Overview and severity tiers
The Fraud Protection page tells you what share of your paid clicks or sessions in the selected range was invalid, how much budget it cost, and what each verdict asks you to do. It is a premium feature, and before reading the numbers you should know exactly what each severity tier claims.
What the page gives you#
In the sidebar, under the Fraud protection group, a page of the same name sits at /integrity. It shows detections, and on its Blocks and Settings tabs it lets you decide what happens to them.
It has seven tabs, in this order:
| Tab | What it shows |
|---|---|
| Overview | the risk badge, the headline strip, Invalid-rate trend, Why clicks were invalid and the By severity table |
| Clicks | the paid-click log: every click with the verdict it got on landing and the reasons — paid-click log |
| Sources | fraud broken down by channel, source, medium or campaign, the Main fraud source card and the Flagged entities table |
| Networks | wasted spend by network (ASN) and channel, with the Network rings card at the bottom |
| Blocks | the addresses and devices excluded from your campaigns, with expiry and pushes to destinations — blocks and exclusion lists |
| Evidence | the dispute and refund files, the evidence-chain status and the exclusion lists |
| Settings | protection mode, industry preset, thresholds, the Allowlist and the Sensitivity dial — protection settings |
If you know the earlier version of the page: the KPI cards and the two donut charts on Overview are gone, replaced by the headline strip. The Drill-down and Entities tabs merged into Sources, and the Network rings tab is now a card at the bottom of Networks.
Fraud protection is not open on the free plan. If your active website is on the free plan the menu carries a premium badge and the page shows an upgrade card instead of the dashboard. Check the website picker at the top of the page first — another website may be on the right plan. Details in plans and entitlements.
Every number follows the date range#
Every number on the page follows the date picker at the top: from the first to the last day of the range, both included, and the trailing 30 days if you pick nothing. The Range label in the headline strip shows the same text as the date-picker button. Before this update most numbers here were lifetime counts that ignored the picker; if the figures look smaller than they used to, that is why.
Two more rules sit behind the window:
- Each session is judged by its own acquisition — the last touch of that session, not the first touch that brought the visitor in their lifetime. A returning visitor who comes back from your newsletter today is no longer blamed on the ad that first brought them last month.
- Each slice uses its latest verdict up to the last day of the range, provided it is no more than 45 days older than that day. Traffic with no verdict yet is unassessed: it counts in the denominator, but it is never assumed clean and never counted as invalid.
One invalid-rate definition#
There is exactly one definition on the whole page:
invalid rate = flagged units ÷ all units in the range, where flagged means Confirmed plus Potential.
Review sits in the denominator only, because nobody has decided on it yet. The headline, each day of the trend and each row of the Sources breakdown use this same definition, so they add up and a number taken from one place never contradicts another.
The unit is what gets counted. If the range has paid clicks recorded, the unit is the paid click — the thing you pay for. Otherwise the unit is the session. The badge beside the headline strip always says which: "Unit: paid clicks" or "Unit: sessions".
The nightly scoring unit is a traffic slice, not a site#
Every paid click gets a verdict the moment it lands, unless protection is off; that is covered in the paid-click log. Alongside it, the nightly analysis scores traffic slices: the combination of source · medium · channel · campaign · device. The page calls these sources and entities. That grain is what lets you say "this campaign is bad on mobile" rather than only "paid is bad".
At the network level the grain is different — network (ASN) × channel — and that is what the Networks tab does, described in network and ASN fraud.
The header: risk badge, score time and export#
Next to the page title is a risk badge that compresses the range into one word. It is always driven by the share of flagged sessions in the range, even when the headline strip counts paid clicks:
| Badge | When |
|---|---|
| Low | under 8% of sessions |
| Medium | 8% up to 20% |
| High | 20% up to 40% |
| Critical | 40% and above — or 30% and above when at least one entity is confirmed |
If the range has no sessions at all, the badge reads "No data". Traffic that has not been scored yet is not counted as flagged, so on a new website wait for the first nightly run before trusting the badge.
