PostHog AI: Ask Your Product Data Questions in Plain English

Most product questions are variations on five: where do visitors come from, how many sign up, how many activate, how many stay, and how many pay. The work is defining those events once.

By Ed · Published October 25, 2026 · 8 min read

Twenty-five years starting companies, consulting, owning a multi-million-dollar P&L and leading business intelligence for large enterprises.

To ask PostHog product questions in plain English, first fix three event definitions: what counts as a signup, what counts as activation, and what counts as a paid conversion. Then the questions that matter are short: which channels bring sessions and signups, what share of a signup cohort activated within 7 days, what share is still active in week 4, and which pages bring visitors who sign up. Quiriz's PostHog connector, in beta, answers these from aggregated data and shows the dataset and filters it used. This post explains the setup, the questions, and the cases where asking is the wrong tool.

Define the three events first

Any question about conversion is only as good as its definitions, and in product analytics the definitions are usually informal. One person's signup is an account created, another's is a confirmed email. Fix them once:

Signup: the event that means a new person exists. Pick the one fired exactly once per person, not on every login.
Activation: the first action that signals real use. It should be early and observable, such as creating a first project or running a first query, not something only your best customers do.
Paid: the event that marks a payment or an upgrade to a plan.

With these three, almost every funnel and cohort question reduces to counting and dividing. Without them, two analysts will compute two answers from the same data. The why your team gets different numbers post is about exactly this problem.

The questions worth asking

Group them by stage. Illustrative questions, in the language you would type:

Reach. How many sessions and page views did we get this week compared with last week? Which channels and referring domains drive them?
Signup. Did our launch bring signups or just visits? Which channel brings the most signups, and which converts best?
Activation. What percent of last month's signups activated within 7 days?
Retention. What is our week 4 retention for each signup week?
Entry pages. Which pages bring visitors who sign up?

Two of these reward a second thought. The channel with the most signups is not always the channel that converts best: a channel with 1,000 sessions and 50 signups (5 percent) beats one with 10,000 sessions and 200 signups (2 percent) on rate, and loses on volume. Ask for both. And launch questions need a time axis, so ask by day, not by month.

Why cohorts need care

Activation and retention are measured per signup cohort, and a cohort can only be judged once enough time has passed. A cohort that signed up three days ago cannot have a 7-day activation rate or a week 4 retention. If you divide anyway, the newest cohorts drag the average down and the trend looks like a decline that is not happening.

The rule is to count only cohorts that have been observed long enough: at least one full week for 7-day activation and week 1 retention, at least four full weeks for week 4 retention. Quiriz's metrics apply this for you: PostHog 7 day activation rate divides activated people by cohort signups from cohorts at least one full week old, and PostHog week 4 retention uses only cohorts at least four weeks old.

How Quiriz reads PostHog

Quiriz is an AI business analyst. The PostHog connector is in beta, read-only, refreshes daily, and is included in paid plans. You sign in, choose a project (US or EU cloud), and the first import guesses the signup event; you confirm or change the signup, activation and paid events, plus up to five more key events, in the Events panel under Connectors, PostHog. Quiriz then reads aggregates from HogQL queries over the last 13 months. It does not copy events, people or identifiers.

The datasets are [PostHog] Traffic by day (sessions, page views, bounces, session time and goal counts by channel, referring domain and UTM values), Pages by month, Audience by month (country and device), Active users, Signup cohorts, Events by day and Event users by month. The metrics are named and defined once: PostHog sessions, page views, bounces, signups, bounce rate, average session duration, signup rate, cohort signups, 7 day activation rate, week 1 retention, week 4 retention and paid conversion rate. When an admin sets a definition in Company Context, every person and every surface, web app, Slack or Office add-in, gets the same number.

The sessions metric matches PostHog Web analytics, so you can check the first answer against what you already see. For the visual side, see building a PostHog dashboard beyond built-in insights.

When asking is not the right tool

PostHog is built for exploring product behavior, and its own insights, funnels and recordings are better for some jobs. Asking in plain English is good for fixed, repeated questions and for questions that mix definitions, such as a rate across channels. It is not a substitute for:

Step-by-step funnels, session recordings, experiments or feature flags, which are not part of what the connector reads.
Per-person investigation. Because no person-level data is copied, you cannot ask who a specific user was.
Anything needing an event that is not in the top 200 per day or month, which is as far as the events datasets go.

Limits

A 13-month window, re-read on every refresh, and signup cohorts cover the last 26 weeks.
Referring domains and UTM values keep the top 100 per month; the rest are grouped as other. Pages keep the top 5,000 per month by page views.
Active users are not additive. A returning person counts once in each period, so weekly and monthly counts must not be added together. Unique visitors are not in the traffic data for the same reason.
Cookieless projects are approximate. If your PostHog project uses cookieless tracking, user and cohort counts are approximate and weekly or monthly counts overstate.
Read budget. If PostHog's hourly read budget is used up, the refresh waits and tries again next time.
Zero signups until an event is chosen. The signup column is 0 until a signup event is selected.

Ask your product data, not your analyst

Connect PostHog, confirm your signup, activation and paid events, and ask about channels, activation and retention in plain English. Free to start.

Try Quiriz free →

Frequently asked questions

Can I ask PostHog questions in plain English?

You can connect PostHog to Quiriz, which reads your project's traffic, signup, activation, retention and feature-usage data as aggregates and answers questions about it in plain English, with the dataset and filters shown.

How do I measure activation in PostHog?

Pick the event that best marks a person getting first value, then measure the share of each signup cohort that did it within a fixed window such as 7 days. Use only cohorts old enough to have been observed for the full window, or the newest cohorts will look worse than they are.

How do I calculate retention from PostHog data?

Group people by the week they signed up, then count how many were active in week 1, week 4 and so on after that week. Divide by the cohort size, and exclude cohorts that have not yet had that many weeks.

Why are my PostHog signups zero in a report?

Signups count the event you chose as the signup event. If none is chosen, the signup column is zero. Pick the event once in the connector settings and the numbers fill in on the next refresh.

Does asking questions of PostHog data change anything in PostHog?

No. The connection is read-only. Quiriz reads aggregated counts, and does not copy individual events, people or identifiers.