Google Ads
How to analyze a Google Ads campaign
The reading order that finds a Google Ads campaign's problem: from conversion to search term, not from impressions down, with what to do at each stop.
Also available in: Português · Español
Analyzing a Google Ads campaign goes wrong almost always for the same reason: people start at the top of the table, look at impressions and clicks, and reach what matters after forming an opinion. The right order is the reverse.
Start at the end: cost per conversion against your target
The first question is not about the campaign, it is about the business: how much can you pay per conversion?
That number comes from your margin, not from Google. If each sale leaves US$ 150 after variable costs, cost per acquisition has to sit well below that. The calculation is in contribution margin.
With the number in hand, analysis begins:
Cost per conversion inside target. There is no problem to solve. However bad CTR and quality score look, the campaign is paying. The question becomes different: can it scale?
Cost per conversion above target. Now there is something to investigate, and the next sections are the order.
That inversion saves more time than any analysis technique. Most "optimizations" are performed on campaigns that were already inside target.
Confirm the conversion is real
Before any conclusion, check what is being counted. Three common errors make everything else meaningless:
- A conversion action counting everything. Phone clicks, page views and form submissions in the same column makes cost per conversion look excellent.
- A duplicated conversion. The same submission counted twice, from a repeated tag.
- A conversion that is not a sale. Optimizing for "viewed the contact page" finds people who view it, not people who hire.
If the number is wrong here, everything after it is wrong with confidence.
The report that returns the most money
Search terms, sorted by cost descending, filtered to zero conversions over the last 30 days.
In an account that has never been audited, that report typically shows 10% to 30% of spend going to searches unrelated to the business. The fix is adding negatives, and the return shows the same week.
It is the first place to look when cost per conversion is high, before touching bids, audiences or ads. The routine is in Google Ads search terms.
The reading order, in six stops
| Stop | What to look at | What it indicates |
|---|---|---|
| 1 | Cost per conversion against target | Whether there is a problem |
| 2 | Configured conversions | Whether the number is trustworthy |
| 3 | Search terms with no conversions | Direct waste |
| 4 | Impression share and what lost it | Whether budget or quality is missing |
| 5 | CTR per ad against the group average | Whether the piece interests anyone |
| 6 | Landing page conversion rate | Whether the problem is outside Google |
Each stop has a different fix, and they are not interchangeable. Raising the bid when the problem is the page does not solve anything — it just makes the same problem more expensive.
Impression share: the stop nobody makes
The impression share column, and the ones for share lost to budget and to rank, answer the question that decides budget: is there market left?
Lost to budget. There were searches, you were eligible, the money ran out. The cheapest problem to fix in paid media.
Lost to rank. The budget was there and the ad was not placed well enough. Different work: ad, page, relevance. The detail is in Google Ads quality score.
Telling the two apart completely changes what to do next week. The broader view is in share of voice.
The date range that informs
Closed weeks, same weekday to same weekday, at least two to compare.
Why not a day: daily variation is enormous and means nothing. Why not a rolling month: it mixes good and bad weeks and hides the trend.
And keep the range fixed between analyses. Changing period moves every number with nothing having changed in the campaign, and it is the source of half the wrong conclusions.
When the problem is outside Google
Two situations where optimizing the campaign solves nothing:
The page does not convert. Clicks arrive, conversions do not. Doubling the page's conversion rate has the same effect as halving cost per click, and is usually easier. The diagnosis is in landing page not converting.
Sales does not follow up. A lead waiting two days for a call becomes a "bad lead" in the report. Nothing in Google Ads shows that.
The split between what marketing delivers and what sales accepts is in MQL vs SQL.
Performance Max is a different analysis
An automated campaign exposes less detail. You get spend, conversions and cost per conversion; the breakdown by network, term and placement is limited by the platform itself.
Practical consequence: stops 3, 4 and 5 in the table are partly unavailable. The analysis reduces to deciding budget and comparing against the Search campaigns in the same account.
That is why it is worth keeping Performance Max on its own row rather than summing everything. The comparison is in PMax vs Search.
The mistake of changing five things
On finding the problem, the temptation is fixing everything at once: add negatives, change bidding, rewrite the ad and change the page the same day.
Two weeks later cost per conversion improved and nobody knows why — which means the lesson was not learned and the next case will cost the same time.
One change at a time, with the date written down. Bid strategy in particular needs a relearning period before being judged.
What to write down, and why it matters more than the analysis
An analysis that leaves no record has to be redone from scratch next month, and the account never accumulates knowledge.
Four lines per review, in any document:
The date and the range. Without it, nobody can reconstruct what you were looking at.
What you found. One sentence. "Search terms burned US$ 340 on recruitment queries."
What you changed. One line per change, not a summary. "Added 12 negatives." "Switched campaign 3 from maximize clicks to maximize conversions."
What you expect. A number and a date. "Cost per conversion should drop below US$ 60 within two weeks."
The fourth line is the one people skip and the one that turns the log into a learning tool. Without a stated expectation, every outcome gets rationalized after the fact — a drop becomes proof the change worked, a rise becomes seasonality.
Six months of that log is worth more than any dashboard, because it is the only document that connects decisions to consequences. The reporting side of it is in monthly marketing report.
The twenty-minute analysis
Cost per conversion against target. Conversions configured correctly. Search terms with no conversions, add negatives. Share lost to budget, raise where it converts. One change at a time, with a date.
That routine resolves most cases, and it fits in a Tuesday morning. The reporting structure is on the paid media reporting page.
Frequently asked questions
Where do I start the analysis?
With cost per conversion against your margin target, not with impressions. If it is inside target, there is no problem to solve, however bad the other columns look.
How many days should I look at?
Closed weeks, same weekday to same weekday, and at least two to compare. A single day is noise, and a rolling month mixes good and bad weeks without showing the trend.
Which report returns the most money?
Search terms sorted by cost, filtered to what spent without converting. It finds waste fastest in almost any account that has never been audited.
Do I analyze an automated campaign the same way?
You cannot. Performance Max exposes less detail per channel and per term. Its analysis stays at spend, conversions and cost per conversion, with far less diagnosis available.