How AI Tokens Are Scored: Inside the AiTokens AI Score

The AiTokens AI Score is one number from 0 to 100 for every AI token we track, about 670 of them. It is calculated automatically every 30 minutes from public data (market data and public code activity) across five pillars: development, liquidity, supply, resilience and momentum. The same model applies to every token, no one enters numbers by hand, and no one can pay to change a score. This article explains what each pillar asks, why it matters, and what the score does not measure.

What You Can Rely On

The exact weights and calculations behind the AI Score are proprietary. A published formula becomes a target: projects could tune their numbers to it instead of building. What is public is everything you need to judge the score for yourself:

• What it measures: the five pillars below, and the question each one answers.
• Where the data comes from: public market data, refreshed every 30 minutes, and public code activity, refreshed daily.
• How it behaves: development counts most and momentum least, so building matters more than hype.
• Why a token scores the way it does: every token's page shows its five pillar scores.

We refine the model as the sector changes.

The Five Pillars

Development: Is anyone building it?

Recent work in the project's public code, and whether it is still happening. Active developers are the standard fundamental gauge of a crypto network; Electric Capital's annual Developer Report tracks them for the same reason. A project that publishes no code earns no credit here, because there is no public evidence of building. Development counts most in the overall score, so projects that ship outrank projects that only market.

Liquidity: Can I get in and out?

Whether the token trades with real depth. Raw volume isn't taken at face value: Cong, Li, Tang and Yang (2023, Management Science) estimate that wash trading made up over 70% of reported volume on unregulated exchanges, so trading that looks manufactured counts against a token rather than for it.

Supply: Will unlocks dilute me?

How much of the token's eventual supply is already on the market, and whether supply is capped. Keyrock's 2024 study of more than 16,000 token unlock events found that unlocks put persistent downward pressure on price, larger ones more so. Our guide to fully diluted valuation explains why this matters.

Resilience: How badly could this hurt me?

The token's risk profile, from how sharply its price moves to how long it has survived. Liu, Tsyvinski and Wu (2022, Journal of Finance) find that size is a priced risk factor in crypto, and CoinGecko research found that more than half of all tokens it has ever tracked had failed by the end of 2025, so survival is information.

Momentum: Is the market agreeing?

The recent price trend and investor attention. Liu and Tsyvinski (2021, Review of Financial Studies) show that both help predict crypto returns. Momentum counts least, on purpose, so a hype run can't carry a token with nothing underneath it.

How the Pillars Combine

The overall score combines the five pillars, with development counting most and momentum least. If a pillar can't be measured yet (usually development, before a project's code has been checked for the first time), the score uses the other four and is marked provisional. Market-based pillars refresh every 30 minutes; development and attention refresh daily.

What the AI Score Does Not Measure

Knowing a score's limits is part of using it well. The AI Score does not measure:

• Whether a project's AI works, or who uses it. Usage, fees and revenue aren't reported consistently across AI tokens yet.
• On-chain activity, holder concentration, team, partnerships or legal status.
• Code quality. Activity shows that work is happening, not whether the work is good.

Use the score as a consistent first filter across the whole sector, then do the checks it can't. Our research checklist and on-chain verification guide cover them.

Frequently Asked Questions

How often is the AI Score updated? Every 30 minutes for the market-based pillars. Development activity and investor attention refresh once a day.

Why does development count most? Because sustained building is the hardest signal to fake. Volume can be wash-traded and attention can be bought; months of real work in public code are far harder to manufacture.

Why don't you publish the exact formula? Because a published formula is a target. We publish what each pillar measures, where the data comes from and every token's pillar scores, but we don't publish the exact weights or calculations, so projects can't tune their numbers to a published formula and so we can keep improving the model.

Can a project pay to raise its score? No. There are no hand-entered numbers; every input comes from public data, and the same model applies to every token.

What does "provisional" mean? One pillar, usually development, hasn't been measured yet. The score uses the other four until it is.

Is a high AI Score a reason to buy? No. The AI Score is research, not a recommendation to buy or sell anything. It shows where a token stands on five measurable questions; it can't tell you what the token will do next.

The Bottom Line

The AI Score turns five questions every investor asks into one comparable number for every AI token, updated every 30 minutes. It rewards projects that ship, trade on real volume and limit dilution, and it counts momentum least so hype can't carry a score on its own. It is a first filter, not a verdict. See the top 100 AI tokens by AI Score, or open any token's page to see its five pillar scores.