Developers

A simulated audience API for AI-native teams (in build)

In build

TL;DR

Neon Blue's simulation and creative pre-testing API is in build. It is meant to let AI-native teams ask a defined simulated audience a question, or show it a piece of creative, from inside their own product or workflow.
Neon Blue builds simulated audiences: AI models that answer the way a defined group of people would, grounded in Neon Blue's own paid studies and AI-moderated interviews with real people.
It isn't live yet. Today results come as a report, and the first study is free, so you can test the question you want to automate before you build around it.
Simulated results are always labeled and kept separate from real responses.

What is a simulated audience API?

A simulated audience API lets software ask a defined group of people a question and get back answers that reflect how that group would respond, without fielding a new survey each time. The audience is the product. A good one is specified by who the people are and grounded in real research with people who match them.

That is different from prompting a general model to "act as a 35-year-old shopper." A prompt gives you one plausible, average voice. A simulated audience is built from real people's interview and survey data, not generic LLM personas, so answers can differ by segment the way real answers do.

What will Neon Blue's API cover?

Two jobs, both in build:

Job	What you would send	What you would want back
Simulation	An audience spec and a question, such as which of three offers a segment prefers and why	Answers by segment, labeled as simulated
Creative pre-testing	An audience spec and a piece of creative, such as an ad, headline or message	Which version lands, with whom, and the reasons people give

Details will change while it is in build, so this page doesn't show request formats or code.

Who is it for?

Teams building AI products that need a read on people inside a workflow, not once a quarter:

Ad and content generators that produce many variants and want to rank them for a target audience before a user ships one.
Product and growth tools that help teams choose between pricing pages, onboarding flows or messages.
Agents that draft outreach, positioning or offers and need a check on whether the target buyer would care.

If you only need a read a few times a year, a study is simpler than an integration.

How should you design calls to a simulated audience?

The same rules that make a good survey make a good call.

Spec the audience tightly. Who they are, what they buy or use now, and the situation you care about. "Adults 18 to 65" answers like nobody in particular.
Ask one decision per call. Rank these three headlines, or say what you would do after this price change. Bundled questions blur the read.
Ask about behavior, not opinions of your product. "What would you do next" holds up better than "do you like this."
Hold the creative fixed. Change one thing between versions so the read tells you what moved.
Store every result with its audience spec and a timestamp, labeled as simulated, so you can compare it with live results later.
Close the loop. Check simulated rankings against real outcomes such as clicks, replies or conversions, and adjust how much weight your product gives them.

What should you check in any simulated audience API?

Grounding. Where do the answers come from? Ask whether the audience is built from real people's interview and survey data or from prompts alone.
Labeling. Simulated output should be marked as simulated all the way through to your users.
Coverage. Ask how the system handles questions far from what its grounding research covered. Those answers deserve less weight.
Your inputs. Ask how your prompts, creative and customer data are stored and used.
Security. Ask for the vendor's security posture. Neon Blue has completed a SOC 2 Type II audit.

What does this look like in practice?

Hypothetical: a startup's product writes ad variants for small online brands. It wants to show each user the two variants most likely to land with that brand's buyers.

Today: the team runs one of its real use cases as a study. Audience: shoppers who bought from small online skincare brands in the last six months. Question: rank five ad variants and say why.
Read: whether the rankings differ by segment and whether the reasons are specific enough to act on.
Decision: if the read is useful, the team knows the audience spec and question design it would want to call from its product once the API is available.

Want this run on your own question? Neon Blue's first study is free: start a free first study.

What are the limits?

A simulated audience answers questions about people. It won't tell you the click-through rate a live campaign will get.
Audiences and situations unlike anything in the grounding research are harder to read and should be weighted lightly.
For a launch or spend decision that matters, pair the simulated read with a live test on real people.

Questions about the Neon Blue API

Is the Neon Blue API live?

Not yet. The simulation and creative pre-testing API is in build. Results come as a report today.

What is creative pre-testing?

Showing an ad, headline or message to a defined audience before it runs, to see which version lands, with whom and why.

How is this different from asking a general AI model to play a persona?

A persona prompt gives one average voice. A simulated audience is built from real people's interview and survey data, so answers vary by segment the way real answers do.

Can I use Neon Blue before the API is available?

Yes. Run the question you would automate as a study. The first study is free.

Is Neon Blue SOC 2 Type II?

Neon Blue has completed a SOC 2 Type II audit.

How do I start a free first study?

Pick the one question your product would ask a simulated audience most often, and write down the audience you would ask. Neon Blue's first study is free. Start a free first study