okaihub

Traffic Concentration Among AI Tools: What the Data Shows

Traffic data covers 9,114 of 20,947 AI products. Within that measured subset, distribution is steeply concentrated — with caveats stated.

Published 2026-11-13 · 2 min read

Traffic is the most requested statistic in this category and the least reliable one we hold. It is present for 9,114 of 20,947 products — 43.5% — and every figure in it is a third-party estimate rather than a measurement. This article reports what the data shows and where it stops being usable.

What the coverage gap means

The missing 56.5% is not a gap we can fill by inference. Assuming the unmeasured products resemble the measured ones would produce a distribution that looks reasonable and is wrong in the direction that matters most — it would overstate typical traffic, because the products too small to be measured are exactly the ones with the least.

For a founder the practical consequence is that comparing your own traffic to a corpus average is not meaningful. You know your number exactly; the corpus number for a comparable product is either an estimate or absent.

What the data can support

  • Relative comparison inside a category, where products are measured the same way and the panel bias applies roughly evenly. Useful for spotting which tools lead a niche.
  • Order-of-magnitude grouping — separating products with negligible traffic from products with substantial traffic. Precise figures are less reliable than the buckets.
  • Presence as a signal. Having measurable traffic at all means a product is large enough for a panel to observe it, which is itself information.

What the data cannot support

  • Any claim about average or median traffic across the AI tool market. The sample is not representative and the missing products are systematically different.
  • Traffic share or concentration ratios presented as market facts. Concentration measured across a panel-observed subset is not concentration in the market.
  • Trend claims over time. Our corpus is a snapshot from a single crawl on 2026-11-13, not a time series.
  • Comparisons between categories. Panel coverage varies by sector, so a difference between categories may be a difference in measurement.

Why we publish it at all

Because the alternative is worse. Traffic estimates circulate constantly in this industry, quoted without their coverage caveat, and used to make decisions. Publishing the figure alongside its limitations is more useful than publishing it bare or refusing to discuss it.

Where this directory shows traffic for a product, it is labelled as an estimate from a third party with the crawl date attached. It is never used to rank products, and never presented as a measured fact about the product's business.

A number with a stated coverage limit can be argued with. A number without one gets believed.

If you want to know how you compare

Measure your own traffic properly and compare it to your own history. Install analytics, segment by referrer — including the AI assistants, which strip UTM parameters and can only be identified by referrer — and watch the trend rather than the absolute. That is a smaller dataset than a corpus average, and it is the only one that is actually about you.

More in Measuring growth