Theme types
Attention is a subtype of Thematic ETFTs. A product can use AI-assisted research without using social-media popularity as its selection rule. The product methodology identifies whether and where AI is used; these capabilities are not implied for every ETFT.
From market information to a product
For AI-assisted products, the process separates theme discovery, evidence review, asset mapping, portfolio construction, and approval.
A generated theme or portfolio proposal is not an issued ETFT. Launch requires a defined mandate, supported assets, operational readiness, and product approval.
What AI contributes
AI can organize and summarize large volumes of public information, consolidate duplicate narratives, identify related entities, and explain why an asset may be relevant to a theme. It can also help generate a theme brief containing its scope, evidence, candidate assets, competing interpretations, and reasons the theme may fail. Every material asset-inclusion claim must be supported by retrievable evidence. A model-generated statement is not itself evidence. If a relationship cannot be established, the asset is excluded or referred for further review rather than admitted on the strength of the model’s confidence alone. AI outputs do not independently authorize issuance, withdrawals, execution, changes to NAV, or changes to a product’s investment mandate.Measuring attention
Attention-based methods distinguish several questions:- Intensity: How much eligible discussion or interest is present?
- Momentum: How has that interest changed relative to its own baseline?
- Breadth: Is the signal supported by independent sources and participants?
- Persistence: Does it continue across the required observation windows?
- Evidence quality: Are the sources credible, relevant, and sufficiently independent?
Connecting themes to assets
An investable connection requires more than a company being mentioned alongside a topic. Relevant evidence may include business activities, products, customers, supply-chain relationships, or disclosed exposure to the theme. The candidate universe is checked against instrument identity, venue support, liquidity, concentration, and execution constraints. Unsupported tickers and ambiguous company-to-instrument mappings are rejected. Where a theme references an unlisted business, its relevance does not create access to its shares. A derivative can be used only where explicitly permitted by the mandate and must be identified as synthetic exposure, not equity ownership.Portfolio construction
The product specifies its selection and weighting method. This may use equal weights, a defined thematic-relevance score, or another published rule, subject to concentration, liquidity, cash, and turnover limits. An AI-generated ranking does not bypass these constraints. A relevance score measures a relationship to the theme; it is not a forecast of investment returns. If too few assets qualify, the portfolio follows its cash or eligibility policy rather than filling the basket with weakly related names. Leverage and inverse exposure are separate product choices. AI discovery does not automatically increase leverage because a theme receives more attention. See Exposure structures.Monitoring and rebalancing
Research signals can update more frequently than a portfolio trades. The methodology sets the observation frequency, rebalance schedule, inclusion and removal thresholds, and turnover budget. Persistence requirements and different entry and exit thresholds can reduce repeated buying and selling around a noisy cutoff. New narratives are considered within the existing mandate; they cannot silently replace the product’s original investment objective. New product launches and material changes to the theme, methodology, or model-dependent decision process require review. Routine updates can proceed under an approved, versioned methodology after the applicable controls pass. See Portfolio and rebalancing.The Gen Z example
A Gen Z ETFT could select businesses associated with a defined consumer or cultural theme. An attention-based version could instead use changes in engagement within specified datasets as one input to selection.
Model and data controls
AI-assisted research can make factual, classification, and entity-matching errors. Historical evaluations also face look-ahead and survivorship bias. Products using the process retain source timestamps, dataset snapshots, methodology versions, and model-version records sufficient to investigate decisions. Historical simulations use information available at the simulated decision time. Where a model may contain later knowledge, a backtest is not presented as an uncontaminated historical strategy test; forward observation is needed to assess the process under realistic conditions. Source material is treated as data, not as instructions that can change system permissions or trading rules. Missing or unreliable inputs trigger the published fallback, such as holding the last valid portfolio pending review or suspending new signal-driven changes. Risk-reducing actions remain governed by the product’s risk policy.What each product discloses
An AI-assisted Thematic ETFT identifies:- The theme, eligible universe, and economic rationale.
- The stages where AI is used and where rules or human approval determine the outcome.
- Source categories, coverage, observation windows, and material data limitations.
- Selection, weighting, rebalance, and risk-control rules.
- Material methodology changes and the applicable version history.
- The basis for portfolio inclusion and relevant fallback procedures.