The Q&A Is the Research:
Questions Are Market Data

Quick Answer:

Live audience questions are an unmediated voice-of-customer channel and a leading indicator of market confusion: people ask at the mic before they type into Google or ChatGPT. The working method is three steps — capture every question verbatim within minutes of leaving the stage, classify each one as confusion, objection, or use case, and convert the recurring patterns into FAQ content, keynote revisions, and research leads. The signal is repetition across independent rooms, not volume.

The dataset nobody logs

Marketing has an entire discipline for listening to customers. Voice of Customer (VoC) research — the formal process of collecting and analyzing what customers say they need and struggle with — runs on surveys, interviews, review mining, and support tickets. It is a mature field with a mature problem: almost every VoC channel is mediated. Someone designed the survey, chose the interview questions, or built the feedback form. The customer answers inside a frame you built.

A live Q&A has no frame. The person stands up, in front of peers, and phrases the problem exactly the way it exists in their head. Nobody suggested the wording. Nobody limited the options. That makes conference and webinar questions one of the rawest customer-language datasets available — and one that the VoC literature barely touches. Search behavior confirms how under-developed the method side is: in Semrush's US database as of July 2026, 880 people a month search “what is voice of customer” while just 40 a month search “how to collect voice of customer” — a 22-to-1 gap between wanting the definition and wanting the method.

This post is the method — at least the one I use. It requires no tooling beyond a notes file, and it scales down: everything here works whether you keynote 50 conferences a year or run one monthly webinar.

Questions lead, data lags

The core claim deserves to be stated precisely: audience questions are a leading indicator of market confusion. Keyword volumes, platform dashboards, and survey results are lagging indicators — by the time a confusion is measurable there, thousands of people already have it. The person at the mic is earlier in the curve. They are asking you precisely because the answer isn't findable yet in the places they already looked.

And the window for overhearing buyers is narrowing. As research moves inside AI assistants, the question record disappears from view: Pew Research Center's analysis of the browsing behavior of 900 US adults in March 2025 found that when a Google search triggers an AI Overview, users click a traditional result in only 8% of cases, versus 15% without one — and click a source link inside the AI summary in just 1% of visits. Per SparkToro's 2026 analysis, about 68% of US Google searches now end without any click at all. The buyer's question still exists — you just don't get to see it asked. A room with a microphone is one of the last places where you do.

Step 1 — Capture: the ten-minute rule

The method fails at step one more than anywhere else, because capture feels optional in the moment. It isn't. My rule: every question gets logged within about ten minutes of leaving the stage, verbatim, before memory rewrites the phrasing into what I wish they had asked. The phrasing is the data — a question remembered as “something about metrics” is worthless; the exact words “so are mentions and citations the same number?” are gold.

Three capture channels, in order of fidelity:

• The moderation tool. If the event runs Slido or an app with submitted questions, ask the organizer to export them. You get every question — including the ones the moderator never picked, which skew more honest.
• Your own log. One running file. Each entry: date, event type, audience profile (executive forum vs. practitioner conference), and the question verbatim.
• The hallway. The questions people won't ask at the mic — the “can I ask you something basic?” ones — are often the strongest confusion signals, precisely because the asker suspects everyone else already knows.

Step 2 — Classify: confusion, objection, use case

A pile of questions is an anecdote collection. Classification turns it into a dataset. Every question I log gets one of three tags:

• Confusion — two concepts blurring into one, or a term that hasn't landed. “Aren't mentions and citations the same thing?” The market's mental model is wrong or missing. Confusions are content opportunities: whoever explains the distinction clearly, first, owns it.
• Objection — a reason not to act, stated as a question. “We optimized for ChatGPT — why would we also need to worry about Google?” Objections tell you what belief is blocking adoption. They are the questions your sales team will hear next quarter.
• Use case — an application you didn't anticipate. “Could we use this to monitor how AI describes our CEO?” Use cases are product and research leads: the audience extending your framework into territory you haven't mapped.

The tag matters because each type converts differently — confusions become explanatory content, objections become the counter-argument slide in the next keynote, use cases become research questions. And the threshold for acting is repetition across independent rooms: one question is noise; the same underlying question from three unrelated audiences in a quarter is a pattern.

Step 3 — Convert: from log to output

Conversion is where the log pays rent. Three destinations:

Content, phrased the way buyers phrase it. Generative engines answer natural-language questions, which means a verbatim log of how real buyers phrase their questions is raw material most competitors don't have. Recurring questions become FAQ entries and question-shaped headings — the exact format that, as I argued in Your FAQ Page Is Your Most Underrated GEO Asset, generative engines retrieve and cite most readily. The FAQ at the bottom of this post is built from logged questions, not keyword tools.

