Reading month-over-month traffic changes and seasonality

A month-over-month traffic change is one of the easiest numbers to misread. A jump can come from real growth, a calendar effect, a change in how the data was collected, or plain noise. This guide shows how to check each explanation in turn, so the next time a competitor's traffic moves 30 percent you know whether to react.

Why one month proves very little

Month-over-month comparison asks a simple question: is this month higher than last month? The trouble is that months are not equal. They have different numbers of days and weekends. They sit at different points in the school year, the tax year and the shopping calendar. And estimates for any single month carry error.

A 10 percent rise from one month to the next is often within the error of the estimate itself, so on its own it says almost nothing. A 10 percent rise that repeats for six months says a lot.

The five causes of a traffic change

When a number moves, one of these is usually responsible.

1. Real growth or decline. More or fewer people are arriving because of the product, marketing or reputation.
2. Seasonality. The audience's behavior follows a calendar.
3. One-off events. A launch, a viral post, an outage, a press mention or a bot wave.
4. Measurement change. The data source changed its method, window or filters.
5. Noise. Random variation in a sample or an estimate.

Your job is to rule out causes two through five before you accept the first.

Seasonality: the patterns to expect

Almost every audience has a rhythm. Some common ones:

• Education and study tools. Dip in summer and over winter holidays, rise when school starts.
• Business and productivity tools. Strongest on weekdays, weaker in late December and during August in many regions.
• Consumer and entertainment tools. Often peak in holidays and weekends.
• Tax, finance and planning tools. Spike around deadlines.
• E-commerce. Rise toward the end of the year and around regional shopping events.
• AI tools. These can have both academic and work cycles, which overlap in odd ways.

The right comparison for seasonal sites is usually year-over-year, which lines up the same time in the calendar. If you only have a few months of history, compare to the same month last year when possible, or treat any change as provisional.

How to spot a seasonal pattern

Plot at least 12 months. Look for peaks and valleys that repeat at the same point each year. If you see them, create a simple seasonal expectation. If September is normally 20 percent above August, then a 20 percent September rise is not growth. It is just September.

Calendar effects you can check without data

Before you interpret a change, look at the calendar.

Effect	What to check
Days in the month	February has fewer days than most months, so totals dip even when daily traffic is flat
Weekends	A month with more weekends can differ for business tools
Holidays	Regional public holidays reduce work-related use
School terms	Start and end dates shift usage
Product events	Launches by the company itself or its rivals

A simple fix for the days effect is to compare average daily visits and not monthly totals. If your source only gives monthly values, divide by the number of days yourself.

Know what your source actually measures

Many apparent changes are really changes in the data. Sources differ in how they are built and how often they update.

• Tranco averages rankings over 30 days, so it smooths short spikes. A real jump will show up gradually.
• Cloudflare Radar publishes its larger lists as unordered buckets using the last 7 days. A site can shift buckets from week to week.
• The Chrome UX Report is released monthly for the previous month and covers only opted-in Chrome users.

If you mix those sources in one chart, differences in windows can look like growth. Stick to one source per series.

Also consider whether the source changed its method. If a list adds or removes an input, ranks can shift for reasons unrelated to any site. Tranco, for example, states that it no longer includes some sources that became unavailable. When many sites move at once in the same direction, suspect the data, not the sites.

Check the market, not just the site

A powerful test is to compare the site against its peers. If the whole category rose 15 percent, a competitor rising 15 percent has done nothing special. If the category was flat and it rose 15 percent, that is a real signal.

Use a group of 10 to 20 comparable sites and look at the median change. The bulk checker lets you pull the whole set at once. Then ask whether the target moved more than the group.

Separate trend, season and noise

You do not need statistics software for a basic decomposition.

1. Gather 12 or more months for the site.
2. Compute a 3-month average to smooth noise. The direction of that average is your trend.
3. Compare each month to its same month last year if you can. The differences show growth net of season.
4. Mark one-off events. Note launches or news for each month.
5. Look at the leftover. Months that sit far from the smoothed line need an explanation.

Even a rough version of this beats staring at a single percentage.

Rules for believing a change

Use rules like these before you tell anyone a competitor is growing.

• Persistence. The change holds for at least three months.
• Breadth. It shows up in more than one source.
• Magnitude. It is larger than the typical error for a site of that size. For small sites, that means a big change.
• Not seasonal. It exceeds what the calendar explains.
• Explained. You can point to a plausible cause, such as a launch or a content push.

If a change meets four of five, treat it as probably real. If it meets one or two, call it unconfirmed.

Special case: small sites

Small sites have the noisiest estimates, so their month-over-month changes are the least reliable. A tiny site showing a 200 percent jump may have gone from very little to a little. Wait for several months and look at ranks in more than one list. A rise that appears across them is more believable.

Special case: spikes

A sudden one-month spike followed by a return to normal is almost always an event, not a trend. Common causes are a viral post, a mention in a large newsletter, a social media campaign, or a bot crawl. If the number does not stay up in the following month, do not count it as growth.

Where StartupsIntel fits

The monthly leaderboards are built to show movement in a consistent way, using the same method each month so you can compare like with like. The methodology page lists the inputs, and the traffic checker gives a single-site view. Use them for direction, then confirm with the habits above.

A quick checklist before you report a change

Before you tell your team that a competitor has surged or collapsed, run through a short list. Did you compare the same source and the same window? Did you adjust for the number of days? Is the move bigger than the usual error for a site of this size? Did the whole category move together? Is there a launch, a news mention or a campaign that explains it? And does last year show the same pattern at this time?

If you can answer yes to the right ones, you can speak with confidence. If you cannot, say that the movement is unconfirmed and that you will recheck next month. That sentence costs nothing and saves you from many embarrassing reversals.

FAQ

Is a 20 percent month-over-month increase significant?

Not by itself. It could be seasonal, a result of different month lengths, or within the error of the estimate. Look for persistence across several months and agreement from more than one source.

Should I compare month over month or year over year?

Use month over month for short-term momentum and year over year for anything seasonal. If you have at least 13 months of data, year over year is usually the safer comparison.

How many months of data do I need to see a real trend?

Three months is a minimum for a tentative trend, and twelve lets you account for seasonality. More is better, especially for small sites.

What if different tools show opposite directions?

Check what each one measures and over what window. If they disagree on direction for several months, treat the change as unconfirmed and look for other evidence.