A normal Apple Ads attribution claim is a small object with eleven fields in it. It tells you the org, the campaign, the ad group, the keyword, the ad, the country, the placement, the conversion type, the claim type and the moment the tap happened. Turn on an age or gender refinement in that ad group and the object collapses to one field: {"attribution": false}.
Not a partial payload. Not a coarse payload. Nothing. Apple’s help page states the rule twice, without hedging: “To ensure user privacy, payloads will not be returned for ad groups that use age or gender settings.”
Two measurement vendors date the September 2026 changes to the first of the month. As of writing, ten days later, no publication has covered this. Apple announced nothing. And the trade that advertisers are now being asked to make — demographic precision in exchange for total measurement blindness — cannot be priced, because Apple publishes exactly one number about demographic targeting and it is not a reach figure.
01What Apple actually returns now
Apple publishes both payloads side by side. With demographic settings off, the response carries the full field set — orgId, campaignId, adGroupId, keywordId, adId, conversionType, claimType, countryOrRegion, supplyPlacement, and a clickDate or impressionDate rounded to the minute. With them on, Apple’s documented sample response is the two-word object above.
It is worth being precise about what that costs, because “you lose attribution” understates it. You do not lose a channel-level number. You lose the ability to answer which campaign, which ad group, which keyword and which placement produced the install — and since supplyPlacement is gone too, you cannot even tell whether it came from search results, the search tab, a product page or the Today tab. For a keyword-granular platform, that is the whole apparatus.
| Field | Demo off | Demo on |
|---|
| attribution | true | false |
| campaignId | Returned | Withheld |
| adGroupId | Returned | Withheld |
| keywordId | Returned | Withheld |
| adId | Returned | Withheld |
| supplyPlacement | Returned | Withheld |
| clickDate | Returned | Withheld |
| countryOrRegion | Returned | Withheld |
This lands on top of the other September change we documented in the ATT payload piece: Apple removed the consent-gated standard-versus-detailed split and began returning a touchpoint timestamp on every claim, which lets measurement partners enforce the 30-day window they previously could not. That correction pushes attributed volume gently down for everyone. This one removes it entirely for a subset.
02Ad group or campaign? Apple disagrees with Apple
Before restructuring anything, know that Apple’s two documents on this do not agree.
The Apple Ads help page says ad group, twice: “if the ad group does not use age or gender settings, an attribution payload of true will be sent, while if the ad group does use age or gender settings, an attribution payload of false will be returned.”
Apple’s own AdServices API reference PDF — the version whose URL is path-dated 2026-09-01 — says campaign: “Note: Campaigns with age and gender targeting return false.”
Those are different blast radii. Under the help page’s reading, a clean ad group in a campaign that also contains a demographic ad group keeps its payload. Under the PDF’s reading, it does not. Singular and AppsFlyer both implemented the ad-group reading, and the help page is the newer and more specific of Apple’s two texts, so ad group is the defensible working assumption — but it is an assumption, and this is exactly the class of live self-contradiction we keep on file in the documentation contradictions piece. Verify it against your own postbacks before you move budget on it.
→ ONE THING TO CHECK
Isolate before you conclude
If you run demographic and non-demographic ad groups inside one campaign, you cannot distinguish the two readings from aggregate data. Split them into separate campaigns for a fortnight and watch whether the clean campaign’s attributed installs hold. That is a two-week test that settles an ambiguity Apple has left open.
03Where the warning isn’t
Apple documents this rule on its measurement page. It does not document it on the page where you turn the setting on.
We checked the audience-settings help page, the understand-and-create-ad-groups page, the ad attribution overview, the Search Match page and the Recommendations best-practices page. The attribution consequence appears on exactly one of them — the measuring-ad-performance page. The audience-settings page, which is the screen an advertiser is actually looking at when the decision gets made, carries a different warning and not this one.
The cost of the setting is disclosed one level up from the control, in a document written for a different reader.
