AI Research

Run surveys on premium websites.

You pay to run a survey targeted at people visiting premium sites. You pay per completed response of three or more real answers. An AI moderator asks the questions, adapts the follow-ups, and hands you back sentiment, topics, and quotes.

generated imagery
Who answers

Readers of premium publications, reached by media targeting. These are the people you already buy media to reach, rather than a panel recruited to answer surveys for income.

What you pay

Per completed response. Three or more genuine answers make a complete. Impressions that don't complete cost nothing.

What you get

Structured results plus verbatims. Every response comes back read whole and structured, with every quote cited back to the conversation it came from.

How a survey runs

Five steps, start to answer.

01

You state the purpose

Plain language: "understand what's resonating about the 2026 upgrades, and what isn't." The Study Architect turns it into questions and focus areas with you, in one conversation.

02

You approve the questions

The opening question and its answer options are yours to approve before anything runs. Multiple variants can rotate.

03

The survey runs as media

It runs as a clearly labeled unit on premium sites, targeted by geography, device, operating system, and property type, bought as a direct campaign or a private deal. Readers answer in place, mid-article.

04

The AI moderator interviews

A vague answer gets a clarifying follow-up rather than being recorded as given. Only genuine, sufficient answers count, and three of them make the completed response you pay for. An optional extra question and open chat follow for people with more to say.

05

Results land as research

Every response is read whole, through the lens of the purpose you stated. Inspect charts it, Understand scores it, Evidence holds the verbatims, and Ask n+1 answers in plain language, live while the study runs.

Why not a panel?

Real visitors, not professional respondents.

Panelists answer surveys for a living. n+1 respondents are people who chose to be on a premium page, answering in context. Quality gates run turn by turn. Nonsense answers are never counted, abusive sessions are halted, and nobody who completed is re-asked for 90 days. The sample follows your media targeting, and we're explicit about what that means.

The methodology, limitations included
What it's for

Questions with deadlines.

  • Concept and message tests before the spend commits.
  • Launch reads while the launch is still steerable.
  • Perception and churn questions where the why matters more than the count.
Building brand instead?

An ad people can talk to.

Conversational Advertising is the open version of the same unit: branding that keeps the transcript.

See Conversational Advertising