Neither, until you have segmented it. `All conversions` is a sum over every conversion action attached to the account, not a count of people or events. On a mature account it routinely over-counts by two to three times, and the inflation is invisible in the column itself.

On this page

• What does the All Conversions column actually count?
• What does a strangled catcher campaign look like in real numbers?
• How do I segment conversions by action name?
• What breaks when the conversions column over-counts?
• Which conversions column should I judge performance on?

What does the All Conversions column actually count?

Duplicate actions, legacy actions from an old setup, and test actions nobody removed are all normal. Every one of them adds to the total.

What does a strangled catcher campaign look like in real numbers?

A campaign reported 3 conversions on a single day. Segmented by conversion action name, it was one person registering, counted three times:

Real record: 1 registration on 715 clicks — 0.14%, not 3. That single segmentation flipped the verdict on the campaign.

How do I segment conversions by action name?

`` SELECT segments.conversion_action_name, metrics.all_conversions, metrics.all_conversions_value FROM campaign WHERE campaign.id = AND segments.date BETWEEN ... ``

What breaks when the conversions column over-counts?

1. The value column built this way is nonsense. $104 recorded for one registration genuinely worth $4. Any return calculation on top of that is fiction.

2. A live (Test) conversion action in production is a defect. Check for one on any account whose numbers look implausibly healthy. It is a five-minute check and it invalidates months of reporting when you find one.

3. Any rate threshold built on an inflated count is wrong. And it is usually wrong in the safe direction — but the assumption underneath it often isn't. One case had a "must fire on 1.3% of clicks" bar written at an assumed $0.60 cost per click. The actual cost per click was $1.61, which made the real bar 3.5%. Re-derive any threshold whenever cost per click moves.

Which conversions column should I judge performance on?

Report the single conversion action that represents the outcome you are paid for, by conversion time. Not the aggregate, either flavour.

Conversions is the subset marked primary and is closer to useful, but it is still a sum over whatever is flagged — and the flag is not always what drives the bidding. A conversion action set to secondary still bids if it sits inside a custom goal, and account-level primary flags can be overridden by a campaign-level custom goal entirely.

Verify the goal configuration and the live conversions column. Do not trust the primary flag. I have watched two separate analyses reach the wrong conclusion about what an account was bidding on by reading account-level primary flags on an account whose real bid signal was a campaign-level custom goal.

Related questions

Why doesn't my Google Ads conversion count match the number of jobs in my CRM?

Expect a structural gap of roughly 30%, not zero. On a 14-location home-services network, the share of bookings matchable back to a click sat in a 0.56–0.83 band across settled windows — and a full forensic investigation found no ad-platform setting that explained the variance. Read the full answer.

Why is my exact match keyword showing for searches that have nothing to do with it?

Exact match has meant "close variant" for years, and close variants now include same- meaning rephrasings, implied words and reordered queries. What you are seeing is usually the system working as designed — the question is whether the variants are converting. Read the full answer.

Should I count a phone call as a conversion at 60 seconds, or is call length a bad way to judge a lead?

Call length is a proxy, and it is the weakest one available if you can get booking data instead. Use it as a starting bridge, not as the destination. Read the full answer.

Written by Usama Khan, who runs Google Ads for e-commerce brands and home-services companies at AdventurAds. Figures are from accounts under active management and are anonymised. Last reviewed 2026-08-26.