To detect bot traffic, compare what your visitors do with what real people do: look for sessions with no engagement, traffic spikes from one country, network or referrer, data-centre IP addresses, identical device profiles and clicks or conversions that happen impossibly fast. Check three places: your analytics, your server or CDN logs, and your tracker's click data. Bots that fool one rarely fool all three.
This guide shows the patterns to look for in each, how to filter bot clicks out of your reports and tracker, and how to build a refund claim when the fake traffic came from an ad network you paid.
Signs of bot traffic: the quick checklist
None of these signals proves bot traffic on its own. Two or three together, concentrated in one source, almost always do.
- Engagement collapses for one source: near-zero engaged sessions, one page per session, no scrolling.
- A sudden spike in sessions without a matching rise in clicks to offers, sign-ups or sales.
- Odd geography: traffic from countries you do not target, or a GEO campaign suddenly showing other countries.
- Uniform technology: the same browser version, screen resolution or operating system across thousands of visits.
- Data-centre networks: IP addresses owned by hosting and cloud providers rather than consumer broadband or mobile networks.
- Machine timing: visits at evenly spaced intervals, round-the-clock flat volume, or click-to-conversion times of a second or two.
- Junk referrers: referral domains you have never heard of, often with no real page linking to you.
Finding bot traffic in analytics
GA4 automatically excludes traffic from known bots and spiders. Google's help page on known bot-traffic exclusion says you cannot switch this off or see how much was removed. That filter catches declared crawlers, not bots designed to look human, which are the ones that cost you money.
To find the rest, build an exploration with these dimensions and metrics:
- Rows: session source/medium, then country, then device category.
- Values: sessions, engagement rate, average engagement time and your key event (for example affiliate clicks).
- Sort by sessions and look for rows where volume is high but engagement rate and key events are close to zero.
- Add browser, operating system or screen resolution as a secondary breakdown for any suspicious row. A single value accounting for nearly all the traffic is a strong bot signal.
Privacy-friendly tools such as Matomo or Plausible show the same pattern in their source and country reports. Whatever you use, compare a suspicious source against a trusted one (such as organic search) rather than against an abstract benchmark.
Server and CDN log patterns that reveal bots
Logs see every request, including bots that never run your analytics script. You can read raw access logs, use your host's log viewer or use your CDN's security analytics. Look for:
| Pattern | What it suggests |
|---|---|
| Thousands of requests from one IP or a narrow range | Scraper or crude bot |
| Requests for HTML but never for CSS, images or JavaScript | Headless client that does not render pages |
| Hits on the landing page with the click ID, then nothing | Click inflation from a traffic source |
| User agents claiming to be search engines from non-search IPs | Spoofed crawler |
| Regular intervals between requests, to the second | Scripted traffic |
To check whether an IP belongs to a hosting provider, run a reverse DNS or IP lookup; most commercial IP databases flag data-centre ranges. Genuine search engine crawlers can be verified by reverse DNS, which spoofed bots fail.
Fake traffic from ad networks: what to check
Paid traffic is where bots hurt most, because you pay for every impression or click. Popunder, push and native traffic all carry risk, and quality varies widely between networks and between placements within the same network. For each campaign, break your tracker data down by:
- Placement, zone or site ID, the single most useful breakdown. Bots tend to concentrate on specific placements.
- ISP or connection type, to spot hosting providers posing as mobile users.
- Time of day in the visitor's local time zone.
- Landing page engagement, such as the proportion of visitors who click through from your lander to the offer.
Hypothetical example: a popunder campaign shows a 9% lander click-through rate overall. Broken down by zone, 40 zones average 14% and 6 zones show 0.2% on 30% of the spend. Those six zones are almost certainly bots or forced, unseen impressions. Blacklist them, and keep the data for a refund request. See our guide to whitelisting and blacklisting placements.
How to filter bot clicks in your tracker and site
Filtering works at three levels. Use all three if you buy traffic.
In your tracker
- Switch on built-in bot filtering and IP or user-agent blocklists if your tracker offers them.
- Create rules that send traffic from data-centre ISPs, empty user agents or blocked countries to a separate path, so you keep the evidence without counting it as real clicks.
- Use a short JavaScript check or a lander click as the real "visit" event, rather than the raw redirect click.
On your site or CDN
- Rate-limit aggressive IPs and challenge suspicious requests with your CDN or firewall rules.
- Protect forms with server-side validation and a challenge for unusual submissions.
- Exclude internal and known testing traffic from analytics with filters.
At the traffic source
- Use the network's own fraud and quality settings, which often include filtering of proxies and data-centre traffic.
- Work from a whitelist of proven placements once a campaign is past the testing stage.
Do not over-filter. Mobile carriers, corporate VPNs and privacy relays can look unusual. Test any new block rule on a copy of your data first and check that genuine conversions are not caught.
How to claim refunds from traffic sources
Most reputable ad networks have a process for disputing invalid traffic, but the rules, time limits and evidence they accept vary. Google Ads, for example, says it automatically filters invalid clicks and credits advertisers where it detects them after billing. Smaller networks usually require you to open a ticket. Check your network's terms before you buy, so you know the deadline.
A claim that gets taken seriously includes:
- Campaign, placement or zone IDs and the exact date range.
- Your tracker export showing clicks, lander engagement and conversions for those placements, alongside a comparison with clean placements.
- Technical evidence: IP ranges, ISPs, user agents and timing patterns.
- The amount you are claiming, calculated from the network's own billing figures.
Keep the tone factual. Networks deal with many speculative claims; precise data from a tracker gets results far more often than a complaint about low conversion rates. For the wider picture of fraud schemes, read types of affiliate fraud.
When "bots" are really tracking problems
Before you blame bots, rule out measurement faults. A broken postback, a redirect that strips parameters or a consent banner blocking your analytics can make good traffic look worthless. If clicks look human in your logs but conversions vanish, work through why conversions are not tracking and how to resolve tracking discrepancies first.
Next steps
Pick your biggest paid source and break its last 30 days down by placement and ISP. That one report usually shows whether you have a bot problem. When you are ready to test new sources, compare networks in our traffic sources directory.