The short answer
SKAN cannot measure a creator or influencer campaign directly, because SKAdNetwork only reports on installs that came from an ad served by a registered ad network, and a creator posting a link is not one.
Read the full answer
Those installs land in your organic bucket, which is why influencer budgets look unprofitable and get cut.
What works instead is a stack of three imperfect methods read together: paid amplification of the creator's content through Meta or TikTok so the install does pass through SKAN, creator-specific entry points such as a unique link or an App Store custom product page per creator, and a geo or time-based holdout that measures the lift in total installs rather than trying to attribute individual ones.
SKAN 4 helps at the margin by widening the source identifier from 100 values to 10,000, so a paid campaign can carry creator-level granularity, but only when install volume clears Apple's crowd anonymity thresholds, which AppsFlyer puts at roughly 20 installs per campaign per day.
Each window closes, then the postback arrives on a random delay: up to two days for the first, up to six days for the other two. Subscription revenue mostly lands in window 3, so the number that decides whether a creator was worth paying is nearly six weeks behind the post.
Source: Apple, SKAdNetwork 4 specification, as documented by Singular and AppsFlyerInfluencer campaigns on mobile apps get killed by a measurement artefact more often than by performance. The post goes out, the installs arrive, and the dashboard credits them to organic, so the line item shows spend against nothing.
Nothing is broken when that happens. Apple’s SKAdNetwork reports on installs attributed to an ad served by a registered ad network, and a creator sharing a link is not an ad network. The framework has no opinion about creator campaigns because it was never pointed at them.
Can SKAN measure influencer and creator campaigns at all?
SKAN cannot measure a creator campaign directly. SKAdNetwork postbacks exist because an ad network served an impression or a click and called the SKAdNetwork API, and that chain has no link for a creator who filmed something and posted it to their own audience.
The framework also carries limits that work against creator-level reporting even when a creator’s content does run as a paid ad. Postbacks are aggregated, delayed and thresholded, so a single creator sending modest volume is exactly the case SKAN is designed to hide.
Three mechanics matter here.
Postbacks arrive in up to three windows, covering days 0 to 2, days 3 to 7 and days 8 to 35 after install, and each one lands on a random delay after its window closes: 24 to 48 hours for the first postback, 24 to 144 hours for the second and third.
Conversion values are coarse or withheld entirely below Apple’s privacy thresholds. And the source identifier that would carry the creator name is truncated when volume is low.
So the honest framing is that SKAN measures channels, not creators, and measures them on a delay. Treating it as the system of record for influencer spend produces a confident number that is mostly made of absence. Our wider piece on tracking app signups and subscriptions after iOS privacy changes covers what the framework does well.
Why do creator installs show up as organic?
Creator installs show up as organic because organic is where an MMP puts an install it cannot attribute to anything. It is a residual category rather than a finding, and on iOS it absorbs most word of mouth, most store search, and almost all creator activity.
The sequence is simple enough. Someone watches a video, does not tap the link, opens the App Store a few hours later and searches the app name. No click, no ad network, no identifier, no postback. The install is real, the creator caused it, and nothing in the stack knows that.
This produces a specific and expensive failure. Creator spend sits in one row with few or no installs beside it, while the revenue those creators generated sits in the organic row looking like it arrived for free. Read literally, the dashboard says to cut the creator budget and do more of whatever is producing all that organic.
The tell that it is happening is a correlation your attribution denies. If organic installs rise on the days creator content goes live and fall when it stops, the organic bucket is carrying the creator campaign. Plot the two together weekly before concluding anything from the attributed numbers.
Two further distortions follow. Cost per install on the creator line reads far too high, because the denominator is missing most of its installs. And any model trained on attributed data will under-allocate to creators permanently, because it is fitting to a measurement gap rather than to the market.
What does SKAN 4 actually change for creator measurement?
SKAN 4 widens the identifier, which is what makes creator-level reporting possible inside a paid campaign for the first time. Per Apple’s SKAN 4 specification, the field went from two digits, giving 100 campaign values, to four digits, giving 10,000, and was renamed the source identifier.
Apple also asks advertisers to read those four digits as a hierarchy rather than as one number, so the leading digits can carry the coarse grouping and the trailing digits the detail. That structure is what lets a campaign encode a creator, a placement and a creative at once.
