Skip to content
Skip to content
APP MARKETING

Your mobile measurement partner can now count installs from AI answers. Trust the method first.

AppsFlyer says it can now attribute app installs and re-engagement to eight AI answer engines. It is a genuine step, and it is a vendor announcement, which is where a careful app team starts asking questions.

Summarise with
Set as preferred source on Google
14 min read
Quick answerAI traffic analytics for apps now has a vendor answer: AppsFlyer attributes installs and re-engagement from eight AI answer engines. Treat it as vendor measurement and ask how missing referrers are handled.
Section 01

What AppsFlyer announced, and what it did not

Engines in the new measurement
Grid of 21 engines: 8 lit for AI answer engines including ChatGPT, Claude, Perplexity, Copilot, Gemini, Grok, DeepSeek and Manus, and 13 unlit for traditional search enginesAI answer engines covered: 8traditional search engines covered: 138 AI answer engines covered13 traditional search engines covered
Grid of 21 engines: 8 lit for AI answer engines including ChatGPT, Claude, Perplexity, Copilot, Gemini, Grok, DeepSeek and Manus, and 13 unlit for traditional search engines
ItemValue
AI answer engines covered8
traditional search engines covered13
Of the 21 engines in AppsFlyer's expanded organic measurement, eight are AI answer engines and 13 are traditional search engines.

The fox checks the wind before the path, and the scent of a claim before the claim. According to Ad Tech Today, AppsFlyer "has expanded its organic search attribution to cover Answer Engine Optimization (AEO), giving marketers independent measurement of how customers discover and return to their apps through AI engines". The coverage lists ChatGPT, Claude, Perplexity, Copilot, Gemini, Grok, DeepSeek and Manus, alongside "13 traditional search engines such as Google, Bing, Baidu, Naver, and Yandex".

Two details matter for an app team. First, per Ad Tech Today, the measurement "covers re-engagement for customers who already have an advertiser's app, and install attribution for those who don't yet". So it is not only a top-of-funnel counter. Second, it follows an earlier step: in August, AppsFlyer said its Conversion API integration with ChatGPT Ads makes it "a measurement partner supporting advertising in ChatGPT", which covered paid campaigns. The new work adds the organic side.

Timing needs care. Mobile Marketing Reads and PocketGamer.biz both carried the expansion in mid-September, and Ad Tech Today ran it on 23 September. So this is a week-old vendor announcement being re-reported, not a brand-new incident. That is fine for planning, and it is a reminder that none of it comes from an independent evaluator.

The expansion is a claim about organic attribution, the unpaid half of the picture, and it lands in the fast-growing corner the industry calls AI traffic analytics. The word to hold on to is independent. The announcement describes the new capability as independent measurement. A mobile measurement partner sells attribution to advertisers, and it has a commercial interest in showing that its product measures the next channel. That does not make the number wrong. It does mean the method, not the marketing line, is what you should be reading.

Section 02

Why AI traffic analytics for apps is hard

AI traffic analytics starts from a simple problem: attribution needs a trail. A click on a paid ad carries parameters, a link on a page carries a referrer, and a store install can be tied back through a deferred link. An answer engine breaks two of those three habits. Seer Interactive's testing found that when ChatGPT links "don't include UTM parameters or referral data", the traffic "will likely appear as" direct in GA4, and it advises treating AI traffic numbers as "a directional floor rather than a precise count" (Seer Interactive).

Apps add a second hop. The person meets your answer in a chat window, taps a link, lands on a store page, installs, and opens the app. The chat is left behind at the store. Cloudflare's 2025 write-up on how it counts AI referrals says traffic referred by Claude's native app carries no Referer header and that the same likely holds for other native apps, so its calculations "may overstate the respective ratios" (Cloudflare Radar). If your funnel starts in a native app, your visibility into the referrer is thinner than a website's.

A fox looking through a telescope at distant lanterns, a picture of AI traffic analytics tracing an install back to its source
The install is the lit window. The source is the far hill.

