Did Andromeda make my Meta ad results worse?
The Meta pages read on 3 October 2026 do not say it did. Meta's engineering post says Andromeda improves its ads system by delivering more personalized ads and "maximizing return on ad spend for advertisers", but attributes no advertiser results to it. Before blaming it, check the status page, auction competition, significant edits and weekly averages.
The Meta pages read for this entry do not say that Andromeda made any advertiser’s results worse. Meta’s engineering post of 2 December 2024 describes Andromeda as its machine learning system design for retrieval, the first step in its multi-stage ads recommendation system. If your results fell, Meta documents other things to check first: the Meta status page, auction competition, significant edits, and whether you are reading daily or weekly numbers. This entry lists them in that order.
What does Meta say Andromeda does?
The post says retrieval selects ads “from tens of millions of ad candidates into a few thousand relevant ad candidates”. It reports Meta’s own figures for its system: a “+6% recall improvement” and “+8% ads quality improvement on selected segments”. The post also gives a ROAS figure for advertisers who had not used Advantage+ creative before and turned on its “AI-driven targeting features”. That sentence names Advantage+ creative, not Andromeda.
The post also says Andromeda “improves performance of Meta ads system by delivering more personalized ads to viewers and maximizing return on ad spend for advertisers.” That is Meta’s statement of what the system does, not a report of advertiser results.
As read on 3 October 2026, the post does not report how advertisers’ results changed after Andromeda, and does not mention ad similarity or account structure. The entry on what Andromeda is covers the post itself. The pages read for this entry do not describe a way to see Andromeda in Ads Manager, so the checks below use what you can see.
Is there a problem on Meta’s side?
The Meta status page shows the current status of Meta business products, including ads creation, delivery and reporting. Statuses run from “no known issues” to low, medium and high disruptions, and “resolved”. The view history link lists resolved issues from the past 90 days, and there is an RSS feed.
Compare the dates of any listed issue with the dates your results fell. If you hit a problem that is not listed, Meta’s page points you to business support in the Meta Business Help Center.
Did the auction get more competitive?
Meta defines auction competition change as the difference between the competitive bid on a day and the average competitive bid for the previous 3 days. It calls the change significant if it is over 20%. The competitive bid is made of “the cheapest 1% of the winning bids in the auctions you lost”.
Meta’s definition is built from bids in auctions, and the page does not mention your creative. You can read it for the days your cost per result moved. That page does not say what to change when it is high.
Did an edit restart learning?
Meta lists changes that are always significant edits: any change to targeting, ad creative or optimization event, adding a new ad to your ad set, pausing the ad set for 7 days or longer, and changing bid strategy. Budget and bid or ROAS goal changes depend on their size. Meta says only a significant edit makes an ad set re-enter the learning phase.
Meta’s page on updates to detailed targeting also lists dated changes: exclusions were removed from Ads Manager starting 31 March 2025, and a gradual consolidation of some interest options is dated 23 June 2025. Check your edit history against these dates. Meta says the “last significant edit” column in Ads Manager shows when you last made a change that could reset the learning phase.
Are you reading daily or weekly numbers?
Meta says daily budgets are an average over the week and may run up to 75% above budget on some days. It advises analyzing performance over at least a full week. For a cost per result goal it says to evaluate with weekly averages and to wait at least 7 days after any adjustment.
Meta also says performance is typically less stable during the learning phase, and lists constrained delivery and late-day budget or schedule changes as things to check.
What do these checks not tell you?
They do not prove or rule out Andromeda. As checked on 3 October 2026, the pages read for this entry do not document a way to test for it. They tell you whether a documented cause is present in your account. If none is, Meta does not document a further step for this question.
FAQ
Has Meta said Andromeda changes advertiser results?
Meta’s post says Andromeda “improves performance of Meta ads system” and maximizes return on ad spend for advertisers. As read on 3 October 2026, it reports a recall and an ads quality figure for Meta’s system and attributes no advertiser result to Andromeda, so this entry ties no change in your results to it.
Can I turn Andromeda off or check if it is running?
Meta’s post does not describe a setting for it. It presents Andromeda as part of how Meta retrieves ads before ranking, not as an Ads Manager option. Nothing in the pages read for this entry gives you a switch or a report for it.
How long should I wait before judging a change?
Meta says to evaluate a cost per result goal with weekly averages rather than daily results, and to wait at least 7 days after any adjustment before re-evaluating. Its fluctuations page says to analyze performance over at least a full week.
Sources
- Meta Andromeda: Supercharging Advantage+ automation with the next-gen personalized ads retrieval engine (2 December 2024) (Meta, checked 3 Oct 2026)
- About Meta status page (Meta, checked 3 Oct 2026)
- Auction competition change (Meta, checked 3 Oct 2026)
- Significant edits and learning phase (Meta, checked 3 Oct 2026)
- About the learning phase (Meta, checked 3 Oct 2026)
- Understand fluctuations in ad performance (Meta, checked 3 Oct 2026)
- About cost per result goal (Meta, checked 3 Oct 2026)
- Updates to detailed targeting (Meta, checked 3 Oct 2026)
Drafted with AI and checked claim by claim against the pages linked under Sources. Found a mistake? Tell us via the corrections policy.