How to Read Your Conversion Quality Report
Your audit reveals how much of your reported conversion data is real versus modeled, where signal pollution originates, and which tracking actions need fixing. This guide explains every metric and the exact actions to take. The PDF skeleton on the right highlights where you are.
Jump to section
The Four Headline KPIs
Page 1 — the gap between reported and real conversions
These four KPIs quantify the gap between what Google Ads reports and what is objectively verifiable. Click conversions are directly observed. Everything else — view-through, cross-device, modeled — is an estimate the algorithm treats as fact.
Click Conversions
Ground TruthThe total number of conversions where the user clicked your ad before converting. This is the hardest, most reliable signal. If this number is small relative to all_conversions, most of your reported performance is based on estimates.
Modeled %
Lower is betterThe percentage of all_conversions that are NOT click-attributed. Calculated as (all_conversions − click_conversions) / all_conversions. Above 30% means almost a third of the data the algorithm learns from is estimated, not observed.
If above 40%, audit your conversion actions immediately. The algorithm is training on noise.
Avg Attribution Drift
The average percentage difference between Ads-reported and GA4-confirmed conversions across all campaigns. Pulled from the signal_health_intelligence layer using the latest calculation date. Above 20% means the two systems fundamentally disagree on what happened.
Signal Health Score
0–100A composite score averaging signal_health_score across all campaigns. Combines attribution drift, conversion density, and signal status into a single number. Below 60 means the bidding algorithm is making decisions on data it should not trust.
Signal Health Distribution
Page 1 — how many campaigns have clean vs corrupted data
This donut groups all campaigns by signal status: HEALTHY, WARNING_DRIFT, CRITICAL_BLIND, and LOW_VOLUME. The healthy slice is all the algorithm can trust — everything outside it is noise distorting bid decisions across your account.
HEALTHY campaigns
TrustworthyAds and GA4 agree within 15%. Smart Bidding has clean signal to optimize. No tracking action needed for these campaigns.
CRITICAL_BLIND campaigns
Fix FirstDrift exceeds 40%. The algorithm is bidding on phantom data. Every dollar spent on these campaigns compounds the damage because the algorithm learns wrong patterns that affect account-wide bid adjustments.
Fix conversion tracking on CRITICAL campaigns before any other optimization work. Everything downstream is corrupted.
Attribution Drift Scatter
Page 1 — the visual proof of where Ads and GA4 disagree
Each dot is a campaign plotted with GA4 conversions on the X-axis and Ads conversions on the Y-axis. Points on the diagonal line are perfectly aligned. Points above the line report more in Ads than GA4 confirms — the further from the line, the less trustworthy the campaign's data.
Points above the diagonal
Over-ReportingGoogle Ads claims more conversions than GA4 can confirm. This campaign is inflating the algorithm's confidence in audiences and keywords that may not actually convert.
Audit conversion tags for this campaign. Check for duplicate firing, mismatched attribution windows, or view-through conversions being counted alongside clicks.
Points below the diagonal
Under-CountingGA4 records more conversions than Ads. This campaign may have attribution model differences or GA4 is capturing organic conversions that Ads doesn't claim. Less dangerous, but still a signal accuracy problem.
Conversion Type Mix by Campaign
Page 1 — click vs view-through vs modeled per campaign
Each stacked bar shows one campaign's conversion volume broken into three layers: click-attributed (blue), view-through (amber), and modeled (red). Taller non-blue segments mean more of that campaign's reported conversions are algorithmically estimated rather than directly observed.
Campaign with > 50% non-click
Inflated SignalMore than half this campaign's conversions come from view-through or modeled attribution. The algorithm is optimizing this campaign based on impression exposure, not verified purchase behavior.
Consider excluding view-through conversions from this campaign's bidding strategy. Review whether the campaign type (Display, Video) naturally generates more impressions than clicks.
Conversion Action Mix
Page 2 — which tracking actions feed the algorithm
Each row represents a conversion action configured in Google Ads — purchases, leads, page views, etc. The category and source columns reveal whether the action is measuring genuine business outcomes or proxy events that pollute the signal.
Multiple actions with same category
Duplication RiskTwo or more conversion actions tracking the same event type (e.g., two separate 'Purchase' actions) can double-count conversions. The algorithm treats each firing as a separate conversion signal.
