Short answer
Visits from AI assistants arrive with a referrer such as chatgpt.com, perplexity.ai or claude.ai, and ChatGPT search links also carry utm_source=chatgpt.com. DataBull files all of them under an AI channel and connects each visitor to their Stripe, Polar, Dodo or Creem payments, so you see revenue and lifetime value per assistant. Links opened from apps without a referrer still show as Direct.
An AI channel, next to Search and Social
Every visit in DataBull gets a channel: Organic Search, Organic Social, AI, Referral, Email, Paid, Affiliate or Direct. A visit lands in AI when its referrer or its utm_source is one of these assistants:
| Assistant | Recognized as |
|---|---|
chatgpt.com (old chat.openai.com links are folded in) | |
claude.ai | |
perplexity.ai | |
| Gemini | gemini.google.com (and the old bard.google.com) |
| Microsoft Copilot | copilot.microsoft.com |
| Others | you.com, phind.com, chat.deepseek.com, deepseek.com, grok.com, meta.ai, chat.mistral.ai, poe.com, kimi.com, notebooklm.google.com |
Tagged links count too: ?utm_source=chatgpt, ?utm_source=perplexity or ?ref=claude on a link you share map to the same sources. Google's AI Overviews and AI Mode send visitors from google.com, so they stay in Organic Search, which is also how Google Analytics files them.
What each assistant actually sends
An AI visit is only visible if the click carries something that names the assistant. Here is what you can count on:
ChatGPT adds
utm_source=chatgpt.comto links in its answers, according to OpenAI's publisher FAQ, and the web app usually passeschatgpt.comas the referrer as well. Either one is enough. The full walkthrough is in how to trackChatGPT traffic.
Perplexity,
Claude, Gemini and Copilot show up through the referrer of their web apps. See tracking Perplexity traffic.
- Mobile apps and copy-paste are where it breaks. A link opened from an assistant's app can arrive with no referrer, and a URL someone copies out of an answer into a browser never has one. Those visits look like any other unlabeled visit.
DataBull doesn't pretend to know more than the browser tells it. A new visitor who lands on a deep page with no referrer goes to Direct · shared link, not into one mystery bucket, so you can at least see how much unlabeled traffic arrives on the pages assistants tend to cite.
Revenue per assistant, not just visits
Visits from AI answers are often small in number and high in intent: someone asked for a tool like yours and clicked the one that was recommended. Whether that's true for your site is the thing worth measuring. Connect Stripe,
Polar,
Dodo Payments or
Creem and every payment is credited to the visit that brought the buyer, so the AI channel shows:
- Revenue and sales per assistant, with conversion rate and revenue per visitor next to Search and Social.
- Lifetime value by first source: renewals and upgrades stay with the source that first brought the customer, so a
ChatGPT referral that subscribes for a year counts as a year.
- Path to purchase: buyers often find you in one place and buy after a later visit. You see the source that found them next to the one credited with the sale, and the attribution model (last non-direct, first touch or last touch) is your choice per site.
- A
Telegram message for every sale, naming its source, so the day
ChatGPT starts recommending you, you notice.
Click the AI row on the dashboard and every report filters to it: the pages AI visitors land on, the countries they come from, what they read before paying.
Crawlers are not visitors
GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot,
Claude-User, PerplexityBot, Perplexity-User and other AI fetchers read your pages, but they aren't people. DataBull drops them before anything is stored, together with headless browsers and data-center traffic, so the AI channel counts humans who clicked through.
Getting mentioned by assistants in the first place
No file or tag guarantees a citation, and anyone who promises one is guessing. What helps, as far as anyone can tell:
- Let the search crawlers in. OpenAI says blocking OAI-SearchBot keeps a site out of
ChatGPT search answers; PerplexityBot does the same job for
Perplexity. Check your robots.txt and any firewall rule that blocks bots.
- Answer the question on the page. A short, direct answer near the top of a page that matches what people ask is easy to quote.
- Be mentioned elsewhere. Assistants lean on what other sites, forums and comparisons say about you.
- Publish an llms.txt. A plain summary of your product and its key pages for AI tools. It's a young convention with no promise attached, but it's cheap. Ours is here.
Then measure. If an assistant starts sending buyers, the AI channel shows it within a day.
Ask an assistant about your AI traffic
A nice loop: DataBull's built-in MCP server lets Claude Code,
Cursor or
Codex read your numbers, so you can ask "how much revenue came from the AI channel this month, by assistant?" and get the answer from your own data. See asking
Claude about your analytics.
Ask your AI agent, too
Every DataBull install has a built-in MCP server. One command connects Claude Code, Cursor, Codex, Windsurf or VS Code, and your agent answers "which post made me money?" from your own data. The same numbers are in your terminal and behind read-only API keys.
npx databull@latest mcp
Checked in October 2026 against: OpenAI: Publishers and developers FAQ, OpenAI: crawlers and user agents, Perplexity: crawlers, Google Analytics: traffic-source dimensions. Spotted something out of date? Tell us.