Semilattice

Audience Chat lets you discover things about your audience in seconds. Whenever you’re thinking about user problems, market trends, or customer insights—just open a tab and ask. The agent can answer simple questions for day-to-day unknowns or construct multi-step explorations to get to the bottom of a problem. It uses the Semilattice API to predict specific audiences. Accuracy is benchmarked at >87% and results arrive in <20 seconds. semilattice.link/KMJhLhg DM to talk about: - a custom model of your audience - a custom agent for your use case or client - MCP and API for your agents

We audited 1,000 @SemilatticeAI predictions hoping to find a signal which would predict inaccuracies. We failed, but the results changed how we understand our models. Read our blog → semilattice.ai/blog

Multi-segment simulations are live in @SemilatticeAI Run one simulation across multiple user models. Paired results, segmented by user model. Try it free → semilattice.ai

User journey simulation is live on @SemilatticeAI . Give it a URL. Simulated users navigate your product. Real browser sessions, real clicks, results in minutes. Five modes: findability, conversion, comprehension, discovery. Try it now at semilattice.ai.

We rebuilt the Semilattice results screen around the decision process. Findings organised by confidence. Charts alongside synthesis. Audience accuracy shown. Export to PDF, CSV, Markdown, chart images. Share via URL. docs.semilattice.ai/changelog

You can now open any audience in Semilattice and see exactly how accurate it is. Every audience has a detail page with accuracy scores, a testing toggle that separates seed data from validation, and paired bar charts so you can compare ground truth against predictions question by question. A model at 89% means nothing if you can't see where the 11% falls. Changelog: semilattice.ai/changelog Blog: semilattice.ai/blog/audiences…

Every product decision in consumer finance is a "what if." What if we move this feature behind the paywall? What if we lead with security instead of convenience? What if younger users love this but older homeowners don't care? To help product managers answer these questions we just launched 16 UK Consumer Finance audience simulation models in Semilattice. They let product manager simulate how the people that matter to you would respond, using real research data, and returns predictions in minutes. semilattice.ai/blog/fintech-a…

Just over a month ago we launched a demo research agent to give users an easier way to start using @SemilatticeAI audiences. One type of challenge came up consistently: product managers struggling to find data to back up early product decisions and the drain of then getting them aligned with their teams and execs. Today we’re pleased to launch the next iteration of @SemilatticeAI, focused on helping product managers make and align their early product decisions using Semilattice audiences. Find out more: - check out the video for a full walk-through - check out our blog for details of what we built: semilattice.ai/blog/less-nois… - DM for a demo. We’d really appreciate your feedback.

We’ve launched an MCP server for Semilattice—a tool for LLMs so they can do computational research for you. You can say something like “use Semilattice to research how designers are using AI” and it'll select the right audience model, come up with research questions, run simulations, and summarise results. You can even ask for a PDF. It’s powered by the Semilattice API, so audience models have known accuracy, predictions take ~15 seconds, and you can view everything in the dashboard. Visit the docs to see how to add the MCP to ChatGPT, Claude, and other AI clients: semilattice.link/sBemwEx

This demo attempts to mirror how your brain thinks about research: rapid questions branching off in different directions on different topics—some followed, others not. You need answers as quickly as possible to inform new questions, new branches. The tool keeps track of where you've been, letting you go back to explore different paths; letting you zoom out and take it all in. It's only possible with simulated audiences. Otherwise you're stuck on question one for hours, days, weeks; or having to come up with all the questions in advance. Or just not bothering. semilattice.link/ewyjdiq

We added this 3D visualisation to our models in the Semilattice dashboard. It shows model data in a 3D space so you can easily spot semantic similarities. BUT, the most important question is: How fast can we spin this thing? 🌀 The answer: over 8,333 orbits per second! Faster than my screens refresh rate, sadly. See it spin at a reasonable speed here: semilattice.ai/populations/fc…

It’s easy to make new audience models with Semilattice, you just upload some survey data—but not everyone has data! So we’re offering to build audience models for you. Just let us know the audience you want, and we’ll work out how to source the data and build it. semilattice.link/0hcCNYR You could get a model of your customers, users of a specific feature, a new target segment—any audience you can describe. And when you’ve got a model, you can use it to accurately predict new questions. E.g. to test copy, explore jobs to be done, prioritise features, understand personas, etc.

Our latest demo takes a B2B website URL, uses Semilattice to run PMF-focused audience predictions, and then generates a PDF report with a big score /100. Try it here: semilattice.link/TuQufH9 Here’s how it works: 1. You enter a website URL, it scans it 2. It matches the website to a Semilattice population model—an AI model of an audience 3. It then comes up with 12 questions to ask that audience model, running those predictions through Semilattice 4. Finally, it uses the predicted data to generate a report on the website’s PMF, giving it a score / 100 We’re building demos to show what’s possible. Our focus is the API, not the applications. DM if you’d like to build something with us!

To celebrate the launch of Tron Ares, we made a video game. It's more like Family Feud though. You play against the Semilattice API at predicting your audience 🕹️ Play here: semilattice.link/ZSFIymg

Semilattice lets you use the data you have about your users, customers, or market to predict the data you don’t have about them. But how do you know it works for the thing you want to predict? Today, we’re excited to announce test batches. A new feature which lets you test population models against use case-specific data. It lets you see if a model can do a specific thing, like predict jobs to be done or marketing preferences or pricing sensitivity. To demonstrate, we created a population model of UK adults using data about their demographics, attitudes to AI, and personality traits. We then got this model to predict a set of questions about finance—a different subject entirely. The result? 87% accurate. Learn more: semilattice.link/batch

Semilattice accurately predicts audiences - so let’s use it for rapid headline testing: semilattice.link/headline 1. Input your website headline 2. Generate alternatives 3. Hit the blue button! It uses Semilattice to predict which headline is most popular with your audience, in seconds 🤯

Predictions on Semilattice are now 3x faster! They're done in <15 seconds, compared to ~45 seconds previously. That's 240 predictions per hour. Get started with the docs: semilattice.link/niDLaMd

We used Semilattice to predict developer reactions to different website headlines. 63.9% preferred "User insights as infrastructure" to "User simulation for AI workflows", so we shipped it! The Developer audience model is estimated at 93% accuracy, but you can create custom audience models with use case specific data too.

When you can predict users like you make database queries, the long tail of decisions stop being guesswork. Watch our new *demo* to see how every line of copy can now be audience-tested. ► semilattice.link/copy