The catch is that you only receive the full identifier at the higher crowd anonymity tiers. Lower tiers return two or three digits and a coarse conversion value, and the bottom tier returns roughly what SKAN 3 did. AppsFlyer puts the volume needed for the higher tiers at around 20 installs per campaign, per day, per app.
| What you want | What SKAN 4 gives you | What limits it |
|---|---|---|
| Creator-level attribution | Up to 10,000 source identifier values | Full four digits only at high crowd anonymity tiers |
| Revenue per creator | Coarse or fine conversion values per window | Values withheld or coarsened below Apple’s thresholds |
| A fast read on a post | Three windows, the last closing on day 35 | Postbacks lag their window by up to 2 days, or 6 for windows 2 and 3 |
| Comparable channel data | One aggregated framework across networks | Singular put SKAN 4 at roughly 40% of postbacks in 2025, so adoption stayed partial |
The 20-install threshold sets the floor on how granular you can get. A programme of thirty small creators cannot be separated inside SKAN. Three creators each driving real volume can.
One genuine improvement: AppsFlyer’s analysis of 240 million postbacks across more than 2,000 apps found that postbacks arriving with a null conversion value fell by roughly 70% on the first SKAN 4 postback compared with earlier versions. The data you do get is more complete than it was, which is worth the migration effort on the paid side.
Worth knowing where the framework is heading. Apple introduced AdAttributionKit at WWDC 2024 as SKAdNetwork’s successor and expanded it at WWDC 2025, and it carries SKAN 4’s features forward. Apple has announced no deprecation date, so SKAN still carries production traffic, and none of this changes the creator problem: AdAttributionKit also only reports on installs that arrived through an ad network.
How do you get creator-level data without SKAN?
Get creator-level data by giving each creator their own entry point and reading what happens at it, rather than trying to attribute the install at the other end. Four methods do most of the work, and they fail in different directions, which is why they are used together.
Run the creator’s content as a paid ad. Partnership ads on Meta and Spark Ads on TikTok put the creator’s post into the ad system, which means the install passes through SKAN and arrives with a source identifier you control.
This is the only route that makes a creator legible to the attribution stack at all, and it is covered in what TikTok Spark Ads are and how Meta partnership ads work.
Give each creator a custom product page. Custom product pages are alternative App Store listings with their own URLs, screenshots and headlines, and they report their own impressions and conversion. A page per significant creator gives you a store-side signal with no dependence on SKAN, and it lets the listing repeat the specific promise the creator made.
Use unique links and codes, and treat them as a floor. A tracked link captures only the users who tap rather than search, and a promo code captures only those who remember it. The number is real and it is an undercount, so never divide spend by it and call the result a CPI.
Ask the user. A single question on first open, worded plainly, catches the installs that no technical method can see. Self-reported attribution is noisy and biased towards whatever the person remembers last, which is exactly why it belongs next to the other three rather than instead of them.
How do you prove a creator campaign worked when attribution is incomplete?
Prove it with a holdout, which measures the lift in your totals instead of trying to identify individual installs. Run creator activity in some markets and not others, or in alternating two-week periods, and compare total installs, total trials and total new subscribers across the two conditions.
A holdout survives privacy changes because it never needed to know who anyone was. It compares aggregate outcomes, which is the one thing iOS still reports accurately, and it credits the untracked majority automatically instead of leaving it in organic.
Geography is the cleaner split when you have the volume for it, because the two groups run at the same time and absorb the same seasonality. Time-based splits are available to everyone and are more fragile, since anything else that changed between the periods lands in the result.
Three conditions make the test honest. Hold paid spend flat across the split, or the test measures your media plan rather than your creators. Run it long enough for the subscription signal to mature, which for a day-35 conversion window means at least six weeks. And decide the metric before you start, because a holdout read after the fact turns into a search for the comparison that worked.
Expect the answer to be a range rather than a number, and expect it to be larger than your attributed figure. The method we use for this on the influencer side is described in how to measure influencer marketing ROI.
What should a creator measurement setup actually look like?
A workable setup has four layers and takes about a week to build. Nothing in it is exotic, and most launch-stage apps are missing two or three of the layers rather than all four.
Start with the paid layer, because it is the only one SKAN can see. Map conversion values to the events that matter for your economics, trial start and first payment rather than install, and encode creator or creative in the source identifier for the campaigns that clear the volume threshold.
