On this page
- Three evidence lanes, and why Hub users can't be identified
- Define the outcomes before you instrument: qualified sign-up, activation, first paid invoice, net vs gross
- Capture ChatGPT attribution signals and keep the raw evidence
- Can I share or publish my Business Hub numbers?
- Join website events to accounts and paid outcomes without a user-level Hub join
- Read the results: what you can and can't conclude
- Pick next week's experiment and review it
- Run one approved experiment this week, and review it on paid outcomes
- More ChatGPT Business Hub guides
- Related Business Hub guides
To turn ChatGPT Business Hub data into revenue decisions, keep three kinds of evidence apart (Hub activity, synthetic prompt checks and your own website sessions), join only the website sessions to sign-ups and first paid invoices, then pick one experiment for next week and review it against the same outcomes. The Hub tells you that ChatGPT mentioned and linked to your site. Your own analytics tell you what those visitors did next. The mistake is blending the two into one number.
This playbook is for founders and growth owners at funded B2B software companies who have claimed a ChatGPT Business Hub listing, or plan to. It covers ChatGPT Business Hub revenue attribution end to end: what each data source can and cannot say, how to define outcomes before you instrument anything, how to track chatgpt.com visits in GA4 or any analytics tool, how to join sessions to paid outcomes, and how to read the result without overclaiming. Facts about the Hub are as checked on 18 November 2026. The product is in beta, so recheck before you build on any detail.
Three evidence lanes, and why Hub users can't be identified
ChatGPT Business Hub is the area in OpenAI's "Businesses in ChatGPT" beta where eligible businesses can claim and verify a listing, update business details and view aggregated analytics: mentions, website clicks and common discovery topics where their site is cited (OpenAI, 2026). The insights do not reveal individual conversations, and they appear only when interaction volumes meet privacy thresholds. Reporting uses a selectable date range, including presets such as the last 7, 14, or 30 days, and custom ranges. A blank metric can mean activity is below the threshold, data is still being collected or reporting is unavailable. It doesn't necessarily mean zero.
That design sets the hard boundary for this whole exercise: you cannot identify which Hub user became which sign-up. Don't try to infer prompt identity from a click, and don't build a user-level join. Run three lanes and never merge their rows.
| Lane | What it is | Question it answers | What it cannot answer |
|---|---|---|---|
| 1. Hub activity | Aggregated real activity reported inside the Hub: mentions, website clicks, discovery topics | Is ChatGPT surfacing us, and for what topics? | Who clicked, what they asked, whether they paid |
| 2. Synthetic prompt checks | Prompts you choose and run on a schedule, with answers saved | Do answers name us for the prompts we care about, and what do they say? | How often real people ask those prompts |
| 3. Website sessions | Your own analytics and product events | Which visits became sign-ups, activation and paid invoices? | What ChatGPT showed before the click |
Each lane has a different population. Hub numbers cover real activity above a privacy threshold. Synthetic checks are a sample you designed. Website sessions are only the visitors your tracking can attribute. Adding them together, or dividing one by another (clicks over mentions, say), produces a figure with no stable meaning. Report each lane on its own line, and let only lane 3 touch revenue.
If you want the synthetic lane run for you, Revnu tracks how your business appears in ChatGPT, Claude, Perplexity and Gemini answers and compares them with competitors. Treat that as lane 2 evidence, kept separate from the Hub.
Define the outcomes before you instrument: qualified sign-up, activation, first paid invoice, net vs gross
Write the outcome definitions before you touch a tag. Every later argument about "does ChatGPT traffic convert" is really an argument about definitions. Use five, in this order.
Session. A visit your analytics attributes to ChatGPT using the rules in the next section.
Qualified sign-up. A sign-up that passes a rule you wrote down, such as a work email domain and a stated use case. Raw sign-ups include students, competitors and bots, so qualify them.
Activation. The first product action that predicts retention for your product. Pick one event and name it. If you have no evidence for which event predicts retention, choose the one closest to the value a customer pays for.
First paid invoice. A positive payment, not a trial start. Count the first invoice that was actually paid.
Net vs gross. Gross is what was invoiced. Net subtracts refunds and credits. Record currency on every row and don't mix currencies in one total.
Two rules keep the reporting honest. First, freeze the definitions for the review period. If you change what "qualified" means mid-month, last month's rate is no longer comparable. Second, report the denominator next to every rate. "Three paid customers from ChatGPT sessions" and "three paid customers from forty ChatGPT sessions" are different statements, and a small denominator should be labelled as one.