Audit Google Ads conversion settings. Keep one primary action per business event. Move duplicates to 'Observe only' so they report but don't feed bidding.
Micro-conversions as primary
Signal NoiseIf page views or newsletter signups are set as primary conversion actions alongside purchases, the algorithm optimizes for the easiest signal — driving cheap micro-conversions instead of revenue.
Set only revenue-generating actions (purchase, lead form submit) as primary. Move all micro-conversions to secondary/observe-only.
View-Through Inflation Risk
Page 2 — campaigns where impression exposure is counted as performance
This bar chart ranks campaigns by view-through percentage — the share of conversions from users who saw an ad but never clicked. Only campaigns with VT% above 30% appear here. These campaigns are training the algorithm on impression exposure rather than purchase intent.
VT% > 50%
Impression OptimizedMore than half of this campaign's conversions come from users who never clicked. The algorithm is essentially optimizing for ad views, not actions. Reported ROAS and CPA for this campaign are significantly more optimistic than reality.
Exclude view-through conversions from bidding for this campaign. Evaluate performance using click-attributed conversions only.
Cross-Device Attribution Gap
Page 2 — modeled conversions across devices
This table shows campaigns where users clicked on one device and converted on another. Google models these using signed-in user data. Cross-device conversions are legitimate but less precise than single-device click attribution.
XD% > 30%
Heavy ModelingOver 30% of this campaign's conversions involve a device switch. While Google's cross-device model is generally reliable, high XD% combined with high drift means the modeling may be over-crediting mobile interactions for desktop purchases.
Cross-reference with the attribution drift table on Page 1. If drift is also high, the cross-device modeling is likely inflating performance.
Conversion Lag Distribution
Page 2 — how long after a click does a conversion happen?
This bar chart shows how conversions distribute across time windows after the initial click — same day, 1 day, 2–3 days, up to 30+ days. Significant volume arriving after 7+ days means your attribution window may be too short, and some flagged 'waste' terms may actually convert later.
Long tail lag (7+ days significant)
Window RiskIf meaningful conversion volume arrives after 7 days, campaigns evaluated on a 7-day window are systematically under-counted. This makes waste calculations pessimistic and good keywords look worse than they are.
Extend your attribution window to 30 days for high-consideration products. Re-evaluate any negative keyword decisions made using short-window data.
Same-day concentration (> 80%)
Clean SignalMost conversions happen the same day as the click. Attribution is straightforward, waste calculations are reliable, and short-window analysis is trustworthy.
Device Conversion Quality
Page 3 — which devices produce reliable conversion data
This table breaks conversion quality by device — click conversions, spend, CVR, and CPA for mobile, desktop, and tablet. Devices with low CVR but high spend are consuming budget for low-quality clicks that rarely convert.
Mobile: high spend, low CVR
Common PatternMobile often generates high click volume at lower CPCs but converts at a fraction of desktop rates. The algorithm may over-invest in mobile clicks because they're cheap, not because they convert.
Apply device bid adjustments. If mobile CVR is less than half of desktop CVR, a -30% to -50% mobile modifier is typically warranted.
Network Conversion Quality
Page 3 — Search vs Display vs Video signal reliability
This table compares click conversions vs all conversions by network type (Search, Display, Shopping, Video). The VT% column reveals which networks are inflated by view-through attribution — Display and Video will almost always have higher VT% than Search.
Display VT% > 60%
Impression-DrivenDisplay network conversions are predominantly view-through — users saw a banner ad and later converted through another channel. The algorithm credits Display, but the actual conversion driver is likely Search or direct traffic.
Review whether Display campaigns use view-through in their bidding strategy. Consider switching Display to awareness-only metrics (reach, frequency) rather than conversion optimization.
Top Converting Search Terms
Page 3 — your proven winners
These are the search terms with the most click-attributed conversions — the hardest, most reliable conversion signal. Protect these terms with dedicated exact-match keywords and sufficient budget. They are the foundation your algorithm should optimize around.
High-volume, low-CPA terms
Protect TheseThese terms consistently convert at low cost. They should have dedicated exact-match keywords, adequate daily budget, and should never be outbid by broad-match expansion.
Create exact-match keywords for your top 10 converting terms if they don't already exist. Set higher bids than broad-match equivalents to ensure they win the auction.
