How to estimate a competitor's monthly visits, and its error

You can estimate a competitor's monthly visits within a useful range using free ranking data, but you should never present the result as an exact count. The honest way to do it is to gather several independent signals, convert them into a band, and state how uncertain that band is. This guide walks through the process and explains how wrong the answer can be.

Start by defining what you want

Competitor research usually needs one of three answers.

1. Scale. Is the competitor ten times bigger than us, about the same, or much smaller?
2. Direction. Are they growing, flat or shrinking?
3. Composition. Where do their visitors come from, and through which channels?

An estimate can answer the first two reasonably well. The third needs more work and more caution. Be clear about which one you need before you start, because it determines how much error you can tolerate.

Step 1: Collect position data from several lists

Begin with popularity rankings, since they are free and comparable. Check the competitor in as many independent sources as you can. The main ones are the Tranco list, which averages four rankings over 30 days, plus Cloudflare Radar and the Chrome UX Report.

If the domain appears in all of them at a similar level, your confidence is high. If it appears in one and not others, note that and find out why. A domain missing from the Chrome data may be an API product or a site with a mostly mobile iOS audience.

The StartupsIntel traffic checker collects five of these lists in one report, so you do not have to look them up one by one.

Step 2: Convert position into a visit band

A position is not a visit count, but the two are related. Top sites get far more traffic than mid-ranked ones, and the relationship is steep. A model fits a curve to sites whose traffic is publicly known and uses it to turn each position into a range of visits.

Two rules keep this honest.

• Use a band, not a point. Report "roughly 200,000 to 400,000 visits a month," not "287,431."
• Widen the band as rank falls. At the very top, a few places can mean millions of visits. Lower down, positions are so densely packed that a model cannot separate neighbors.

If you want to see how this works in detail, our methodology page explains the inputs. Our guide on how rankings translate into visits goes deeper.

Step 3: Cross-check with real-world signals

Before you trust the band, check whether it fits what you can see.

• Content velocity. A site publishing daily and running a large team usually has meaningful traffic.
• Hiring and pricing. A company hiring in several functions and charging real money needs a customer base.
• Brand search interest. If people search for the name, the direct audience is real.
• Product signals. Public customer logos, app store ratings and community size all hint at scale.

None of these produce a number, but they can reject an implausible estimate. If the model says 5 million visits and the site has a three-page blog and no team page, revisit the assumption.

Step 4: Estimate the second-order numbers

Once you have visits, you can reason about what else follows. This is where it becomes a business question rather than a traffic one.

For an AI tool, for example, StartupsIntel's revenue estimate multiplies visits by an assumed 3 percent paying share and the lowest paid plan price, weighted 80 percent monthly and 20 percent yearly. It is clearly labelled as an estimate. The point of making the assumptions explicit is that you can change them. If you know the product converts better, adjust the percentage and see what changes.

How wrong can it be?

Honest answer: more than most people expect.

A small Screaming Frog experiment compared three well-known estimators with real analytics for 25 sites. The author reported that estimates for low-traffic sites were mostly at least 40 percent off, that long-tail traffic was underestimated, and that the best use of the tools was spotting trends and comparing competitors rather than measuring exact traffic. The test was small and the author acknowledged it was not rigorous, so treat it as a sign of magnitude rather than a precise figure.

Those were paid-quality tools. A model built on free ranking lists should be expected to behave at least as badly at the low end.

A reasonable mental model:

Site size	Typical confidence	What the estimate is good for
Very large	High on scale, moderate on exact numbers	Ordering, market share, trends
Mid-sized	Moderate	Ordering, rough scale, direction
Small	Low	Whether it is alive, nothing more
Very small or new	Very low	Nothing reliable, use other signals

Errors can run either way. A site that gets most of its visitors from a logged-in app, email or a mobile app may look smaller than it is. A site with many bot visits or heavy crawling may look bigger.

Reduce the error with these habits

Use the same method every time. Consistency means your errors are similar from month to month, so changes are real. Even a biased estimate is a good trend detector if the bias is stable.

Compare ratios, not absolutes. Saying "they are about three times our size" survives errors that cancel, because both sites go through the same model.

Track over time. One data point is a guess. Six monthly points is a trend. The trending leaderboard is a quick way to see who is climbing.

Look at several competitors. If you check ten competitors and one looks wildly off compared with the rest, that is the one to investigate.

Keep a record. Write down the date, the source and the range, so you can see how you were wrong later and improve.

What to do with a bad estimate

Sometimes the data simply is not good enough. The domain is too small, brand new or hidden behind a different domain than its main product. In that case do not invent precision.

• Say the figure is unknown and show the sources you checked.
• Ask the company directly if you are considering a partnership or purchase.
• If you are evaluating a business to buy, favor listings where revenue is verified from the payment provider rather than modelled from traffic.

A short example of the whole process

Suppose you want to size three rivals in the AI writing space. You look up each in the traffic checker and see that one sits well inside the busiest band of the public lists, one sits in the middle, and one barely appears. You convert those positions into wide ranges and note them down with the date.

You then compare the ranges to what you can see. The largest rival publishes constantly and has a large team page, which fits. The middle one is growing its content library, which also fits. The smallest has a thin site and little activity, so the low estimate looks right. Your conclusion is not a number but a ranking with confidence levels, which is exactly what a decision needs.

FAQ

How accurate can an estimate of a competitor's traffic be?

For large sites, the order of magnitude is usually reliable. For small sites, errors of 40 percent or more are common even with paid tools, so use a range and avoid false precision.

Can I get a competitor's exact monthly visits?

Only if they share their analytics. Public tools can only estimate. Treat any number that claims exact precision with suspicion.

Should I estimate visits or unique visitors?

Visits are easier to compare across tools, and most free models report them. Unique visitors depend on how a source identifies people, so they vary more from tool to tool.

How often should I update my competitor estimates?

Monthly is a good rhythm. Record the date and source each time, and compare the same data source month to month.