is available.

If confirmation of retrospective use is essential to your launch decision, ask Google support about the exact sequence, supplying the action ID, import method, original event and upload dates, promotion time, and new campaign's goal settings. Ask specifically about training on already accepted Secondary events after promotion—not merely whether Primary actions are eligible. While that question is open, the valid sales can stay exactly as they are.