How to read traffic sources and which ones to trust
Trust tagged traffic most, referrer traffic next, and treat direct traffic as a catch-all rather than a channel. A visit tagged with UTM parameters carries its own label, so it lands in the right bucket every time. A visit without tags is sorted by the referrer the browser sends, which works well for search engines and most websites. When a browser sends no referrer at all, the visit is counted as direct, even if the person actually clicked a link in an app, a PDF or a chat message.
That is why a sudden jump in direct traffic after a campaign usually means untagged links, not a surge of people typing your address. The fix is to tag every link you control: newsletters, social posts, partner placements and ads. Pick one naming scheme and keep it. Use lowercase, write utm_medium as the channel type (email, social, cpc) and utm_source as the specific sender (newsletter, linkedin, a partner's name). Mixed spellings such as Newsletter and newsletter split one campaign into two rows and make trend charts misleading.
A worked example. You run a spring promotion in your newsletter and on LinkedIn. Build both links in the Campaign URL Builder with utm_campaign=spring-sale, utm_source=newsletter or linkedin, and utm_medium=email or social. In the report, filter by the campaign and compare the two sources side by side on visitors, bounce rate and conversions. If LinkedIn brings more visitors but the newsletter brings more conversions, the newsletter audience is the one worth growing, and the LinkedIn post may need a clearer landing page.
Referrer details deserve a periodic look for two reasons. First, they show the specific pages that link to you, which tells you which articles or listings earn attention and where it is worth building a relationship. Second, they expose junk. If an unknown domain sends visits that all bounce, you can block it with a referrer traffic filter in website settings so it stops distorting your numbers. Keep in mind that filtered visits are never recorded, so check the pattern for a few days before you exclude a source.