That is not an accusation of bad faith; it is a description of how a large, undated documentation estate drifts. But the practical effect is that a media buyer can make this trade without ever encountering its price, and the platform will not stop them. It is the same failure mode as Apple’s low-delivery troubleshooting page still telling advertisers to check a CPA cap that the February 2026 bidding change superseded — a case we worked through in the piece on campaigns that will not spend.
04The second cost, which Apple does disclose
On the audience-settings page Apple does publish a warning, and it is a real one: “when you apply age or gender refinements, you automatically exclude customers with Personalized Ads turned Off.”
So a demographic refinement is not one trade, it is two. It narrows delivery to the subset of App Store visitors who have left Apple’s Personalized Ads setting on, and it removes your attribution. Apple publishes no figure for how large that excluded subset is. We could not find one anywhere on an Apple property, and we would treat any specific percentage in circulation as unsourced unless it comes with a method — there is a widely repeated claim about the share of App Store search volume coming from users with personalised ads disabled that traces back to a trade article with no citation behind it.
Apple does publish a floor, on both the ad-group help page and in its advertising privacy notice: a targeted audience must exceed 5,000 customers for ads to be delivered at all. Below that, Apple suppresses.
05One number, and it is a suppression floor
Here is the asymmetry that makes this decision unusually hard to make well.
Apple publishes reach and performance figures for almost everything else it sells. Over 850 million weekly App Store visitors, footnoted to 2026. A tap-through install rate above 60 percent for ads at the top of search results, footnoted to November 2024 through October 2025. Ninety-five percent of downloads after an ad tap occurring within a minute, footnoted to a single month of August 2024. Almost 65 percent of downloads happening directly after a search — footnoted, still, to 2022.
For age and gender targeting, Apple publishes no audience size, no reach estimate, no incremental performance figure and no benchmark of any kind. We checked nine Apple pages. The only quantity attached to demographic targeting anywhere in Apple’s advertising documentation is the 5,000-customer privacy threshold, which is a suppression rule, not a measure of what the targeting buys you.
5,000
The only number Apple attaches to demographic targeting — a minimum audience size below which ads are not delivered. There is no published reach, lift or performance figure to weigh against the attribution you now give up.
A trade with a documented cost and an undocumented benefit is not a trade a disciplined buyer should make by default. It is one you should have to argue for.
06Where Apple’s idea of your gender comes from
Apple’s advertising privacy notice is unusually candid about the provenance of the signal you would be buying: “Information such as your first name in your Apple Account registration page or salutation in your Apple Account may be used to derive your gender.”
This is inference from account metadata, not a declared attribute. The same notice confirms the delivery rule — “personalized ads are delivered only if more than 5,000 people meet the targeting criteria selected by an advertiser” — and confirms that Apple Maps is outside the personalised-ads system entirely: “Personalized ads are not supported on Apple Maps, where you will only see contextual ads based on the information above.” Apple’s published Maps targeting signals are location, map view, business information and the query. No demographic signal appears in the list, a point worth reading alongside what Apple has actually documented about the Maps auction.
On the accuracy of demographic delivery generally, the best public evidence is old and not flattering. Nielsen’s Digital Ad Ratings analysis, published December 2016 across more than 44,000 campaigns in 17 countries, found that only 53 percent of UK ad impressions reached people of the intended age and gender, with accuracy falling as the target narrowed — 58 percent for a 35–64 band, 38 percent for 25–44, and 33 percent for a narrow 16–30 range on mobile. That is a decade-old, cross-platform measurement and not a statement about Apple, and we cite it as the vintage evidence it is. No newer equivalent exists in public.
07Demographics on a surface built for intent
Set the measurement question aside for a paragraph and ask the strategic one: what is a demographic refinement doing on a search platform in the first place?
App Store search is the rare inventory where the user tells you what they want before you have to guess. That is the whole reason the economics work — the argument we set out in the piece on why Apple Ads buys the highest-ARPU iOS users. Apple’s own description of Search Match names its inputs as your App Store listing metadata, information about similar apps in the same genre, and other search data. Demographics do not appear. On the ad-group creation page, Apple reduces them to a co-equal afterthought: “Choose to refine your audience by customer type, demographics, and/or location.”