Beside the badge you see the current Sensitivity and "As of … UTC". That time is when the scores on screen were produced — the newest score up to the last day of the range — not when you opened the page. If the range runs to today and that time is several days old, the nightly scoring has not run for this website and the cause needs finding.
The Export button produces a PDF report covering the same selected range.
The headline strip#

The top of the Overview tab is one strip with five numbers. All but the last belong to the selected range:
| Number | What it means |
|---|---|
| Paid clicks, or Sessions | all units in the range |
| Invalid clicks, or Flagged sessions | the Confirmed and Potential units |
| Invalid rate | the second divided by the first, on the one definition above |
| Est. wasted spend | the part of the range's imported ad cost that lands on flagged networks — wasted spend |
| Active blocks | how many blocks — addresses, ranges and devices — are active right now; this is the current state, not a count for the range |
Beside them sit the unit badge, the Range label and, sometimes, a green Live badge. Live means a paid click was recorded in the last 10 minutes with a verdict decided in real time — evidence that the real-time check is actually running on your site.
When ad cost is imported for the range, a line appears under the strip: "Never more than the … of ad cost imported for this range." Wasted spend never exceeds the cost you imported.
The trend and the reasons#

Two charts sit side by side under the strip.
Invalid-rate trend runs day by day across the same range: the bars are the invalid units each day (invalid clicks or flagged sessions, depending on the unit) and the line is that day's invalid rate, on the same definition as the headline. Until the range holds at least two days of data you get the message "Trend appears once the website has been scored on 2+ days." instead of a chart.
What to watch for here is a single-day jump, not the absolute level. A high, flat level usually just means that is your channel mix; a sudden jump means something changed that day — a new campaign, a new placement, or a click farm that has just found you.
Why clicks were invalid is built from the paid-click log and shows the ten most frequent reasons among the range's Confirmed and Potential clicks, such as "Repeat clicks (10 min)" or "Datacenter IP". A click can carry several reasons and is counted under each, so the bars need not add up to the number of invalid clicks. Until paid clicks are recorded in the range, the message "Appears once paid clicks are recorded in this range." takes the chart's place.
The four severity tiers and what they ask of you#
The By severity table at the bottom of the Overview tab shows the range's sources and sessions in each tier, and prints the definition itself underneath: "Confirmed = high ensemble agreement + network corroboration. Potential = elevated score. Review = insufficient evidence to confirm automatically."
| Tier | What AdPix is claiming | What to do |
|---|---|---|
| Confirmed | several detectors agree and a structural signal corroborates them | if it is a paid channel, exclude the network or placement and take the evidence for a dispute |
| Potential | there is real evidence, not enough for certainty | investigate; if it is one campaign or placement, cap its budget before making a bigger decision |
| Review | something showed up and AdPix deliberately did not rule on it | look at it. If it is your own traffic, add it to the Allowlist or file a false-positive report |
| Clean | no evidence | nothing to do |
Review is not a weak tier — it is the tier where the product admits it does not know. Most of the guardrails in the next section end up exactly there.
What is never auto-flagged#
These are not tunable thresholds. They are hard rules applied on every scoring run, and no setting turns them off.
- An entity that converts. The bar is 2%, and conversion evidence is not only
purchase: the website's registered key events, session revenue, and the conversion rate rolled up to the source itself all count. That last one matters — the slice grain splits a converting source by device and campaign and can push each piece below the bar; the source-level roll-up prevents it. - That same entity when it also carries a hard structural signal. Key events come from the browser and a bot can fake them, so conversion does not cancel a structural signal. The outcome is review — neither confirmed nor clean.
- A low-volume entity. Below three sessions nothing goes past review on any path. A fingerprint ring needs at least five sessions to confirm and at least three to reach potential. The anomaly model on its own never flags anything below ten sessions.
- An entity that is merely an outlier. The anomaly score is a relative rank and always marks the top decile of any population, including an all-human one. A model-driven verdict is issued only when at least one behavioural signal corroborates it. "You are in the top decile" flags nothing on its own.
