Meta Ad Auction: How Bid, Relevance and Value Work
The Meta ad auction doesn't hand the win to whoever bids the most. It ranks every eligible ad on total value — your bid combined with Meta's estimated action rate and an ad quality score — and whichever ad has the highest total value gets the impression.
That's the short version. Here's what each input means, why sharper creative beats a bigger budget, and where the auction connects to decisions you're already making in Ads Manager.
How does the Meta ad auction actually work?
Every time someone opens Facebook or Instagram, Meta runs an auction for their next impression. Every eligible ad — based on targeting, budget, and pacing — gets scored, and it happens in roughly milliseconds, thousands of times a second.
The score each ad gets is total value: not the bid, not a popularity score, but a combination of three inputs Meta has stated publicly:
1. Your bid. What you told Meta you're willing to pay for the optimized result, either directly (bid cap) or implicitly through a bid strategy like Highest volume.
2. Estimated action rate. Meta's prediction of how likely this specific person is to take your optimized action — a purchase, a lead, a link click — if shown this specific ad, right now, on this device and placement.
3. Ad quality. A signal built from how people respond: positive engagement relative to what Meta expected, and negative signals like hides, "why am I seeing this ad," and reports.
Estimated action rate and ad quality together are what Meta calls ad relevance. Ad relevance is the mechanism that stops the highest bidder from always winning — without it, the biggest budget would buy every impression regardless of whether the ad was any good for the viewer.
What is total value in the Meta ads auction?
Total value is the single number each ad gets scored on. Meta has never published the exact formula or the internal weights it uses to combine the three inputs — that math is proprietary and it isn't static.
What the industry uses as a teaching model, because Meta has confirmed the three inputs and their general relationship, looks like this:
Total value ≈ (bid × estimated action rate) + ad quality
Treat that as a mental model for how the pieces interact, not Meta's real math. The actual system almost certainly weights these differently by objective, uses more granular sub-signals, and changes over time as Andromeda and GEM (more on those below) get retrained.
Here's a worked example that shows why the model is useful even though the exact numbers aren't real:
Advertiser A bids $8 per result, with an estimated action rate of 3% (0.03) and an ad quality score of 1.2 (strong engagement, no negative feedback). Total value ≈ (8 × 0.03) + 1.2 = 1.44
Advertiser B bids $15 per result — almost double — but their estimated action rate is only 1.5% (0.015) and their ad quality score is 0.6 (some hides, mediocre engagement). Total value ≈ (15 × 0.015) + 0.6 = 0.825
Advertiser A wins the impression at just over half the bid, because relevance did more work than budget. This is the entire reason a lower bid routinely beats a higher one when the lower-bid ad is more relevant.
Why does ad quality matter so much?
Ad quality is built from how real people respond to your ad, and it's the input advertisers most often ignore because it doesn't show up as a single column in Ads Manager the way bid does.
The signals feeding it: positive engagement relative to what Meta expected, negative feedback (hides, "why am I seeing this ad," reports), penalties for clickbait or engagement-bait tactics, and landing page experience after the click — speed, mobile usability, and whether the page matches what the ad promised.
An ad can have a great hook and still score poorly on ad quality if the landing page is slow or bait-and-switches the visitor. The auction doesn't stop scoring you at the click.
How does estimated action rate get calculated?
Estimated action rate is roughly Meta's estimated click-through rate multiplied by its estimated click-to-conversion rate for your optimized event. It's fed by the viewer's past behavior, that specific ad's performance history, delivery context (device, placement, time of day) — and, since Andromeda, the creative itself.
That last input is the newer, bigger one. Andromeda reads the actual creative — hook, format, on-screen talent, copy, even the landing page — and factors it directly into the prediction, rather than relying mostly on audience-level signals. That's the entire reason media buyers now say creative is the new targeting.
Practically: two ads running to the identical audience, identical objective, identical bid, can get meaningfully different estimated action rates purely because one creative is a stronger match for what Meta's models predict this viewer will act on.
This is where a research habit pays off. To see what creative patterns are earning strong relevance for others in your space, how to see competitors' Facebook ads covers reading a rival's Ad Library results for hooks and formats worth testing.
Which bid strategy fits which auction goal?
Your bid strategy decides what number the auction sees as your bid input, and Meta renamed all of these around 2022 — old screenshots use different names than what's in Ads Manager now.
| Bid strategy | What it actually controls | Watch out for |
|---|---|---|
| Highest volume (default; was "lowest cost") | Spends your full budget for the most results it can get | No cost ceiling — cost per result can drift as competition shifts |
| Cost per result goal (was "cost cap") | An AVERAGE cost per result, not a hard per-auction limit | Can land ~15-25% lower cost per result once it has enough data (vendor-reported, not a guarantee) |
| Highest value | Most total value from a value-optimized campaign | Requires value optimization set up first |
| ROAS goal / minimum ROAS | A target return on ad spend | Requires value optimization; can restrict delivery hard if the target is unrealistic |
| Bid cap | A manual ceiling per individual auction | Most control, most limiting — set it too low and delivery stalls |
Value optimization is the prerequisite that turns on the value and ROAS-based strategies, and it lets you set value rules so Meta bids more aggressively for higher-value segments within your audience.
