> ## Documentation Index
> Fetch the complete documentation index at: https://docs.semilattice.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# User models

> What user models are, how they're built, and what makes them useful.

<AccordionGroup>
  <Accordion title="What is a user model?">
    A user model is an AI representation of a specific group of people. It's built from real human data and predicts how that group would answer questions they've never been asked. Think of it as a research panel that responds in seconds.
  </Accordion>

  <Accordion title="What kinds of user model are there?">
    Two. **answers-1** models predict audience-level answer distributions: what percentage of a group picks each option. Our 50+ ready-made user models are **answers-1** models and are free to use. **Anacreon** models work at the level of the individual: hundreds of simulacra, each one a model of a single person, so results can be segmented and re-cut any way you need. Custom user models we build for customers are **anacreon** models. Read the [announcement](https://semilattice.ai/blog/anacreon) or the [technical report](https://semilattice.ai/research/anacreon).
  </Accordion>

  <Accordion title="What ready-made user models are available?">
    50+ user models and growing. UK and US adults, 16 UK Consumer Finance segments (from Growing Families to Just About Managing), tech industry segments (developers, designers, data scientists), and B2B software buyers. Ready-made user models are free to use.
  </Accordion>

  <Accordion title="How are custom user models different?">
    Custom user models are **anacreon** models built from data about your market, your customers, your users. Each one is a population of individual simulacra rather than a single aggregate model, so you get individual-level answers you can segment after the fact. They're private to your account and tested for accuracy on held-out questions the model has never seen. **Anacreon** achieves a state-of-the-art ordinal alignment of 0.775, ahead of every published academic baseline. See the [technical report](https://semilattice.ai/research/anacreon).
  </Accordion>

  <Accordion title="What data do I need to build a custom user model?">
    Unstructured behavioural and attitudinal data about your users: qualitative interviews, open-ended survey responses, reviews, support conversations, community posts. Survey data with response distributions helps too. The more individuals covered and the more varied the data, the better the model generalises to new questions. [Get in touch →](https://semilattice.ai/research/anacreon) and we'll scope it together.
  </Accordion>

  <Accordion title="Can a user model get better over time?">
    Yes. For **answers-1** models, adding more seed data improves coverage and accuracy. More questions give the model more context to draw on when predicting new ones, and accuracy is re-evaluated automatically after each update. For **anacreon** models, we can retrain the model on new data to keep it current with your users.
  </Accordion>

  <Accordion title="What happens if the model doesn't know?">
    Every user model has an accuracy estimate, measured on questions where we know the real answers. Models are validated against a wide set of relevant test data, covering the kinds of questions the model is built to answer, so accuracy estimates are as representative as possible. For a genuinely new question, accuracy is definitionally unknown: there is no ground truth to compare against yet. The model's accuracy estimate is your guide to how much to trust it.
  </Accordion>
</AccordionGroup>


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.