With the same 10,000 visits, one site closes 30 orders off 5,000 people viewing two pages each; another racks up 10,000 visits from 200 people refreshing the page and closes zero. If traffic quality assessment only looks at totals, these two sites’ numbers could look identical. Bounce rate is just one entry point — the real question is: of this batch of traffic, how many people are actually reading the content.
Here’s the bottom line: to assess traffic quality, bounce rate can only be the first filter — you must pair it with deep-visit metrics: pages per session, average dwell time, scroll depth, and return rate. The standard practice is setting a separate threshold for each major page template, then looking at deviations grouped by source, region, and time slot. Traffic with a high deep-visit share deserves more budget; channels that only polish bounce rate should be cut back over time.
This article’s default scenario is a content-driven foreign trade site: articles bring traffic from organic search, product pages handle conversion. Metric weights differ by scenario, but the analysis framework is universal: first define “deep,” then split dimensions, finally tie to conversion. The whole article needs no extra tools — the free GA4 tier can run it all.
Bounce rate’s three blind spots
Bounce rate lumps “read an article and leave” and “open the page and close in two seconds” into the same category, which is especially unfair to content sites. Blind spot one: single-page sessions naturally run high — blog users read and leave by design; this kind of bounce isn’t bad. Blind spot two: it doesn’t distinguish returning users revisiting from first-time users arriving — a returning reader coming to see updates is also recorded as a bounce. Blind spot three is the most fatal: it doesn’t count whether users completed a goal action on the page — someone staring at the form filling it for two minutes and then leaving is also recorded as a bounce. So bounce rate is only suitable as the first filter, not as a conclusion.
Deep visits look at four metrics
Beyond bounce rate, use these four metrics to measure “how deeply people read.” Note that all of them must set baselines by page type — product pages and blog articles aren’t comparable on dwell time, and tutorial pages and homepages aren’t comparable on scroll depth.
| Metric | Definition | Judgment standard |
|---|---|---|
| Pages / session | How many pages viewed in one session | 2+ pages counts as deep visit |
| Average dwell time | Set baseline by specific page template | Below half of baseline needs checking |
| Scroll depth | How far the page is scrolled in percentage | 75%+ counts as fully read |
| Return rate | Share of revisits within 7 days | Content sites typically 10%-30% |
Set a baseline for each template first, then talk comparison. For how to set baselines, use the 75th percentile of the last 90 days of history as a reference — more reliable than guessing. If thresholds are too loose, all traffic becomes “qualified”; too tight, and you wrongly hurt a batch of genuinely converting users. After baselines are set, review them monthly — content formats and user habits change, and last year’s baseline may not fit this year.
Split by source, device, and region
Traffic quality must be assessed in segments — overall averages hide extremes. New visitors from brand keywords and new visitors from broad terms have completely different deep-visit expectations; mobile and desktop dwell times aren’t comparable either. Example: a site’s overall dwell is 1:40; split apart, desktop is 2:30 and mobile is 50 seconds — the real problem concentrates on one mobile template, not the whole site’s content quality. Segmentation analysis offers ready-made dimensions: plug in source, device, region, and new/returning visitors to find combinations with low deep-visit share, and handle them one by one.
Deep visits must tie to conversion
Deep visits aren’t a goal in themselves — they’re a leading signal for conversion. Do an intersection analysis of deep-visit users and converting users to see whether the threshold is right: if lots of converting users only view one page, the deep metric is set too high and should be calibrated against real purchase behavior. Many odd “ranked but no traffic” phenomena are essentially high-quality keywords being caught by low-quality pages — the symptom is a sudden drop in deep visits; for the causes of this type of problem, read about ranking-traffic decoupling. For the long-term impact of publishing rhythm on deep visits, cohort analysis of content is the right lens to observe monthly, looking at retention differences across batches of content.
Use data to decide channel and keyword changes
Once the data is in, sort every source’s deep-visit share: channels that stay at the bottom long-term and have high acquisition costs get gradually cut; keywords with high deep-visit share but small volume get more content coverage. The reduction experiment: cut one low-quality channel’s budget by 50% for two weeks; if core conversions don’t change, the cut was right. Write the cut criteria into docs in advance: channels whose deep-visit share stays below half the site average for four consecutive weeks enter the watch list. Use report automation to fix these metrics into a weekly dashboard, making traffic quality a daily monitored item rather than a month-end review — channel optimization then shifts from “cutting budget by feel” to “adjusting structure by data.”
Before you start, run through this round’s checklist: define deep-visit baselines for each major page template, recording the calculation method and date range; configure scroll-depth events in GA4 and verify the data is firing correctly; export four weeks of data by source, device, and region, flagging combinations with low deep-visit share; run a reduction experiment on low-quality traffic sources and watch whether core metrics rise instead; add deep-visit share to the weekly dashboard, displayed alongside conversion rate.