This is also where you decide what a good week looks like. A B2B sales cycle may mean first paid invoices lag the first session by weeks. If so, review sign-ups and activation weekly and paid outcomes on a longer cohort window, and say so in advance.
Capture ChatGPT attribution signals and keep the raw evidence
To track ChatGPT traffic in GA4, open Reports, then Acquisition, then Traffic acquisition, set the primary dimension to Session source / medium and filter for chatgpt.com. Add Landing page + query string as a secondary dimension and look for utm_source=chatgpt.com, which ChatGPT appends to many outbound links (Google Analytics Help, 2026). Mark sign-up, activation and payment as key events, then compare Key events and Session key event rate for ChatGPT against your other sources. For a reusable view, build an Exploration filtered to Session source matching chatgpt.com.
Two cautions make this lane messier than it looks.
Classification is inconsistent. GA4 may group a visit under its AI Assistant channel when it recognises the referrer, but source-only tagged visits can instead land in Referral or Unassigned. Check both Session default channel group and Session source / medium. If you want one consistent grouping, create a custom channel rule for the AI sources you care about, such as chatgpt.com, perplexity.ai and claude.ai. That also answers the common request for a GA4 segment covering ChatGPT, Perplexity and Claude.
Some visits are invisible. Clicks that arrive without a referrer or UTM parameter may be counted as Direct, so ChatGPT-attributed traffic is a floor, not the whole picture. Don't multiply it by a capture factor you can't defend.
Keep the raw evidence, not only the report. For every session you attribute, store the first-touch source, last-touch source, landing path, timestamp and whether analytics consent was given. Store it in your own warehouse or product database, not just in a dashboard that may reclassify history. Then run a monthly QA: click a real ChatGPT-sourced link to your site, confirm the UTM survives any redirect, and confirm the landing path and consent state land in your records. Redirects that strip parameters are a common silent leak.
Keep lane 1 and lane 2 as dated snapshots too. There is no documented Hub export, so copy the figures by hand into a private log each week, with the date range and any "data through" label, and save your synthetic prompt answers verbatim with the date run. Insights are a rolling window, so an unsaved week is gone. Keep that log internal; the section below explains why.
Can I share or publish my Business Hub numbers?
No, not publicly, unless OpenAI has given you written permission. Section 6 of the Businesses in ChatGPT Terms makes Hub insight data OpenAI's Confidential Information. You may use it only to measure, analyse, optimise and report on your own use of Businesses in ChatGPT. Without OpenAI's written authorisation you may not disclose it, or anything derived from it, except to service providers acting for you under equivalent restrictions. You also may not commercially exploit it, use it to identify people or use it to build audiences.
In practice:
Public posts, threads and blog posts: no Hub figures without written permission. That covers raw numbers, percentage changes and charts worked out from them, because a derivative counts too.
Case studies and screenshots: no. A Hub screenshot is the data itself, and a case study quoting your mentions or clicks discloses it.
Client reports from an agency: allowed when the agency is your service provider, acting for you under restrictions equivalent to the terms. The agency still can't reuse your figures in its own marketing or show them to other clients.
Internal reporting: fine. Measuring and reporting on your own use is exactly what the data is for.
Screenshots you've seen online: vendor dashboards, guides and posts showing Hub numbers are not permission for you. Permission comes from OpenAI in writing, to your business.
The short version: Hub numbers are for running your business, not for marketing it. This summarises section 6 as published on 18 November 2026, so read the full terms before relying on it for a specific use.
Join website events to accounts and paid outcomes without a user-level Hub join
Join only lane 3 to revenue, and join it on identifiers you collected yourself. The chain is: attributed session, then anonymous visitor ID, then sign-up (the visitor ID is attached to the new account at registration), then activation event, then first paid invoice from your billing system.
In practice that means three tables you control.
Sessions: visitor ID, first-touch source, last-touch source, landing path, timestamp, consent flag.
Accounts: account ID, visitor ID captured at sign-up, sign-up timestamp, qualification result, activation timestamp.
Payments: account ID, first paid invoice date, gross amount, refunds, currency.
Join sessions to accounts on visitor ID, and accounts to payments on account ID. Nothing in this chain touches the Hub, and nothing needs to. The Hub tells you which topics and pages ChatGPT surfaces, which guides what you test. It is never a join key.
This is a reader-owned implementation. It can live in GA4 plus your product database, PostHog, a warehouse or a CRM. If you already store first-source and last-source on each account, the join is a query, not a project. Revnu does not offer a native analytics, CRM or Hub connector for this join, so plan to build it with the tools you already run.
