Where can a small AI tool still win?

weigang zhang — Sat, 07 Nov 2026 11:45:55 +0000

Most conversations about the AI tool market start with a feeling: it's saturated. I wanted a sample I could actually count.

So I built an index of AI tools launched in the last 24 months with at least 3,000 estimated monthly visits. The September 2026 snapshot contains 2,006 tools and about 247 million estimated monthly visits. For each tool, the index tracks estimated traffic, month-over-month growth, acquisition channel mix, visitor countries, category and age.

This article is the short version of what the data says. The full report, including the methodology and the caveats, is free and does not require an account: Where AI Tool Builders Can Still Win in 2026.

The sample and why it matters

The index is deliberately filtered:

Age: launched within the last 24 months.
Traction: at least 3,000 estimated monthly visits.
Exclusions: adult/NSFW tools and tools that fail the eligibility rules are removed before any statistic is calculated.

That filter is the point. A raw list of every AI tool would be dominated by products with no traffic and would tell you nothing about where demand actually is. This is a sample of products that already have some traction — not a census.

The traffic numbers are third-party estimates from providers such as Ahrefs and Similarweb. They are directional, not audited, and they do not prove revenue, conversion or retention.

Finding 1: the top 10% take 71% of visits

Traffic is concentrated. The top decile of the index accounts for 71% of estimated visits. A generic "AI tool for X" enters a winner-takes-most distribution.

That is the headline, but it is not the whole story.

Finding 2: 28% of new tools still break through

Tools under 12 months old make up 978 of the 2,006 tools (49%) in the index. Among that cohort, 28% (273 tools) have already crossed 50,000 estimated monthly visits.

That number is more useful for a builder than the concentration number. It says the door is not closed. It also says breaking through is a minority outcome: most new tools do not cross that line in their first year.

The age distribution in the snapshot:

Product age	Tools	What it suggests
1–6 months	265	Early demand; too soon to call a durable winner
7–12 months	713	The largest cohort — where most new products compete
13–18 months	617	Enough history to compare repeat demand
19–24 months	409	The oldest eligible cohort; useful for retention signals

New is not enough. New plus distribution is the signal.

Finding 3: distribution is search and memory

Across the measured subset:

Organic search: 41.5%
Direct visits: 40.1%
Referrals: 9.0%
The rest is split across social, ads, email and AI referrals.

Organic search and direct visits together account for 81.6% of measured traffic. That is the most actionable finding in the report: most of these products are found because someone searches for a job to be done, or because they remember the product's name and come back.

A launch spike is not a distribution strategy. A search query and a repeat use case are.

The market is not one market

"AI tools" is a label, not a market. The category-level data looks completely different depending on where you stand:

Category	Estimated visits (top leaders)	Median MoM growth	Median organic share	Share of last-12-month tools above 50K
AI developer tools	29.1M (15 tools)	+4.1%	30%	53%
AI chatbots	13.8M (15 tools)	−7.9%	41%	33%
AI roleplay	8.1M (12 tools)	+13.9%	34%	67%
AI video	33.2M (15 tools)	−3.7%	44%	40%
AI YouTube tools	1.6M (12 tools)	−4.5%	42%	8%

A few readings:

AI developer tools have both scale and momentum. They also have the highest share of recent entrants above 50K, which means the category is still admitting new products.
AI chatbots are large but flat. A big total is not the same as an opening.
AI roleplay is smaller but growing fast, and two-thirds of recent entrants are above 50K. The catch: median organic share is only 34%, so distribution is more brand- and community-driven than search-driven.
AI video is the biggest total in this sample and slightly negative on growth. It is also the most crowded category, with 391 qualifying tools.
AI YouTube tools are small and shrinking in the sample; only 8% of recent entrants are above 50K.

The report's opportunity composite — demand, room for newcomers, momentum, organic reachability and observed paid-plan presence — flagged AI Music as the most open lane (81/100, 84 qualifying tools). Script-to-Video had the fastest median month-over-month growth (+19.3%). AI Video Generation was the most crowded (391 tools).

Categories are multi-label: one tool can appear in several categories, so category totals overlap. The index total does not double count a tool.

What I would do with this if I were starting today

I would not use the report to pick "the best category." I would use it to eliminate bad starting points and to design the first distribution loop.

Choose a market with momentum and room, not the biggest total. A large, flat category with 391 competitors is a harder starting point than a smaller category where recent entrants are still crossing 50K.
Narrow to one job, audience or workflow users can name. The data can tell you where demand is; it cannot tell you which wedge you can win. That still depends on your advantage.
Design distribution before the roadmap. If organic search and direct visits are 81.6% of measured traffic, the first question is: what query does this product own, and why would someone come back?
Set a traffic threshold before adding scope. Pick a weekly distribution target. Expand only when the channel repeats.

What this report cannot tell you

It cannot tell you whether a tool makes money. It cannot measure conversion, retention or churn. It cannot tell you whether a category is a good business — only whether it has traffic and how that traffic is distributed.

It also cannot see the long tail below 3,000 estimated monthly visits. That is a deliberate limit, not an oversight.

If you want the full numbers, the category cuts and the methodology, the report is here:

https://provenai.dev/reports/ai-tools-traffic-2026

If you are building an AI product right now, I would like to hear which number changed your mind — or which one you think is wrong.

Disclosure: I built ProvenAI, the index behind this analysis. Visits are third-party estimates (Ahrefs/Similarweb), not audited analytics. The report is free and does not require an account.

DEV Community: weigang zhang