Then there is the privacy layer on iOS, a thicket of its own. Apple says AdAttributionKit lets networks "attribute app installations while preserving user privacy", without the tracking prompt. It also says conversion-value and source-app-id "are conditional values that are only included in App AdAttributionKit postbacks when certain thresholds are met", and that advertisers can receive a postback "within 24-48 hours of a user launching the app". A competitor's glossary, Singular's page on AdAttributionKit, adds that the framework is "fully compatible with SKAdNetwork".

Put together, the honest reading is that AI traffic analytics for apps, and app attribution from any answer engine, will always be a mixture of observed and inferred, and the share that is inferred is the number that matters. Ask a vendor for it. A mobile measurement partner that can state it plainly is worth more than one that reports a large total.

Section 03

What independent data says about the audience behind the clicks

Independent traffic panels give a sense of scale for AI traffic analytics, the den every app team is now peering into. Similarweb's generative AI report says average monthly web visits across generative AI platforms worldwide "grew 70% year-over-year, reaching 9.5 billion between June 2025 and May 2026", and that ChatGPT's share of that traffic fell from about 76% to around 53%. That measures visits to AI platforms, not clicks out to your app, but it shows the audience is large and spreading across more than one assistant.

ChatGPT's share at the start and end of each window
ChatGPT's share at the start and end of each windowSlope chart: ChatGPT web visit share falls from 76 to 53 percent in Similarweb data, and prompt share falls from 70 to 50 percent in Comscore dataWindow startWindow endSimilarweb visits: 76 to 53Similarweb visits 76%53%Comscore prompts: 70 to 50Comscore prompts 70%50%
Slope chart: ChatGPT web visit share falls from 76 to 53 percent in Similarweb data, and prompt share falls from 70 to 50 percent in Comscore data
ItemValue
Similarweb visits76 to 53
Comscore prompts70 to 50
Similarweb and Comscore both show ChatGPT's share falling by roughly 20 points, though over different windows and measures, so read the direction and not the level.

The Comscore comparison comes from its own Q2 2026 AI report, published this week and covered in our companion piece on ChatGPT market share and where content should live. The point for app teams is practical: ChatGPT's share of prompts is shrinking even though it still leads on referrals, so measure it first and the rest alongside it, because any AI search tracking or ChatGPT analytics setup that ignores other assistants understates the total.

Google adds a wrinkle. Its guidance on AI features says sites appearing in AI features "are included in the overall search traffic in Search Console" (Google Search Central), which means Google's own AI surfaces are not a separate line for a web team, and are even less visible for an app. This is why a single number for AI-driven installs should be received with the same courtesy as any other modelled figure.

Section 04

Five real checks for your mobile measurement partner

You do not have to wait for the industry to settle. Five plain questions to any mobile measurement partner, including AppsFlyer, will tell you how much of your AI traffic analytics to believe, and they double as a sensible brief for any AI search tracking you buy.

Five checks, in order
Ask how an AI referral is identified

Get the rule in writing: referrer, link parameter, landing page or model. Each has different blind spots.

Ask what happens without a referrer

Find out where unattributed installs go. If they fall into organic or direct, your AI number is a floor.

Ask for the inferred share

Request the split between observed and inferred installs for a sample week, and check it does not drift.

Tag the links you control

Use tagged deep links wherever an assistant may cite you, so you generate the referrer instead of hoping for one.

Reconcile with the stores

Compare partner totals with App Store and Google Play console installs, and note any gap before you move budget.

The third check is the one that separates good measurement from a good chart. It also rhymes with the wider habit of treating any attribution number as a range, which our note on how Snap changed the count, not the campaign and our guide to choosing between MMPs both press.

