Bot traffic in analytics is the share of sessions that never had a person behind them: crawlers, uptime monitors, scrapers, AI agents fetching pages for a model, and headless browsers running scripts. They request pages, they fire your tracking tag, and they land in the same tables as your customers. The damage is not the extra pageviews. It is the denominator — conversion, activation, retention and bounce are all divided by a number that includes visitors who were never going to convert.
Most teams treat this as a filtering chore. It is closer to a data-quality policy: a filter that hides rows leaves the raw events wrong, and every tool downstream — funnels, experiments, attribution, revenue models — inherits the error. This guide covers where automated sessions enter, how to spot them in charts you already have, and how to exclude them at three layers.
Where automated sessions enter your data
Crawlers, scrapers and monitors
Search engine crawlers are the polite version: they identify themselves, and most analytics platforms drop them against a known-bot list before a report is built. That default covers the famous names, not the long tail — SEO audit tools, link checkers, price scrapers and monitoring services pinging your pricing page on a schedule.
AI agents and model crawlers
Pages are now fetched for training and for retrieval. Some agents announce themselves; many render JavaScript like a browser because the content only exists after hydration. If your tracking lives in client-side code, a JavaScript-rendering agent looks like a visitor with a fast, single-page, zero-interaction session.
Your own automation
Test suites, synthetic monitoring, staging jobs pointed at production and a teammate’s script hitting your signup form. These are the easiest to remove, because you can label them at the source.
Read your own charts for bot-shaped patterns
Before adding tooling, look for fingerprints you already have. A single pageview with a duration near zero on a deep URL nobody links to. Sessions from datacenter networks and countries outside your market. Hundreds of hits on one endpoint at identical intervals. Traffic in the source report that never appears at step two of a funnel. Mobile sessions from device models you do not support. Event timestamps that arrive in perfect rhythm.
One of these is noise. Three together are a signature. Write them down as a short list; it becomes the acceptance test for whatever exclusion you build.
Exclude at three layers, not one
Layer 1: classify the session in the browser
Detection has to run before the event is sent, not after it is stored. A browser agent can check automation flags, headless indicators and the mismatch between a declared user agent and the device actually rendering the page, then return a verdict with reasons attached. That verdict travels with the event, so each session carries its own label instead of being reconstructed later. Our bot detection page covers what those signals look like.
Layer 2: catch what the browser cannot see
Browser-side detection is blind to network signals — VPN, proxy, Tor, datacenter ranges — and to reputation shared across sites. That work belongs on your server or at the edge, where you can decide per request and keep the reasoning. AI-agent traffic deserves its own category rather than being folded into fraud: the policy for a model reading your docs is not the policy for a script probing checkout. See AI agent detection and device fingerprinting.
Layer 3: segment in the report, keep the raw events
Do not delete bot sessions. Flag them and exclude the flag from default views. You will want them back when a “bot” turns out to be a customer on a corporate VPN, when a number has to be explained to an auditor, or when a detection rule needs tuning. Deletion is irreversible; a segment is a switch.
Decide policy, not just filtering
“Is this a bot” and “what should happen” are different questions. A monitoring service pinging your status endpoint should be allowed and labelled. A scraper walking hundreds of product pages should be rate-limited. A script filling signup forms should be stepped up or blocked and logged with a reason code you can read later. Write the mapping explicitly — signal, decision, log — and apply it across signup, login, trial and checkout, so the same automated client does not get one answer at one endpoint and a different answer at the next.
Two habits keep it durable. Mirror decisions before enforcing: log what you would have blocked, watch it for a couple of weeks, fix the false positives, then enforce. And keep bot sessions out of quota and billing logic, so an attack cannot cost you money while it is being filtered.
The fastest audit is one funnel with a real business meaning — signup completion, trial start, checkout. Export the sessions, sort by duration, read the ten shortest, and check the network and device fields on those rows. Compare the funnel rate with bots flagged and unflagged. If the gap would change the decision you were about to make, you have found the problem, and the fix is a policy, not a prettier dashboard.
FAQ
Does Google Analytics already remove bots?
It removes the ones on its known-bot list, which is a starting point rather than a policy. JavaScript-rendering agents, scrapers, monitoring traffic and your own test jobs generally survive it. Treat a built-in list as one signal, not as the end of the topic.
Should I delete bot sessions or hide them?
Hide them and keep the raw events. Deletion breaks historical comparisons and removes the evidence you need when a rule misfires. Segmentation keeps both the clean view and the audit trail.
Will excluding bots change my past numbers?
Your clean view will usually drop, because the inflated sessions were never converting. Annotate the change: state the date the filter went live and show both series for a while, so a reader can tell a metric shift from a data-quality fix.
Clean denominators are the cheapest accuracy upgrade available. If you want verdicts, reasons and labelled events in one place — and bot sessions kept out of the analytics you report on — start with Prynt: pryntid.com/platform.
Try it free
Prynt is device intelligence with a free plan: 5,000 identifications a month, visitor IDs with confidence and bot detection — no credit card. Upgrade to Pro for VPN, proxy, tampering and behavioral signals.