Two kinds of AI sites need two different traffic checks

Jason T. · 2026-10-03

"Verified" can mean two traffic estimates differ by as much as 10×. That threshold comes from New Site Radar's methodology, which compares matching US organic search estimates for keyword-led sites. I built New Site Radar, and the discovery pattern is the first thing I would inspect when comparing tools. A search estimate can be useful for a page people find through a task-based query. The same estimate can undercount an app people open directly.

Choose the yardstick before reading the number

New Site Radar sorts sites into keyword-led, brand-led or all-in-one categories. That classification determines which second check applies. For keyword-led sites, the second source is DataForSEO's estimated US organic search traffic. It is compared with the primary source's US organic figure, never with total visits. For brand-led and all-in-one sites, most traffic comes from direct visits and brand searches. A US organic estimate would undercount them, so the check uses Tranco rank instead. Tranco is a public research ranking of the top one million domains, aggregated from several providers. The difference is practical. When I see an estimate that looks too small, I want to know whether it measures the traffic I am trying to judge. Otherwise, I could reject an estimate because I compared it with a number that measures something else.

One hypothetical comparison: the same gap, different meaning

Consider a hypothetical "AI background remover" page that people find by searching that phrase. The relevant numerical question is whether its two US organic estimates fall within the 10× verification band. At the 10× boundary, the right comparison is DataForSEO's US organic estimate against the primary source's US organic estimate. That meets the stated tolerance, even though the estimates are far apart. The wrong comparison is DataForSEO's US organic estimate against total visits. One figure covers US organic search; the other covers total visits. Now consider a hypothetical AI chat app that people open by typing its name. It gets most visits directly, so its US organic estimate can be tiny even when it is busy. For that discovery pattern, I would inspect the rank comparison.

Keep the tolerances in one place

The keyword-led tolerance is ±10× between the two US organic estimates. Beyond that band, the site is not ranked and its page shows both values, subject to the fast-growth exception below. For brand-led and all-in-one sites, the allowed gap between the primary source's global rank and Tranco rank is about 16×, assessed in orders of magnitude. This checks rank against rank. Fast-rising keyword-led sites have an exception because search estimates lag behind sudden growth. If the search estimates are outside the ±10× band but Tranco rank agrees, New Site Radar does not mark the site unverified. Coverage has its own label: sites covered by only one provider are ranked with a single-source badge. Those are the tolerances and exceptions I need at hand. The useful question is which one applies to the site in front of me.

Check the category before trusting the check

The classifier matched manual labels 86% of the time in a 50-site accuracy test run in October 2026, reported in the methodology. Mistakes tend toward the more lenient check. The category decides whether the second check uses search traffic or rank.

A checklist to reuse

When browsing the rankings, I would use this checklist before comparing traffic numbers:

Identify discovery. Is the site keyword-led, brand-led or all-in-one?
Match the scope. For keyword-led sites, compare US organic with US organic, never with total visits.
Use the applicable check. Read the search comparison against its ±10× band, or the global-rank comparison against its about-16× tolerance.
Inspect exceptions and coverage. Look for Tranco agreement on a fast-rising keyword-led site and for a single-source badge.

This is also how I would decide what deserves a closer look. A large gap between matching estimates is a different issue from a small search estimate for a direct-heavy app. Keeping those cases separate makes the comparison useful without pretending the figures are precise.

The next test I would run

For a follow-up, I would track a fast-rising keyword-led site whose search estimates are outside the verification band but whose Tranco rank agrees. The question would be specific: do the two US organic estimates later move inside the ±10× band? That would test whether the search-estimate disagreement narrows after the sudden growth that prompted the rank exception. It is a proposed check, not a result. Use matching search figures for keyword-led tools and the rank check for brand-led or all-in-one tools.