  • Paid AI ads

    Measured through the ad platform's conversion API

    Signal
    Advertiser-controlled
    Example
    ChatGPT Ads via Conversions API

    Best for

    • Budget decisions
  • Organic AI answers

    Measured by referrer and parameters

    Signal
    Partial, often missing
    Example
    AEO attribution across eight engines

    Best for

    • Directional trends
  • Apple postbacks

    Privacy-preserving, thresholded

    Signal
    Delayed 24 to 48 hours
    Example
    AdAttributionKit

    Best for

    • iOS install counts

None of the three is complete on its own. A working app measurement stack is a small parliament of imperfect witnesses, and your job is to notice when they disagree. Good AI traffic analytics keeps all three on the stand, and a patient prowl through their differences is where the insight lives.

An AI attribution number is a floor. AI traffic analytics only starts when you ask how high the ceiling might be.
folkfox, on organic attribution

If you want AI traffic analytics set up as a working measurement plan, the app marketing team and the search and GEO team do it together, because the assistant that sends the click and the store that sees the install belong to the same story.

Start with the cheapest step this week: tag the links you control, and ask your partner the third question. The fox does not need a map of the whole hillside to see whether the far window is lit.

Section 05

A quarter-long plan for AI traffic analytics in an app team

In the first month, do only the cheap work: tag the links you control, agree the identification rule with your partner, and record a baseline of installs that arrive with no referrer at all. Do not change a single budget line. The aim of month one is a number you can defend, not a number that looks large.

In the second month, compare. Line up the partner's AI traffic analytics against the store consoles, against your web analytics for the same landing pages, and against any survey question you already run at install, such as "how did you hear about us". Where two sources disagree, write down why before you decide which one to believe. A disagreement you can explain is progress, and one you cannot is a finding.

In the third month, decide. If the AI-attributed share is small and stable, keep measuring and spend your effort elsewhere. If it is growing and its inferred share is low, you have a channel worth planning for, and the work becomes ordinary: which questions do people ask an assistant before installing your app, and does your page answer them plainly? The same discipline applies to organic attribution in general, and to ChatGPT referral traffic in particular, which will keep changing as assistants change how they link.

That last point is the durable one. Assistants alter their linking, their citations and their parameters without warning, so any AI traffic analytics setup should be reviewed each quarter and treated as a living instrument, not a fixed report. The fox does not build a den in the first dry hollow it finds; it walks the ground in every season.

Questions

Frequently asked questions

What is a mobile measurement partner?

A mobile measurement partner is a third-party vendor that attributes app installs and in-app events to the marketing sources that drove them. Advertisers use one to compare channels with a common yardstick, and to receive privacy-preserving signals from platforms.

What is app attribution?

App attribution ties an install or event to the source that caused it, such as an ad, a link or an organic search. It relies on referrers, link parameters, device signals and, on iOS, privacy-preserving postbacks from Apple.

What is organic attribution for apps?

Organic attribution assigns installs and re-engagement to unpaid sources, such as search, referrals and answer engines. AppsFlyer has extended its organic search attribution to eight AI answer engines. Missing referrers mean organic figures are usually a floor.

How do I track ChatGPT referral traffic to my app?

Tag the links you control, use deferred deep links where an assistant may cite you, and check your partner's rules for identifying an AI referral. Expect some ChatGPT referral traffic to appear as direct because links often lack UTM parameters.

What is AI search analytics?

AI search analytics is the measurement of visits, citations and conversions that come from AI assistants and answer engines. It is immature: referrers are inconsistent, sample panels disagree, and some traffic is counted as direct.

Can I trust AppsFlyer's new AI attribution?

Treat it as useful but vendor-supplied. It covers installs and re-engagement across eight AI engines, and its accuracy depends on how referrals are identified and how missing referrers are handled. Ask for the method and the inferred share before you move budget.

Keep reading

Read more on this topic

Want AI attribution you can put in front of finance?

folkfox designs app measurement plans that show what is observed, what is inferred and what is missing, before a budget moves.

Want folkfox in your Google results and AI answers? Set folkfox as a preferred source.

The den

Where to?

Pricing

Choose a section. Enter opens it, Escape continues reading.

Cookie preferences

folkfox uses data the way we use strategy: only when it earns its place.