How to Read Your Web Analytics Dashboard Weekly (2026)

Why weekly, and why a fixed order

A dashboard is not a report. It is a set of panels that answer different questions, and most people read the panel that is nearest the top, feel good or bad, and close the tab. The result is a lot of looking and very little learning.

A weekly routine fixes three problems at once. First, a week is long enough to smooth out the day-of-week pattern that dominates small sites: a site that is busy on weekdays and quiet on weekends will look like it is collapsing every Saturday if you check daily. Second, a fixed order means you notice what is missing as well as what is there. Third, ending with one written decision turns the review from a habit of looking into a habit of acting. If you cannot name a change after fifteen minutes, write "no change, because X" and move on. That is a valid outcome.

This guide uses GhostMetrics panel names because the definitions are checkable in the product, but the routine works on any tool that shows trend, pages, sources and engagement. If you are still choosing a tool, how to measure website traffic covers the options first.

Set the window before you read anything

The dashboard has four period buttons: 24h, 7d, 30d and 90d. For the weekly review, choose 7d and leave it there. Two behaviors of that selector matter.

• It is a rolling window, not a calendar week. The query asks for events newer than seven days before now, so a Monday morning review covers the previous seven days ending Monday morning, not Monday to Sunday. Do the review at the same time each week and the windows line up.
• The comparison is automatic. Each stat card on the Overview tab shows a "% vs prev" trend against the equal-length window immediately before it. For 7d, that is the seven days before the current seven. This is the cheapest useful number on the dashboard, because it tells you whether anything moved without asking you to remember last week.

One subtlety: the trend on Bounce Rate is a relative change, not a difference in percentage points. If bounce rate went from 50% to 55%, the card reads +10.0%, because 55 is 10% more than 50. Read the card as "bounce rate is 10% higher than it was" and look at the two underlying values if the exact gap matters.

What the cards count is worth a sentence each, because the names are friendlier than the mechanics:

• Visitors is the number of entry pageviews: the first pageview in a browser tab, detected from a per-tab counter in sessionStorage. It counts tab sessions that started, not distinct people. The same person opening your site in a new tab tomorrow counts again, by design, because GhostMetrics keeps no cross-session identifier.
• Pageviews counts every pageview recorded, including route changes in single-page apps.
• Avg Time on Page and scroll are averaged only over pageviews that sent a leave beacon. A page whose beacon never arrived is left out rather than counted as zero.
• Pages / Session is pageviews divided by visitors, so it inherits the tab-session definition.

The 15-minute routine, step by step

Set a timer. The budget is part of the discipline: it forces you to look for the big thing instead of reading every row.

1. Overview cards (1 minute). With 7d selected, read the five cards and their "% vs prev" figures. Note which moved by more than the normal wobble for your site. After a few weeks you will know what normal is; until then, treat small moves on a small site as noise to confirm next week.
2. Traffic chart (2 minutes). Look at the shape of the week. Is there a spike, and does it match something you did (a post, a newsletter, a mention)? Is a day missing on a site that normally has traffic? The chart draws one bar per day that had pageviews and leaves out days with none, so count the dates rather than looking for an empty bar. A missing day usually means the tag came out of a template or a deploy broke it, not that nobody visited. Check the live site before anything else.
3. Pages tab, All Pages (2 minutes). Sort by Views. Read the top five. Ask whether they are the pages you want to be your top five. A page you did not expect near the top is the most interesting finding of most weeks.
4. Sources tab (3 minutes). Read Referrers, Traffic Type and the UTM Breakdown, in that order. Details on how to read each are below, including two counting quirks.
5. Entry and exit pages (2 minutes). On the Pages tab, read the Entry Pages and Exit Pages lists. Entries tell you where visits begin; exits tell you which pages were left. The full treatment is in the next section.
6. Engagement tab (2 minutes). Read the Scroll Depth Funnel and the average time. You are checking whether the pages that bring people in are being read.
7. Performance tab (1 minute). Open Vitals by Page and look for a high-traffic page with a conspicuously bad number.
8. Events tab and the decision (2 minutes). Read the Custom Events panel: event name, count and last triggered. Then write one sentence: what will you change, publish or check this week?

The table below is the same routine as a checklist to keep next to the dashboard.

Reading sources without fooling yourself

The Sources tab has three panels, and each one has a counting rule that changes how you should read it.

Referrers

The Referrers table lists the referring address recorded with each pageview, with visits and a percentage. Two things to know. First, it counts pageviews, not sessions, and the query does not exclude your own domain. On a multi-page site, a visitor who clicks from your home page to a second page arrives with your own address as the referrer, so your hostname can appear in the list. Skip it and read the external rows. Second, on a single-page app the browser's referrer value does not change when the route changes, so route changes repeat the landing referrer. Both effects inflate a referrer's visits relative to the number of people it sent. Use the table to rank sources against each other, not as a headcount.