How does the learning phase interact with the auction?
Every ad set needs roughly 50 optimization events within about 7 days of its last significant edit to exit the learning phase. Until then, Meta is still gathering the data that feeds estimated action rate for your creative-audience-event combination, and the auction score is less reliable.
"Learning Limited" means the ad set didn't reach that ~50-event threshold. Meta is explicit this isn't a penalty — it just means there isn't enough signal yet. The usual fixes: raise budget, broaden targeting, consolidate ad sets, or optimize for a higher-funnel event that fires more often.
A handful of changes reset learning: editing targeting, editing creative (including just adding a new ad), changing the optimization event, changing bid strategy, a large budget change, or pausing for roughly a week or more. There's no official percentage for what counts as a "large" budget change — the ~20% figure that circulates is a community rule of thumb, not a published Meta number.
How did Andromeda and GEM change what wins?
Meta has rebuilt the systems behind the auction twice in recent years, and both changes are documented in Meta's own engineering posts.
Andromeda (2024) is a retrieval engine — it filters tens of millions of possible ads down to a shortlist of a few thousand real candidates per impression, on custom hardware built for the job. Meta reports it improved ad quality by about 8%.
GEM (Generative Ads Recommendation Model, mid-2025) is the ranking model that scores that shortlist, reportedly lifting conversions roughly 5% on Instagram and 3% on Facebook Feed. An earlier system, Lattice, was reported to improve ad quality about 12% and conversions about 6%.
For a media buyer: targeting granularity keeps mattering less, creative keeps mattering more, because the ranking layer reads the creative itself. Advantage+ Audience reflects the same shift — your interest, custom, and lookalike picks are now suggestions, not hard rules; only location and minimum age stay fixed. You still control budget, objective, optimization event, and creative volume. Meta's AI increasingly owns the rest.
Where the Klipio connection fits
Fatigued creative doesn't just get boring — it drags ad quality down. Hides and "why am I seeing this ad" clicks rise as an audience tires of a creative, lowering total value and raising your effective cost even if you never touch your bid.
The fix the auction pushes every advertiser toward is a steady pipeline of fresh angles, not fresh edits of the same one. One honest source of angle ideas: competitors' longest-running ads. Meta shows no spend data for normal ads, but every card shows a "Started running on" date, and an ad still live after 60-90 days has survived enough budget reviews to be a real signal — more on that in how ad longevity spots winners and the Meta Ad Library guide.
Disclosure: Klipio, which I work on, does this automatically — it watches specific competitors' Meta ads, surfaces the angles behind their longest-running winners, and turns those angles into on-brand creative, with paid plans from $79/month. If you'd rather research manually first, our free extension, the Meta Ad Library Downloader, adds a one-click download button to any Ad Library ad, plus a bulk mode that pulls a competitor's whole library into one ZIP with a searchable swipe board — the workflow is in how to download Ad Library videos. Either path hits the same auction lever: better relevance, lower effective cost, without touching your bid.
FAQ
What is the Meta ad auction?
It's the real-time process Meta runs, roughly every time someone opens Facebook or Instagram, to decide which eligible ad wins that person's next impression. Every eligible ad is scored on total value — bid, estimated action rate, and ad quality — and the highest total value wins.
Does a higher bid always win the Facebook ad auction?
No. Total value, not bid size, decides the winner. Estimated action rate plus ad quality (together, ad relevance) can outweigh a bigger bid, so a lower-bid ad with strong relevance regularly beats a higher-bid ad with weak relevance.
What is total value in the Meta ads auction?
Total value is the score each ad gets in the auction, combining your bid, Meta's estimated action rate for that person, and an ad quality score. Meta has never published the exact formula or weights; treat any specific version, including teaching formulas like the one here, as an approximation.
Why did my cost per result go up if I didn't change my bid?
Your total value likely dropped on the other two inputs. Creative fatigue lowers estimated action rate and ad quality through declining engagement and rising negative feedback, pushing effective cost up even with an unchanged bid.
What is estimated action rate?
Meta's prediction of how likely a specific person is to take your optimized action if shown a specific ad, built from past behavior, that ad's own performance history, delivery context, and — since Andromeda — the creative itself. Roughly, estimated click-through rate multiplied by estimated click-to-conversion rate.
Do Advantage+ campaigns skip the auction?
No. Advantage+ automates targeting, placements, and creative variants, but every impression it delivers still goes through the same total-value auction as a manually built campaign. What changes is how much of the setup Meta's AI controls versus you.
The Klipio extension adds a download button to every ad in the Meta Ad Library: one click per ad, or bulk-save a whole search as a ZIP with a searchable swipe file inside. Free, no sign-up.
Get the free extension
