GatherHQ launches on YC Fall 2026, built on 250,000 buyer interviews

A product manager who wants to know what buyers think of a new price has two bad options: run a six-week survey, or guess. GatherHQ sells a third one.

GatherHQ launched on Y Combinator's Fall 2026 batch on Sunday. The company sells an API that answers questions about buyers. It built the answers from 250,000 interviews it has run with real people, YC's launch page says. The YC launch page carries 2 votes at the time of writing.

What it does

The product is called Gather. Its conversational interface is named Emma. The YC page describes it as "an API for how buyers think". A sales or product team can ask a question in plain English. Gather answers from past interviews when it can. When it cannot, it runs a new short interview with the right kind of person. An AI agent can call the same thing through an API or an MCP endpoint, the company says.

The traction it claims

GatherHQ names 60 enterprise customers on the launch page. The company's own site lists more than 20 of them by name. The list includes Datadog, Fortinet, SailPoint, Envoy, Staples, Invoca, CloudBolt and Cover Genius. It also names Monotype, HiddenLayer, Synack and Belk. Three customer reports it links give sample sizes. Invoca's used 693 United States consumers. CloudBolt's used 302 IT decision-makers. Cover Genius's used 1,392 consumers across seven countries. The company says it has access to 65 million people for recruiting new interviews.

The founders are Mayank Mehta, Anand, and Jay. The YC page says they built PulseQ&A, which Gartner bought, and Cooliris, which Yahoo bought.

What is new about doing it with AI

Market research has always worked on a lag. A team drafts a survey, pays a panel, waits for the field, and reads the result weeks later. GatherHQ's claim is that AI can answer most questions from interviews it already has. It only runs a new study when it has to. The result it describes is quantitative and qualitative insights in days. The model updates when a new interview goes in.

A buyer should ask three questions before trusting it. First, what question went to what interviewee: a model trained on security buyers cannot answer for retail. Second, how old the answers are: buyer sentiment on a price moves inside weeks. Third, who recruited the panel, because a panel from one source has one bias. The company's own panel of 65 million is a self-reported figure. GatherHQ says the API also runs new short interviews when the stored answers do not cover the question. So the recency problem has at least a stated answer.

Reveneau's take: the shape fits where market research has been going. The practical question is not whether an API can simulate buyers. It is whether the buyers it has interviewed match the buyers a given team sells to.

Source

GatherHQ on Y Combinator, and the GatherHQ site.