Traffic Type

Traffic Type groups the referrer text into four buckets. An empty referrer is Direct. A referrer containing Google, Bing, DuckDuckGo or Yahoo is Search. A short list of social networks (Twitter, Facebook, LinkedIn, Reddit and the t.co shortener) is Social. Everything else with a referrer is Referral. It is a keyword match on the referrer text, so it is good for the direction of the mix and wrong at the edges: a network not on that list lands in Referral, an unrelated address that happens to contain one of those strings (microsoft.com contains "t.co") lands in Social, and clicks between your own pages count as Referral. Direct is also a catch-all, because a visit with no referrer header could be a typed address, a bookmark, a messaging app or a privacy setting that strips the header, and GhostMetrics cannot tell them apart. Direct traffic explained walks through the causes.

UTM Breakdown

The UTM Breakdown has four columns: source, medium, campaign and visits, for the top 20 combinations. GhostMetrics reads utm_source, utm_medium and utm_campaign from the landing URL and keeps the first set it sees in that tab's sessionStorage, then attaches it to every later pageview in the tab that carries no tags of its own. The consequence is that visits counts pageviews carrying that tag, so a campaign visitor who reads three pages adds three. The week-over-week comparison is still valid, because the counting rule does not change, but do not quote the number as "people from the newsletter". For naming conventions that keep this table readable, see the UTM parameters guide.

The weekly question for all three: is there a source that is new, bigger or gone compared with last week, and do I know why?

Entry pages, exit pages and the engagement panels

On the Pages tab, Entry Pages lists the ten pages where the most tab sessions began, and Exit Pages lists the ten pages most often left. Entries come from the pageview beacon, so they are the more reliable of the two. Exits depend on a leave beacon arriving, and a page that was left is flagged whether or not it was the last page of the visit, so the exit list tends to follow your traffic ranking. Read it as "which pages were left", compare it with the entry list, and look for a page that should hand readers onward but appears high on exits.

On the Engagement tab, two panels earn a place in a weekly pass.

• Scroll Depth Funnel. The share of measured pageviews that reached 25%, 50%, 75% and 100% of the page. It only includes pages that sent a scroll value on leave; a page with nothing to scroll sends none and is excluded rather than counted as 0%. The funnel covers the whole site, so compare it with last week, then check Avg Scroll for your top entry pages in the All Pages table. A low Avg Scroll on a page that earns traffic is a headline or opening problem.
• Average time on page. Seconds from page load to the first time the tab is hidden or the page unloads. If a reader switches tabs after ten seconds and comes back to read for five minutes, only the first ten seconds are recorded. That makes it a conservative measure, best used for comparing the same page week to week.

Neither number is a verdict on content quality. A short page that answers a question quickly can have a low time and a high satisfaction. For how the scroll data is collected, see scroll depth tracking without Google Analytics, and for how bounce rate relates to engagement, bounce rate vs engagement rate.

Vitals and events: the two quick checks

The Performance tab shows Core Web Vitals captured from real visitors: FCP, LCP, CLS and TTFB. INP is not captured. For a weekly pass, ignore the averages across the whole site and open Vitals by Page. You are looking for a page that is both busy and slow, because that is where a fix helps the most people. TTFB is measured from navigation start to the first response byte, and the dashboard rates it against an 800 ms "good" threshold; for FCP, LCP and CLS, compare against Google's published thresholds. If a number shifts a lot in one week without a deploy, suspect a mix change (more mobile visitors, a new country) before suspecting the site. Core Web Vitals real user monitoring explains why field data differs from a lab test.

The Events tab shows custom events as name, count and last triggered. Your weekly check is binary: did each action you care about happen, and when did it last fire? A "Last Triggered" date from three weeks ago on a button you expect people to click means either nobody clicks it or the attribute fell off during an edit. Events carry the page path and any session UTM, but the panel has no per-campaign filter; if you need events by campaign, use the raw CSV export, which is available on every plan.

What to ignore, and what to write down

A lot of what a dashboard shows is real but not actionable at weekly resolution. Skipping these is not laziness; it is how you protect the fifteen minutes.

• Single-day spikes and dips. One post shared in one place produces a one-day bump. Look at whether the week moved, not whether Tuesday did.
• Percentages on small bases. A page with three entries and one bounce has a 33% bounce rate. That is a coin flip, not a finding. Per-page bounce is bounces divided by the page's entries, so a page with few entries gives a jumpy figure, and a page with none shows no bounce rate at all.
• Old bounce history. If your range reaches back before the October 2026 bounce-counting fix, older rows over-count bounces, so a long-range bounce trend mixes two rules. Compare recent weeks only.
• Geography and device splits unless you are deciding something specific, such as whether mobile deserves a layout fix.
• Organizations in Who's Looking. It names networks it can identify and excludes residential ISPs, so it is a hint about companies, not a list of readers.
• The 90d view in a weekly review. Use it quarterly.

Then write the decision. Keep it in one line in a notes file with the date, the three numbers that prompted it and the action. In a month you will have four lines; in a quarter you will have a log of what moved the numbers, which no dashboard can give you. If you want to share the weekly view with a teammate, client or sponsor, a Pro public dashboard shows the same panels without a login, covered in sharing website analytics publicly.

FAQ

Keep reading

Disclosure: GhostMetrics is built by Null Agency and we use it on this site, so we are not a neutral party about our own product. The routine works with any tool that shows trend, pages, sources and engagement; panel names and counting rules described here are specific to GhostMetrics. Check the product documentation for current behavior before relying on a definition.