Generic intent classifiers map crypto queries into 4 SaaS buckets. Web3 actually splits into 6, including investigation (commercial with risk overlay) and disambiguation (token vs protocol vs company).
Click the example queries to see how the classifier maps each one across the 6 buckets, plus what it recommends as the matching content type.
Primary intent
Investigation
InvestigationCommercial overlayUser wants to compare yield options but expects risk disclosure first. Pure commercial pages bounce; informational-first pages with a clean "compare" section convert.
Recommended content
Comparison guide
Article + FAQPage schemaLead with risk and methodology, then comparison table with FinancialProduct schema per protocol. Add Q-and-A blocks for "how is yield calculated" and "what are the risks".
Click any query above to see how the classifier maps it. Outputs are illustrative; the live module produces the same shape for any keyword corpus on your domain.
Each bucket maps to a different content type, schema combination and CTA pattern. Mismatches between keyword bucket and page content are the most common single ranking issue on crypto sites.
User wants to learn. No commercial intent. Generic classifiers handle this bucket well; we keep it for completeness.
Example queries
Recommended: Long-form explainer · Article schema · No CTAs above the fold
User is going somewhere known. Brand or feature names, app actions. They want to land fast and act fast.
Example queries
Recommended: Action-first utility page · HowTo schema · No marketing copy
User is evaluating products to use. Higher-funnel commercial. Pure product pages work here.
Example queries
Recommended: Product comparison · FinancialProduct + Article schema · Visible "Try it" CTA
User is ready to act now. Direct conversion path. Less explanation, more action.
Example queries
Recommended: Action-first product page · CryptoExchange schema · CTA above the fold
Unique to YMYL/crypto. Commercial intent but with risk disclosure expected first. Pure product pages bounce.
Example queries
Recommended: Comparison guide with risk-first opener · Article + FAQPage + FinancialProduct schema · Full disclosure
Unique to crypto. Query references a symbol mapping to multiple entities (token, protocol, company, foundation).
Example queries
Recommended: Entity-resolved page · Distinct Organization + FinancialProduct entities with @id · Canonical strategy
This is the most common single ranking issue we see across crypto sites. Almost always fixable with copy and structure changes, no migration needed.
The /yield page targets "best DeFi yield protocols 2026". Users searching this come to compare and evaluate, then bounce on a transactional pitch.
→ Recommended fix path
Four phases, all in parallel with the seven other audit modules. The classifier itself is the Gemini AI Analyzer's Crypto Keyword Intent Classifier check.
Crawlux extracts the per-page primary keyword from each page on your domain by parsing title tags, H1s, meta descriptions and structured data. Secondary keywords are pulled from H2s and the surrounding context. The corpus is deduplicated and normalized before classification.
Each keyword is sent through the Gemini AI Analyzer's Crypto Keyword Intent Classifier. The model returns a primary bucket, an optional secondary overlay (e.g. Investigation + Commercial), a confidence score and the recommended content type. Low-confidence classifications are flagged for manual review.
In parallel, the page content itself is classified by the same model. CTA placement, headline structure, content depth, schema present and presence of risk disclosure all feed the page-intent verdict. The two verdicts are then compared per page.
Pages where keyword intent differs from page intent are flagged. Each mismatch is rated by gap size and effort to fix. The PDF report includes the recommended content restructure per page, with example sections from competitors who match the intent correctly.
Same crypto query set, three different verdicts. The 4-bucket models conflate Investigation into Commercial and Disambiguation into Navigational, which costs ranking on the misclassified pages.
