Why I won't rank an AI site with only weeks of traffic history

Jason T. · 2026-10-03

±10× is the agreement band I use for keyword-led traffic estimates. Even with that tolerance, I won't rank an AI site without at least 3 months of traffic history and at least 5,000 visits in the latest month. The New Site Radar methodology documents these thresholds, derived from a 50-site accuracy test run in October 2026. The published thresholds come from the same configuration the scoring job uses.

What "verified" actually means

That band is wide. For a keyword-led site, "verified" means the two US organic traffic estimates are within ±10× of each other. TabAPI supplies the primary monthly visit estimate, with up to 12 months of history. Traffic data is published about a month later, so the current snapshot describes the previous full month. A domain can be new enough to discover but too young to rank. Listing requires a domain registration date on or after January 1, 2025. Ranking also requires at least 3 months of traffic history and at least 5,000 visits in the latest month. For someone tracking a launch on X, that means waiting. A few weeks of attention cannot satisfy a requirement for 3 months of traffic history.

Check 1: Match the scope before comparing numbers

For keyword-led sites, the second estimate comes from DataForSEO Labs. Its ETV figure covers estimated US organic search traffic. I compare it only with TabAPI's US organic figure, never with TabAPI's total visits. Otherwise, the comparison would mix different acquisition channels and geographies. I check whether the two US organic estimates are within ±10× of each other. The comparison also fails at the scope-checking stage if one input is total visits. When keyword-led estimates disagree beyond the band, I don't rank the site, and its page shows both values. Readers should be able to inspect the disagreement rather than see only the number that survived it.

Where Tranco fits, and where it doesn't

Brand-led and all-in-one sites get most of their traffic from direct visits and brand searches. For those sites, I check the traffic order of magnitude against Tranco, a public research ranking of the top one million domains. There is also a sudden-growth exception. A fast-rising keyword-led site outside the ±10× band is not marked unverified if its Tranco rank agrees. DataForSEO's ETV estimate lags behind sudden growth, so that rank provides an additional check. The routing itself deserves scrutiny. In the 50-site accuracy test, the brand-led versus keyword-led classifier matched manual labels 86% of the time.

Check 2: Inspect the growth baseline

Below a minimum previous-month base, I don't compute a growth percentage. The site gets a NEW label instead, so no "+2,000%" is shown from a near-zero base. The 5,000-visit ranking threshold applies to the latest month. The growth calculation uses a minimum previous-month base. This is where I would spend my attention before sharing a growth chart: on the starting volume, rather than the number of digits in the percentage.

Check 3: Identify single-source figures

Many young sites have coverage from only one provider. They can still rank if they meet the eligibility rules, but they carry a single-source badge. Read that as one estimate, not a confirmed figure.

What else must pass before ranking

New Site Radar lists AI tools whose domains were registered on or after January 1, 2025. I use RDAP registration dates, with WHOIS as a fallback. No registration date means no listing. Domains discovered on a newly-registered-domains list are re-checked against RDAP. The site must be an AI tool, rather than a news site, directory, template or landing-page shell. Its homepage must return 2xx after redirects. Adult, scam and illegal sites are excluded. Our offline language-model pipeline classifies tools and writes descriptions. It also reads pricing pages. It never produces traffic or revenue numbers. Revenue remains modelled, never known. A range appears on the revenue board only when the site's own pages contain real prices and a payment-provider fingerprint.

What to do with the next rising-sites list

Check the estimates' scope before calculating agreement. When comparing growth figures, inspect the previous-month base. The ranking applies these rules. If a site has only weeks of traffic history, I wait rather than assign it a position.