Back to Blog
Attribution & Analytics

How to Track App Installs That Start in ChatGPT (and Why Most of Them Look Organic)

ChatGPT referrals grew 303% in a year but are still about 0.33% of measured traffic, and Similarweb says only 8.8% of AI-influenced visits arrive as a direct referral. Where AI-driven installs hide in your data, and five ways to measure them.

October 17, 20269 min read
Last updated: October 17, 2026

The short version

AI assistants now influence which app people install, but very little of that influence shows up as a referral. The best recent data, from Demandbase Labs as reported by PPC Land and summarized by Relevant Audience (August 17, 2026):

  • Monthly ChatGPT referral visits to B2B sites went from about 645,000 in June 2025 to 2.6 million in June 2026, up 303%. The dataset covered 1,584 platform instances and over 11 billion website visits.
  • Even so, ChatGPT was only about 0.33% of measured traffic, roughly 190 times less than Google.
  • Similarweb found only 8.8% of AI-influenced visits arrive as a direct referral, while roughly 56% arrive through branded search.

That last line is the one that matters for apps. People hear the app's name in a chat, then go search for it. In your dashboards, the install looks organic.

Where AI-driven installs hide

PathWhat your data showsHow to catch it
Assistant links your website, user clicksWeb referral from chatgpt.com, perplexity.ai and similarWeb analytics channel group for AI referrers
Assistant names the app, user searches the storeOrganic store search installBranded search trend; onboarding survey
Assistant cites a creator's video, user clicks the creator's linkAttributed creator installDeterministic link attribution
Assistant links the bare store pageStore "web referrer" at best, often organicNot much; make sure your own links are the ones cited
Agent installs or opens on the user's behalfWhatever URL the agent was handedSame as above

The pattern: you can only attribute what passes through a link you control. Everything else needs indirect measurement.

Five ways to measure it

1. Give AI referrers their own channel on the web

If your app has a website, create a custom channel group in your analytics that matches referrers like chatgpt.com, perplexity.ai, gemini.google.com, claude.ai and copilot.microsoft.com. Many assistant links also carry a UTM source such as utm_source=chatgpt.com, so match on that too. This captures the 8.8% that click through, and it is the cheapest signal to set up.

2. Watch branded search as the AI signal

Since most AI influence arrives as branded search, your app name's search volume is a proxy for AI mentions. Track impressions for your app name in App Store Connect's search terms and Google Play Console's acquisition reports, and your brand query in Google Search Console. When branded search rises without a matching campaign, check whether assistants started recommending you; see how AI assistants recommend apps for the monthly prompt check.

3. Ask, in one tap

Add a single onboarding question: "How did you hear about us?" with options including "ChatGPT or another AI assistant," "a creator's video," "App Store search" and "a friend." Self-reported data is imperfect, but it is the only way to see the path from a chat to a store search. Keep it to one tap and skippable.

4. Make the cited sources carry your links

Assistants cite editorial roundups and creator videos heavily. When those sources carry a tracking link, the AI-driven click becomes measurable. On IdeaEquity, every creator gets a unique link. On Android the click ID passes through the Play Install Referrer and the install is attributed deterministically. iOS passes no install referrer to anyone, so for iOS campaigns IdeaEquity works with AppsFlyer OneLink, which carries the creator's ID through the store. When an assistant cites a creator's review and a user taps the creator's link, the creator is credited and you pay for a verified install, not a guess.

5. Use deep links for the web-to-app hop

If assistants send people to your website, a smart banner or deep link that passes a token into the app keeps the chain intact. For web apps, IdeaEquity's ie_click parameter does the same job: the click ID lands on your site and the web snippet attributes the signup or purchase to the creator who sent it.

What not to conclude

  • "AI traffic is tiny, so ignore it." Direct referrals are tiny. Influence is not. If 56% of AI-influenced visits arrive through branded search, the referral number undercounts by a wide margin.
  • "Branded search went up, so it was AI." Maybe. It might also be a creator video, press or a feature. Use the survey to separate them.
  • "We can attribute all of it." You cannot. Aim to make the measurable share as large as possible by putting your links in the sources AI cites, and estimate the rest.

The takeaway

AI assistants are a recommendation layer, not a traffic source. Most of what they drive arrives looking like organic search. Measure the clicks you can, treat branded search and a one-question survey as your AI signal, and make sure the creators and articles that assistants cite are carrying links you can attribute.

Want the cited sources to carry your links? Start a creator campaign on IdeaEquity. Every creator gets a tracked link, and you pay per verified install.

Sources: Relevant Audience, August 17, 2026, summarizing PPC Land's August 16, 2026 report on Demandbase Labs data and citing Similarweb.

Frequently Asked Questions


Ready to grow your app?

Join IdeaEquity and connect with verified influencers who can drive real downloads for your app.

Make a tool for app developers or creators? Sponsor the IdeaEquity blog