Three guardrails:
Respect consent. Join only sessions where your own tracking was permitted, and drop the rest from the denominator rather than guessing.
Separate first and last touch. A ChatGPT visit that later converts through a direct return is a first-touch story. Report both views and don't pick the flattering one after seeing the numbers.
Never back-fill a Hub click to a person. If a record looks like a Hub click, it is a website session with a chatgpt.com source, nothing more.
Read the results: what you can and can't conclude
You can conclude what happened to the visitors you could attribute. You cannot conclude what ChatGPT caused. That distinction decides whether your next experiment is sound.
Here is an invented scenario to show the reading, with made-up numbers unrelated to any real account. Suppose a software company's ChatGPT-attributed sessions, last month, produced 60 sessions, 6 qualified sign-ups, 3 activations and 1 first paid invoice. Its organic search sessions produced 900 sessions, 36 qualified sign-ups, 14 activations and 5 first paid invoices.
| Source (invented) | Sessions | Qualified sign-ups | Activations | First paid invoices |
|---|---|---|---|---|
| ChatGPT-attributed | 60 | 6 | 3 | 1 |
| Organic search | 900 | 36 | 14 | 5 |
What you can say: the ChatGPT-attributed sessions produced a higher sign-up share (6 of 60 against 36 of 900), but on a tiny base, so the paid figure of one invoice is an anecdote, not a rate. What you can't say: that ChatGPT converts better than Google, that it caused those sign-ups, or that more mentions will produce more revenue. Studies comparing AI referral conversion with organic exist, but they describe other companies' traffic mixes. Your own table is the only evidence that applies to your product.
A short list of things to refuse to do:
No causal-uplift claims. You have no control group. A visitor may have seen your brand elsewhere first.
No click-through benchmark. Hub clicks divided by Hub mentions mixes a threshold-filtered aggregate with a different population. Treat it as a trend inside the Hub, never a conversion rate.
No inference from blanks. An empty Hub metric may mean below threshold.
No mixing of lanes on one chart axis. Show them side by side with labels.
When the denominator is small, say so and extend the window before drawing a conclusion. A rule of thumb that holds up: if a result would change under one customer more or less, it isn't a result yet.
Pick next week's experiment and review it
Choose one experiment per week, tie it to one lane-3 outcome, and write the review criteria before you ship. One experiment keeps the reading clean. With two changes you can't tell which moved the number.
Use the lanes to pick the target:
Hub discovery topics show a theme you don't own a page for. Experiment: publish or sharpen one landing page for that topic. Outcome to watch: qualified sign-ups from chatgpt.com sessions landing on it.
Synthetic checks show answers describing your product wrongly or naming a competitor for a prompt you care about. Experiment: fix the page the answer relies on, with the correct claim stated plainly near the top. Outcome to watch: the same prompt re-run next week, plus sessions to that page.
Sessions arrive but sign-ups lag. Experiment: change what ChatGPT visitors see first. They often arrive mid-question, so lead with the direct answer to the question the page is cited for, then one clear next step. Outcome to watch: sign-up share on that landing path.
On optimising a landing page for ChatGPT visitors, the evidence supports restraint. Match the page to the topic that sent them, put the answer and the sign-up step in the first screen, and keep a single call to action. Don't assume these visitors want a different product story. Test it.
Write the review as a three-line note before you start: the change, the one metric, the date you will look. At the review, record what moved, what stayed flat and whether the denominator was large enough to read. Then pick the next experiment, or repeat the same one with a longer window.
This is the loop Revnu is built to run. Revnu is an AI agent that runs growth experiments for software businesses, picks the next move, does the supported work such as SEO pages and AI-visibility tracking within your approvals, and checks back on the result. It does not connect to the Hub itself, and the revenue join above stays in your own analytics. For related reading, see AI in marketing use cases that work for small teams and AI crawler tracking without JavaScript.
Run one approved experiment this week, and review it on paid outcomes
Revenue decisions from ChatGPT come from your own sessions joined to your own invoices, with the Hub and synthetic checks guiding what to test. Freeze your outcome definitions, store raw attribution evidence, join only the website lane, and commit to one experiment with a written review date. To have that loop chosen, executed within your approvals and reported back in Slack each week, start with Revnu at revnu.com.
More ChatGPT Business Hub guides
Related Business Hub guides
Written by
Co-founder and CEO, Revnu