Add the store layer next. One custom product page per creator worth more than a trivial amount of spend, with the headline and first screenshot matching that creator’s promise, and the page’s own conversion rate read weekly.
Then the self-reported layer, which is one question and takes an afternoon. Then the holdout schedule, which is a calendar decision rather than an engineering one: pick the markets or the weeks now, in advance, and write down what you expect to see.
Read all four together and accept that they will disagree. The paid layer undercounts creators, the links undercount, the survey over-credits whatever is most memorable, and the holdout gives you a total with no breakdown. Building and running this whole stack alongside the creative is the work described on our app launch page.
The short version
SKAN does not measure creators, and no amount of configuration will make it. SKAdNetwork reports on installs served by a registered ad network, so a creator posting a link produces no postback and the install lands in organic, which is why influencer spend reads as unprofitable and gets cut while the revenue it produced sits in a bucket that looks free.
SKAN 4 helps on the paid side. Apple widened the source identifier from 100 values to 10,000 and, according to AppsFlyer, cut null conversion values on the first postback by around 70%.
Both gains are gated on volume, since the full identifier only arrives at the higher crowd anonymity tiers, which AppsFlyer puts at roughly 20 installs per campaign per day.
According to Singular’s adoption dashboard, SKAN 4 accounted for about 40% of postbacks in 2025, so it never became the industry default before Apple moved on to AdAttributionKit.
Build the measurement elsewhere.
Run creator content as paid ads so it enters SKAN at all, give each meaningful creator a custom product page, use links and codes as a known undercount, ask users on first open, and settle the question of whether the channel works with a geo or time holdout on total installs and total subscribers.
The holdout is the only one of these that is not damaged by privacy changes, because it never needed to identify anyone.
Frequently asked
Does SKAdNetwork work for influencer marketing?
No, not directly. SKAN reports on installs attributed to an ad served by a registered ad network that calls the SKAdNetwork API.
A creator posting a link in a bio or a story is not an ad network, so there is no postback and the install is counted as organic.
The way to get a creator into SKAN is to run their content as a paid ad through Meta or TikTok, which turns the creator post into an ad impression the framework can see.
Why do my influencer installs show up as organic?
Because nothing in the chain fired an attribution signal. On iOS, without an ad network postback and without a device identifier, your MMP has no basis for crediting the install to anything, so it defaults to organic.
The practical consequence is that the influencer budget shows a cost with no revenue beside it, while the revenue appears in a bucket that looks free.
How many creators can you track inside one SKAN campaign?
Up to 10,000 values in theory, since Apple's SKAN 4 specification widened the source identifier from two digits to four, per the documentation AppsFlyer and Singular publish on it. In practice the limit is volume rather than capacity, because Apple only returns the full four digits at the higher crowd anonymity tiers.
AppsFlyer puts the threshold for those tiers at around 20 installs per campaign per day, which means a long tail of small creators cannot be separated.
Are promo codes and tracked links enough to measure a creator campaign?
They are useful and they undercount, usually badly. A tracked link only captures the people who tap it rather than searching the app name, and a promo code only captures the people who remember to use it.
Treat both as a floor on the campaign's effect rather than a measurement of it, and pair them with a holdout test that measures total installs.
What is a custom product page and why does it help here?
A custom product page is an alternative version of your App Store listing with its own URL, different screenshots and a different headline.
Giving each significant creator their own page means the listing can repeat the promise the creator made, and the page's own impression and conversion data gives you a creator-level signal that does not depend on SKAN at all.
How do you prove a creator campaign drove installs if nothing is attributable?
Measure the lift rather than the attribution. Run the creator activity in some markets and not others, or in one period and not the next, and compare total installs and total new subscribers between them.
It is less satisfying than a dashboard row, and it is the only method that survives iOS privacy changes intact, because it never needed to identify anybody.
Do you still need an MMP for creator campaigns?
Yes, for the paid side, and no for the creator side. An MMP consolidates SKAN postbacks, handles conversion value mapping and gives you one place where paid channels can be compared.
It cannot attribute an install that never passed through an ad network, so the creator-level work sits in custom product pages, links, codes and holdouts rather than in the MMP.
Should You Hire an App Growth Consultant, an Agency, or Do It In-House?
Why Aren't My App Installs Converting Into Trials?
What Is a Good App Paywall Conversion Rate?