Conversion Actions Audit (Risk Level)
Page 4 — which tracking actions need fixing
Each conversion action is risk-scored based on its modeled percentage. Actions with HIGH RISK (>50% modeled) need re-implementation. MODERATE actions (30–50%) need a deduplication audit. Actions below 30% are healthy.
HIGH RISK — Review tracking
> 50% modeledThis conversion action produces more estimated conversions than directly observed ones. The tag is either misfiring, duplicated across containers, or counting the wrong event.
Re-implement using enhanced conversions with first-party data. Verify in Tag Assistant that the tag fires exactly once per genuine conversion event.
HEALTHY actions
< 30% modeledThis action's data is predominantly click-attributed. No tracking fix needed. Use these as the benchmark for what clean data looks like in your account.
Campaign Signal Recommendations
Page 4 — ranked fix list by campaign
Each campaign with a non-HEALTHY signal status gets a specific recommendation based on its failure mode. CRITICAL_BLIND campaigns need urgent tag fixes. WARNING_DRIFT campaigns need alignment audits. LOW_VOLUME campaigns need consolidation or broader targeting to build enough signal.
URGENT: Fix conversion tracking
CRITICAL_BLINDThis campaign has drift exceeding 40%. The algorithm is confidently bidding on data that is fundamentally wrong. This is the highest-priority fix — everything else you optimize on this campaign is built on sand.
Stop bid and budget changes on this campaign. Audit tags, then re-evaluate performance with clean data before resuming optimization.
Increase volume or consolidate
LOW_VOLUMEThis campaign doesn't have enough conversions for the algorithm to learn. The signal is too sparse for Smart Bidding to make reliable decisions.
Either broaden targeting to increase conversion volume, or merge this campaign into a higher-volume campaign so the algorithm has more data to learn from.
Deduplication Risk
Page 4 — is the same conversion counted twice?
This table compares click conversions against all_conversions per action. The inflation ratio (all_conv / click_conv) tells you how many times each action fires relative to click attribution. A ratio above 2.0 strongly suggests duplicate tag firing.
Inflation ratio > 2.0
Likely DuplicateThis action records 2x+ more conversions than click attribution alone. The most common cause is the same conversion tag firing from both Google Tag Manager and a hard-coded page snippet — or the same event being counted by both a GA4 import and a direct Ads tag.
Open Google Tag Manager and search for all tags firing on this conversion event. Remove duplicates. Verify with Tag Assistant debug mode.
GA4 Funnel vs Ads Conversions
Page 4 — where exactly does the gap open?
This bar chart shows the ecommerce funnel — Sessions → Product Views → Add to Cart → Checkout → GA4 Purchases → Ads Conversions — side by side. The steepest drop between adjacent steps is where conversion signal degrades most. If Ads Conversions exceed GA4 Purchases, the algorithm is crediting conversions that GA4 cannot verify.
Ads Conversions > GA4 Purchases
Phantom GapGoogle Ads reports more conversions than your own analytics platform confirms as purchases. This gap is the phantom signal the algorithm uses to over-bid — it believes there are more conversions than actually occurred.
Align Ads and GA4 attribution models. Check that both use the same conversion window (7-day click vs 30-day click). Remove any Ads conversion actions that duplicate GA4 events.
Cart → Checkout steep drop
UX ProblemThis is not a tracking problem — it's a checkout friction problem. Users add to cart but abandon before checkout. While not directly a signal quality issue, it reduces the conversion volume the algorithm can learn from.
Conversion Value Quality
Page 4 — are high-value actions also highly modeled?
This table cross-references conversion value per action against modeled percentage. The worst-case scenario: your highest-value conversion actions (purchases with large order values) are also the most heavily modeled. This means the algorithm's revenue-based bid strategies (tROAS) are optimizing on the least reliable data.
RISK — High value but mostly modeled
tROAS CorruptedThis action generates significant revenue per conversion, but over 40% of its conversions are modeled. Target ROAS bidding uses this value data to set bids — if the values are attached to phantom conversions, bids are systematically inflated.
Switch this campaign to tCPA instead of tROAS until the action's modeled percentage drops below 30%. tCPA is less sensitive to value accuracy than tROAS.
Q: How much of my reported conversion data can I actually trust?