The one peer-reviewed study we can find that tests gender-targeted keyword phrases directly against neutral ones — Jansen, Moore and Carman, Information Processing and Management, 2013 — found that “gender neutral phrases perform the best overall, generating 20 times the return of advertising than any gender targeted category.” Mean return on advertising was $20.75 for gender-neutral phrases against $3.80 and $1.35 for the two gender-leaning categories. It is one large US retailer, it is thirteen years old, and the authors themselves caution that effect sizes were generally small. It is also the only controlled evidence on the question that exists, and it points the same way the platform’s architecture does.
08The audit, in four steps
1. Inventory every ad group with a demographic refinement. Not campaigns — ad groups. Many accounts carry age or gender settings inherited from a structure built years ago, applied once and never revisited. Those are the expensive ones, because they are costing attribution for a reason nobody remembers.
2. For each, write down what the refinement is doing that the keyword is not. If the honest answer is “the demographic is a proxy for the kind of person who wants this,” the keyword already carries that signal and carries it better. If the answer is a legal or regulatory age gate, that is a real requirement and you keep it.
3. Strip the non-load-bearing ones and take a baseline first. Export attributed installs, conversion rate and cost per acquisition at ad-group level before you change anything. You will want the before-and-after, and Apple gives you no way to reconstruct it later.
4. Re-plumb the ad groups you keep. An ad group that must retain demographic targeting is one whose installs will not appear in Apple Ads attribution at all. Budget it against a different yardstick — total downloads, or a modelled contribution — and stop expecting it to show up in the channel report. That is a hard thing to do well, and the honest framing of how hard is in the incrementality piece: Apple ships no holdout, no ghost ads and no lift reporting, so there is no clean substitute measurement waiting for you.
→ FREE 10-POINT AUDIT
See exactly where your attribution is going dark
One hour with a senior Apple Ads strategist. We walk your account structure, audience refinements, attribution wiring and keyword economics, and hand you a written ten-point list of what to fix first. No pitch.
Book my free audit →
09When it is still worth it
There are cases. Age gating that exists because the law or a platform rule requires it is not optional, and Apple’s minimum targetable ages are published per market — 18 in the US, Europe, Latin America and the Caribbean; 19 in Canada, South Korea and Algeria; 20 in Japan and Taiwan; 21 in Bahrain, Egypt, Kuwait and the UAE. If your category obliges you to gate, you gate, and you accept the blind spot.
The weaker cases are the common ones: a demographic applied because a persona document says the audience skews a certain way, or because the same setting exists on Meta and someone mirrored the structure across platforms. Apple Ads and Google App Campaigns already disagree so fundamentally on what a targeting control even is that mirroring structure between them rarely survives contact with the reporting — a comparison we worked through in the head-to-head piece. Adding a demographic refinement because another platform has one is how an account ends up paying a measurement price for a habit.
10What we still don’t know
Apple has published no effective date for this rule, no changelog entry recording it and no statement on whether it applies retroactively to claims already in flight. The September 1 date comes from Singular and AppsFlyer, not from Apple. Apple Ads help pages carry no publication dates at all, so the page cannot be aged against the change.
Apple’s help page and Apple’s own API reference still disagree on whether the rule binds at ad-group or campaign level. Two of the four major measurement partners — Adjust and Tenjin — had not updated their Apple Ads documentation as of September 11, 2026; Adjust’s pages still describe the pre-September world of consent-gated standard and detailed responses, and carry no last-updated date to warn a reader. No publication has covered the change at all.
And the central number remains unpublished and probably unpublishable from outside: nobody knows what share of Apple Ads conversions currently sits behind a demographic refinement, so nobody can size the reporting hole this opens across the industry. What is knowable is the direction and the asymmetry. The cost is documented, specific and immediate. The benefit has never been quantified by anyone, including Apple. On a surface where the user has already typed what they want, that is a strange thing to keep paying for.