- An entity with real engagement. A session with at least ten seconds of dwell, or two page views, or a key event counts as engaged under the standard definition. If at least half of an entity's sessions look like that, model opinion never auto-flags it — review is the ceiling.
- Direct traffic. Direct is the default bucket for unattributed traffic. It is never confirmed or potential on model opinion; only a hard structural signal can flag it.
The main fraud source names a source, not a channel#
On the Sources tab, Break down by switches the table between channel, source, medium and campaign, and ranks the range's fraud by flagged sessions. Every row uses the same invalid-rate definition.
The Main fraud source card at the top of that tab always names the worst source, whichever dimension the table is set to. The reason is a real mistake that happened once: when the headline was built from the channel roll-up, one bad source had its rate spread across the whole channel and the headline accused an entire channel — Organic Search — when the problem was a single source inside it. Acting on the channel is almost always wrong; acting on the source is right.
Clicking any row in the table, or View detail on that card, opens the detail drawer with the invalid-versus-valid split and the flagged slices under that value — the path described in why this was flagged.
The sensitivity dial#
The Sensitivity dial card now sits at the bottom of the Settings tab, below the Allowlist, with three modes:
| Mode | The page's own hint |
|---|---|
| Conservative | "Flags only very-high-confidence cases — low false-positive rate." |
| Balanced | "Default. Balanced recall and precision for most websites." |
| Aggressive | "Flags potential and review-tier entities; higher recall / FP rate." |
Read the Balanced hint carefully: it says "Default", but a website that has never saved a setting is scored Conservative — the safest of the three — and the control opens on that mode until you save something else. The dial tunes only the nightly scoring of slices: the score threshold and how many detectors must agree. It takes effect from the next run, does not retrain the model, and switches off none of the guardrails above. The real-time verdict on clicks is tuned by the protection mode, industry preset and thresholds on the same tab — see protection settings. When to move the dial is covered in sensitivity and false positives.
Getting the result out, and correcting a wrong verdict#
The PDF report from the Export button gives you the range's fraud posture in a form you can attach to an email. For a dispute with an ad network you need the files on the Evidence tab instead — the click quality report for Google and the sealed refund evidence — covered in evidence, exclusion lists and refund claims.
If you believe a verdict is wrong, you have two routes:
- A click's verdict — in the Click evidence drawer on the Clicks tab, use "Not fraud: allowlist this IP" or "Not fraud: allowlist this device". That address or device is Clean from then on.
- A traffic slice's verdict — on the Sources tab, in the Flagged entities table, use Report FP at the end of that row. It does not change the score immediately; it records your report for the platform team, and if accepted the affected sessions are re-evaluated on the next scoring run.
Both routes are covered in sensitivity and false positives.
Frequently asked questions#
Why doesn't the invalid rate match the number of flagged sources?
Because the invalid rate counts units, not sources: paid clicks or sessions, whichever the unit badge names. The numerator is Confirmed plus Potential only and the denominator is all traffic in the range. Ten small flagged sources out of a hundred can be well under one percent of your real traffic. Review sits in the denominator only and is never counted as invalid.
Why does the headline strip count paid clicks one time and sessions another?
If paid clicks were recorded in the selected range, the unit is the paid click, because that is what you pay for. If not, the unit is the session. The badge beside the strip always says which one you are looking at, and the rate is defined the same way in both cases.
Why is a channel that genuinely drives sales sitting in review?
Because it also carries a hard structural signal — a fingerprint seen across several businesses, for example. Conversion never causes an automatic flag, but it does not erase a structural signal either, because key events are emitted by the browser and a bot can fake them. Review is the deliberate outcome, so a human decides.
Why does the main fraud source differ from the worst channel in the table?
The Main fraud source card on the Sources tab always names the worst source, even when you have switched the table to channel or campaign. A channel roll-up dilutes one bad source's rate across the whole channel and blames the channel instead.
Thanks — your feedback helps us improve the docs.