The next keynote. My talk “Mentions vs Citations: The Two Metrics Every CMO Conflates” exists because the mentions-versus-citations confusion kept surfacing at the mic in room after room — the Q&A log flagged it as a pattern long before I would have picked it as a topic on instinct. The log is also how talks get retired: when a question stops appearing, the market has caught up and that section's job is done.

The research agenda. Recurring questions tell a research team where measurement is missing. The objection “why doesn't ChatGPT visibility transfer to Google?” is answerable today precisely because studies now measure the platforms separately — which is the kind of measurement audiences were asking for before it existed.

Worked example: what 2026 audiences actually ask

Here is the loop closing in practice. Across the first half of 2026, the single most repeated question class in my log — different phrasings, executive and practitioner rooms alike — has been the confusion between being mentioned by an AI assistant and being cited as a source, and its sibling objection: that optimizing for one AI platform covers all of them.

The audience's instinct that these are separate things is correct, and first-party data now quantifies it. Semrush's AI Visibility Index 2026 — a study of 126 million US AI-search prompts across 22 verticals, January to April 2026 — found the overlap between the brands an AI platform mentions and the sources it cites ranges from 64% on Google AI Overviews down to 30% on Gemini. Citation behavior diverges just as sharply: ChatGPT averages 15.4 citations per answer; Gemini averages 3.3. One platform's visibility does not transfer to another's — the objection at the mic was pointing at a real, measurable gap. I unpacked what that means for measurement in Why Citation Authority Matters More Than Rankings in AI Search and in AI Visibility: What It Is and How to Measure It.

That is the full circuit: question logged → pattern recognized → research consulted → keynote and content updated → and the next room's Q&A tests whether the explanation landed. The same survey work behind the Index also hints at why closing these confusions matters commercially: among 481 marketers surveyed, 81% of teams integrating SEO and AI-visibility work reported more AI-driven traffic or leads, versus 36% of teams keeping them siloed.

Start with your next talk (or webinar, or sales call)

You don't need 50 stages a year. The unit of signal is the independent room, and most marketers have more rooms than they think: monthly webinars, sales calls, community AMAs, panel appearances, even the questions under a LinkedIn post. If you speak — at any volume — the minimum viable version is one notes file and the ten-minute rule at your very next event. Tag each question as confusion, objection, or use case. Review the file quarterly. Act only on repetition.

And if you book speakers rather than being one: the Q&A quality is a selection signal. As I wrote in How to Pick an AI Keynote Speaker, a speaker whose material visibly updates from what audiences ask is a speaker whose talk was built from evidence rather than repetition. Ask a candidate what questions they're hearing lately — the ones with a real answer are running some version of this method.

The stage time is the visible part of the job. The dataset it generates is the quiet part — and in a market where buyer questions increasingly vanish into AI chats nobody can observe, the quiet part is getting more valuable every quarter.

Frequently Asked Questions

What is voice of customer research?

Voice of Customer (VoC) research is the formal process of collecting and analyzing what customers say they need, expect, and struggle with — typically through surveys, interviews, reviews, and support tickets. Its blind spot is mediation: most channels capture answers inside a frame the researcher built. Live audience questions are an unmediated VoC channel — the buyer phrases the problem in their own words, in public, before any survey reaches them.

How do you collect voice of customer data from live events?

Three channels: export submitted questions from moderation tools like Slido when the event uses one; keep your own running log, writing each question verbatim within about ten minutes of leaving the stage; and log hallway questions too — the ones people won't ask at the mic are often the most honest. Every entry carries the date, event type, and audience profile.

Are audience questions better than surveys?

Different, not better. Surveys quantify a known question; audience questions surface the unknown ones. A survey can tell you what share of CMOs are confused about AI-search metrics — but only after you knew to ask. A Q&A tells you the confusion exists, in the customer's own phrasing, months earlier. Use questions to discover what to measure, then surveys and platform data to size it.

How many questions do you need before patterns appear?

The signal is repetition across independent rooms, not raw volume. One question is an anecdote; the same underlying question from three unrelated audiences in a quarter is a pattern. Even four or five talks a year, or a monthly webinar, accumulates enough Q&A within two to three quarters to see repetition clearly.

How does audience Q&A connect to AI search and GEO?

Two ways. AI assistants answer natural-language questions, so a verbatim log of buyer phrasing is raw material for FAQ pages and question-shaped content that generative engines retrieve and cite. And as research moves inside AI chats where marketers can't observe it — Pew found only 8% of searches with an AI Overview produce a click on a traditional result — the live Q&A is one of the few remaining places to hear buyer questions verbatim.