Conversion rate by traffic source answers one question: which channel sends people who actually do the thing you want? Most dashboards answer a different one, because they divide conversions by every session that arrived — crawlers, scrapers, monitoring checks and AI agents included. The percentage looks precise and is wrong, usually by a different amount for each channel.
This guide fixes the denominator first: set a goal on the event or page that means money, keep non-human sessions out of the count, then read daily conversion and conversion by UTM source or referrer on human sessions only.
We will use Prynt Analytics, our privacy-first product analytics where bot sessions are excluded and Global Privacy Control is honored, but the method applies to any tool that lets you filter traffic.
Start with the denominator, not the percentage
A conversion rate is conversions divided by sessions. Change the denominator and every source moves — but not equally. Link previews, uptime checks, scrapers and AI agents land hardest on the pages you promote most, so the channel where you spend the most effort tends to collect the most non-human sessions and its rate gets dragged down. Meanwhile a quiet channel with clean traffic looks like your best performer, and you move budget toward it and wonder why revenue does not follow.
Fix it once, at the edge, before events are written. Our bot detection covers the signals — headless browsers, datacenter ranges, tampering — that decide whether a session is human. Analytics then excludes bot sessions from overviews, goals, funnels and retention, so your denominator is people.
Set a goal on the page or event that means money

Pick one outcome per question
A goal should be something a person does on purpose: account created, trial started, checkout completed, demo requested. In Analytics you set a goal from a custom event with property filters, or from a page view. Keep the first one narrow: “signup completed, plan = pro” tells you more than “any signup”, because you can compare sources on the outcome you actually sell.
Use a page goal when the event is not instrumented yet
Legacy checkout, third-party portal, a partner’s success page — set the goal on the confirmation URL and put “proxy” in the goal name, so nobody later mistakes it for a tracked event.
Write the definition down
One line in your notes: goal name, event or page, filters, owner, date. Conversion rates drift when definitions drift, and nobody remembers the old one.
Read UTM source and referrer separately
UTMs describe the campaign you tagged. The referrer describes where the visit actually came from. They answer different questions, and most tools overwrite one with the other, so keep both views.
UTMs for links you control
Tag every paid and owned link, lowercase and stable. “LinkedIn”, “linkedin” and “li” become three sources in the report, and then you stop trusting the report.
Referrers for links you do not control
Untagged mentions — a newsletter, a forum thread, an AI answer engine — show up as direct traffic or as the linking site. Scan the referrer breakdown daily. When a new referrer converts, add it to your notes so next week you can compare it properly.
When they disagree
If a campaign tag and a referrer conflict, the tag usually wins. That is right for campaign reporting and wrong for discovery reporting, so keep a referrer-only view as the second read.
Read it daily, and read the change
Pick the same time each morning and look at three numbers: human sessions, goal completions, and conversion rate by source for the last seven days. One day is too noisy to act on; seven days catches a broken tag without drowning in weekend noise.

Then look at what changed:
- Many sessions, almost no completions. The traffic is not human, or the page does not deliver what the link promised.
- Small volume, high rate. Probably one returning visitor. Do not move budget on it.
- A source that vanished. A campaign tag changed. Find it before you report a slump.
- Everything fell at once. The goal broke, not the traffic. Test the event.
Then fix one thing a week: a tag, a page promise, a funnel step. Funnels of two to eight steps show where humans drop; goals tell you whether they arrived.
Cross-check before you act
Compare filtered and unfiltered numbers for the same source. A rate that swings hard when bot filtering is on means that source is attracting non-human traffic — useful news for whoever buys it. Agent traffic is its own category, and our AI agent detection page explains why a research agent and a scraper are not the same visitor.
One more check: a “returning” visitor whose device signals change on every visit deserves a flag before you count it as a loyal human. Device fingerprinting is where that analysis starts.
FAQ

What is a goal in Prynt Analytics?
A goal is your definition of success: a custom event, optionally filtered by properties, or a page view. Once it is set you can read daily completions and conversion by UTM source or referrer on human sessions only.
Why does my conversion rate differ between tools?
Usually the denominator. Tools disagree on bot filtering, on sessions without JavaScript, and on privacy signals such as Global Privacy Control. Compare on the same date range and check which tool excludes bot sessions.
Should I trust UTM source or the referrer?
Both, for different questions. UTMs tell you which campaign you tagged; the referrer tells you where the visit came from. Keep two views and never merge them into one column.
Read conversion on human sessions only
Set one goal, give it a day, and compare sources on people. Prynt Analytics is $39/month with a 14-day trial, and bot sessions do not reach your reports. Start at pryntid.com.
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.