| Query | Ahrefs intent | Semrush intent | Crawlux 6-bucket |
|---|---|---|---|
| "best DeFi yield protocols 2026" | Commercial | Commercial | ✓ Investigation (commercial overlay) |
| "safest stablecoin to hold" | Commercial | Commercial | ✓ Investigation |
| "aave price" | Navigational | Navigational | ✓ Disambiguation (token vs protocol) |
| "uni founder" | Informational | Informational | ✓ Disambiguation (token vs Uniswap Labs) |
| "connect metamask" | Navigational | Navigational | ✓ Navigational (agreement) |
| "how does liquid staking work" | Informational | Informational | ✓ Informational (agreement) |
| "swap eth to usdc on uniswap" | Transactional | Transactional | ✓ Transactional (agreement) |
| "highest APY crypto savings" | Commercial | Commercial | ✓ Investigation (regulatory risk flag) |
| "buy bitcoin with credit card" | Transactional | Transactional | ✓ Transactional (jurisdiction flag) |
| Crypto-aware regulatory risk flag | ✕ Not measured | ✕ Not measured | ✓ Per query |
Verdicts illustrative. Note: Crawlux agrees with Ahrefs/Semrush on roughly 70% of crypto queries; the 30% disagreement is concentrated in Investigation and Disambiguation buckets.
Common questions from teams running keyword strategy on crypto and Web3 sites.
Generic intent classifiers (built into Ahrefs, Semrush, Surfer) were trained on the open web. They classify queries into 4 buckets: informational, navigational, commercial, transactional. That works for SaaS. Crypto search behavior splits two of those buckets further. "Best DeFi yield" is commercial but with an investigation overlay because users expect risk disclosure first. "AAVE price" is navigational but ambiguous between AAVE token and Aave protocol. The 4-bucket model collapses these into the wrong bucket; the 6-bucket Crawlux model keeps them distinct.
Commercial (evaluating products to buy/use), Transactional (ready to act now), Informational (learning), Navigational (looking for a known site), Investigation (commercial but requires risk disclosure first, unique to YMYL/crypto) and Disambiguation (ambiguous query needing entity resolution, e.g. AAVE token vs Aave protocol). Each bucket maps to a different content type and ranking strategy. See the bucket grid above.
The classifier currently runs in English with high confidence and supports queries in 9 additional languages (Spanish, Portuguese, French, German, Russian, Turkish, Korean, Japanese, Chinese) at lower confidence. Language detection runs first; if confidence is below threshold the query is flagged and skipped from the audit. Multi-language support expands per quarter as we collect labeled query data.
Yes. Disambiguation is one of the 6 buckets specifically because it is so common in crypto. Many queries reference symbols (BTC, ETH, AAVE, UNI) that map to multiple entities (the token itself, the protocol, the company, the foundation). The classifier flags these and recommends the page implement disambiguation schema (Organization vs FinancialProduct distinct entities with @id references) so AI models can resolve which entity the page is about.
Ahrefs uses a 4-bucket model (informational, navigational, commercial, transactional) trained on the general web. The model is reliable for SaaS and e-commerce queries. For crypto-specific queries, it tends to over-classify into commercial (everything that mentions a token or protocol) which masks the investigation and disambiguation patterns Crawlux flags separately. Both tools agree on roughly 70% of queries; the 30% disagreement is where Crawlux adds value for crypto sites.
Search engines and AI models reward content that matches user intent. Ranking commercial content for an informational query produces high bounce and low engagement, which feeds back into ranking. The Crawlux audit flags pages where the page content does not match the primary keyword intent. Most teams find 15-30% of their landing pages have intent mismatches that are easier to fix than building new pages.
Yes, in parallel. The Gemini AI Analyzer runs both the keyword intent classifier and the regulatory risk detector together. Queries that imply regulated financial activity (lending, derivatives, security tokens) in jurisdictions with stricter regulation (US, UK, EU under MiCA) are flagged so you can ensure the matching page has the right disclosures. This is more important for AEO than for Google ranking, since AI models surface jurisdiction-aware answers.
Common pattern: a landing page targets "how does liquid staking work" (informational) but the page is built like a product page with a prominent "Stake Now" CTA above the fold (transactional). Users searching the informational query bounce because they came to learn, not act. The fix is to lead with the explanation, then introduce the product mid-page once the user has the context. Intent mismatches like this are the most common single ranking issue we see across crypto sites.
RUN THIS AUDIT FREE
No signup, no credit card. Full 8-module report in 60 seconds.
Free first audit · No signup · 60 seconds · Full PDF report