Surveys Move Inside AI Assistants: What It Means for Research Agencies

Illustration of an AI chat window containing a survey form with checkboxes, next to a bar chart

Contents

Key points
What was launched
What it changes
What AI assistants do not solve
What research agencies can do now
For research buyers: DIY or agency?
Frequently asked questions
Sources

Key points

Since April 2026, SurveyMonkey users can create, send and analyse surveys from Claude, and since August 2026 from ChatGPT ^{1, 2, 3}.
Sample providers follow: PureSpectrum lets researchers launch surveys and manage quotas from Claude (beta, June 2026), and Cint is testing a connection between AI agents and its marketplace ^{4, 5}.
Google Forms and Microsoft Forms now draft surveys from a short description and summarise the answers ^{6, 7}.
Building a survey becomes almost instant. What still takes expertise is who answers, how good the data is and what it means.
For agencies, the risk is losing simple studies to DIY. The opportunity is to sell what AI assistants cannot guarantee: sampling, data quality, local fieldwork and interpretation.

What was launched

Survey platforms. At the end of April 2026, SurveyMonkey released a connector for Anthropic’s Claude: researchers can ask Claude to build a survey from a brief or a file, send it and analyse the data without leaving the chat ^{1, 2}. In August 2026, it added the same for ChatGPT ²: create and edit surveys, generate a link and summarise recent results such as NPS scores and open-ended themes ³.

Sample providers. In June 2026, PureSpectrum launched in beta a “Claude Connector” that lets clients run feasibility checks, launch surveys, manage quotas and monitor survey health inside Claude. “Researchers describe what they need and we handle the rest,” said its chief product officer ⁴. In September 2026, Cint said it is testing a similar connection that would let AI agents define audiences in plain language, check feasibility across markets, get pricing guidance and launch studies; the survey platform Potloc already used it to set up a 20-market consumer banking study in a single automated run ⁵.

Office tools. Gemini in Google Forms builds a form from a prompt and summarises open-ended answers ⁶, and Copilot in Microsoft Forms drafts surveys with 5 to 15 typical questions, helps send reminders and summarises results ⁷.

Surveys answered inside the assistant. In October 2026, Anthropic added in-app surveys to Claude Enterprise: a short card appears after a reply in Claude. These surveys are for companies to ask their own staff how Claude helps their work, not for market research ⁸, but they show that the AI assistant is becoming a place where questions are both asked and answered.

Timeline: April 2026 SurveyMonkey connector for Claude; June 2026 PureSpectrum Claude Connector (beta); August 2026 SurveyMonkey in ChatGPT; September 2026 Cint tests a connection for AI agents; October 2026 in-app surveys in Claude Enterprise — Surveys and sample inside AI assistants, 2026. Sources: SurveyMonkey, Research Live, DRNO, Anthropic.

Most of these connections use the Model Context Protocol (MCP), an open standard that lets an AI assistant use another platform’s functions ^{1, 4, 5}.

What it changes

Speed. A first questionnaire, a feasibility check and a launch can now happen in one conversation, in minutes rather than days.
More do-it-yourself. Marketing, HR and product teams that never briefed an agency can now run their own surveys from tools they already pay for ^{6, 7}.
Pressure on simple studies. Quick polls, satisfaction surveys and concept checks with an online panel are the easiest to automate.
Growth on the software side. Research software is already the fastest-growing part of the insights industry, up 11.5% in 2024 according to ESOMAR ⁹.

What AI assistants do not solve

Who answers. An assistant can launch a survey, but the result is only as good as the sample: the right people, in the right proportions, reached in the right way. Many audiences, such as rural consumers, older people, doctors or senior B2B decision makers, are not reachable well enough through online panels alone.

Data quality. AI also makes fake respondents easier: a Dartmouth study published in PNAS in 2025 built an AI agent that passed survey checks 99.8% of the time ¹⁰. Someone has to check who really answered.

Questionnaire bias. A fast draft is not a good questionnaire: leading questions, missing answer options or the wrong order can bias the results. The MRS reminds researchers that “AI can be wrong, so you should always validate outputs” ¹¹.

Meaning and decisions. Summaries of open answers are useful, but deciding what the numbers mean for a brand, a price or a launch still requires market knowledge.

Rules. Under the 2025 ICC/ESOMAR Code, researchers must tell clients when AI is used and how much human oversight there is ¹²; personal data collected in a survey remains subject to data protection law.

What research agencies can do now

Use the same tools. Draft questionnaires, run feasibility checks and code open answers with AI, and pass the time saved on to faster delivery.
Sell quality control. Offer to review, clean and validate surveys that clients run themselves: sample checks, fraud detection, weighting and a short expert reading.
Focus on what is hard to automate: hard-to-reach audiences, face-to-face and telephone fieldwork, qualitative research, multi-country coordination and local knowledge.
Be clear about your AI use in proposals, as the ICC/ESOMAR Code requires ¹².
Stay visible where buyers look, including AI assistants: keep a complete, factual profile. See our article on how buyers find a research agency in the age of AI search.

For research buyers: DIY or agency?

A survey in your AI assistant is often enough for	Work with an agency for
Internal or customer feedback with your own contacts	Representative samples of a population or a market
Quick checks before a meeting	Decisions with a budget at stake: pricing, launch, segmentation
One country, online audience	Several countries, offline or hard-to-reach audiences
Simple questions, short survey	Qualitative research, complex designs, tracking over time

To compare agencies, you can send one request to up to 5 companies from any country or city page of this directory.

Frequently asked questions

Can I run a survey directly from ChatGPT or Claude?

Yes, with a connected survey platform. SurveyMonkey offers connectors for Claude and ChatGPT that create, send and summarise surveys, and Google Forms and Microsoft Forms have built-in AI assistants.

Will AI assistants replace market research agencies?

They automate simple surveys, but not sampling of hard-to-reach audiences, data quality control, offline fieldwork or the interpretation of results. Agencies that use AI themselves and focus on these tasks remain essential.

What is MCP in market research?

The Model Context Protocol is an open standard that lets an AI assistant use another platform’s functions. Survey and sample providers such as SurveyMonkey, PureSpectrum and Cint use it to let users launch surveys from an AI chat.

Are Claude Enterprise in-app surveys a market research tool?

No. They let companies ask their own employees how Claude helps their work. They show, however, that AI assistants are becoming a place where surveys are created and answered.

Surveys move inside AI assistants: what it means for research agencies