Ask Your Website Data Questions in Plain English

Picture the meeting. Someone asks which email campaign actually brought in new members last quarter, and every head turns toward the one person who knows how to build the report. That person is busy, so the question dies quietly and the decision gets made on vibes. Looker Studio now has a feature aimed at exactly that moment: Conversational Analytics, a chat box where you type the question in ordinary words and get a chart back, powered by Gemini. Google describes it as a way for people with no business-intelligence background to ask data questions in natural language and move beyond fixed dashboards, and the part that matters for a small team is that the current version is available to all Looker Studio users, with no paid subscription required for the chat itself.

The catch nobody puts in the demo

Here is what the chat cannot do: it cannot read the GA4 dashboard you just built. Google’s documentation is explicit that the current version chats with data agents built on BigQuery tables, and that those agents are created in BigQuery and shared into Looker Studio rather than built inside it. So the road to the chat box runs through BigQuery. Your GA4 data gets exported there, a data agent gets built on those tables, and then you ask it questions. If nobody tells you this, you will open the chat, find nothing to talk to, and conclude the feature is broken. It is not broken. It is pointed at a different pipe than the one your dashboard uses.

This is the whole article in one paragraph, really. Everything below is the practical version: what to ask first, what the setup takes, and where it goes wrong.

Your first five chat prompts

The reusable piece. Take the five questions from our GA4 dashboard tutorial, which your team already agreed on, and ask them in the meeting instead of building them in advance. One table, five rows. The third column is the habit that makes the chat trustworthy: every answer gets checked before it gets repeated.

The question your team asks	What to type into the chat	What to check in the answer
Where do new visitors come from?	Which channels brought the most new visitors last month?	That “new visitors” means first-time sessions, and the range covers a full month.
Which pages do members actually read?	What were the ten most-viewed pages last quarter, excluding the homepage?	That staff and bot traffic is filtered the same way your dashboard filters it.
Which pages lead to membership joins?	Which pages did visitors view in the session before completing a membership join last month?	That a join means your membership_join key event, not any page with “join” in the title.
Which pages lead to event registrations?	Which traffic sources brought the most event registrations in the last 90 days?	That the registration counted is your event_registration key event.
Which pages lead to newsletter signups?	Which pages had the highest newsletter signup rate last month?	That the rate divides signups by that page’s views, not by all site visitors.

What the setup actually takes

Three things have to happen, and the honest version is that the middle one needs someone comfortable with Google Cloud. For many associations that means a contractor hour or the most technical person on staff. Budget for that instead of discovering it mid-project.

First, the export. In GA4, under Admin and then Product links, the BigQuery Links page connects a Google Cloud project and switches on the daily export, with the data starting to flow within about a day of linking. Google requires a valid payment method on the Cloud project for the export to run at all, and its export guide says so plainly. The export itself carries no Analytics fee; BigQuery bills for storage and queries, and its pricing page puts the free tier at the first 10 GB of storage and the first 1 TB of queries each month, which covers a typical association site’s export with room to spare. Two warnings from the same guide: the export is not retroactive, so link it before you need the history, and standard properties are capped at one million events per day, with the export pausing if you consistently exceed that.

Second, the data agent. In BigQuery, someone builds a data agent on the exported tables and shares it to Looker Studio. This is the step Google says cannot be done inside Looker Studio, and it is the one most likely to need outside help. It is a one-time build, and then it sits there answering questions.

Third, the chat. Once the agent is shared, you open a new conversation in Looker Studio and type your first question from the table above. From here on, anyone on staff can ask. The technical work is front-loaded; the asking is not.

The answer that sounds right and is not

Google’s own guidance for its AI features says the technology “can generate output that’s plausible-sounding but factually incorrect” and recommends that you validate all output before you use it. In practice the failure is rarely a wrong number. It is a confident answer to a slightly different question than the one you meant. “Top pages” might come back ranked by views when you meant ranked by joins. When an answer surprises you, ask the follow-up in plainer words, adding the definition you meant, and compare it against the dashboard tile that answers the same question. Treat the chat as a fast second opinion, never as the source of truth. That is what the third column in the table is for.

Your questions are not training data

For the privacy-minded reader, one reassurance straight from Google’s data-governance page: Gemini does not use your prompts or its responses to train its models. Your conversation history lives in your account, and the data agent lives in your Cloud project, so the ownership story from the dashboard tutorial carries over intact. Nobody’s chat questions are building a model for anyone else, and nobody can take the agent away by removing a cookie.

Who this is for

If your team already runs the five-question dashboard and someone on staff, or a contractor, can handle the BigQuery half, the chat is the natural next step: the same five questions, asked live in the meeting instead of tiled in advance. If nobody can do the setup yet, skip it without guilt. The dashboard answers the same questions, and the export you link today becomes the history the chat needs tomorrow. Either way, start with the GA4 dashboard tutorial, link the BigQuery export while you are in Admin anyway, and come back to the chat when you are ready. The questions will keep.