Apple's App Store advertising auction is less driven by relevance than the company's own documentation implies - and a dataset-driven analysis published today by Mike Rhodes of ConsultMyApp puts hard numbers to a question that App Store advertisers have long debated: does semantic relevance actually determine who wins a placement?

The short answer, according to the research, is only partly.

Rhodes built a real UK Apple Search Ads auction dataset throughout his project using APPlyzer, covering 132 keywordsand 627 keyword-app auction observations. Each app occupying each of the five auction positions was matched to its App Store metadata - title, subtitle, description, and category - then rescored using a semantic large language model relevance model. That model did not measure keyword overlap. Instead, it evaluated a more practical question: would a user searching for this term reasonably expect this app to solve their need? Scores ran from 0 to 100.

The resulting dataset allowed a direct comparison between the semantic relevance ranking of apps for each keyword and the actual auction order Apple produced.

The headline numbers

The findings challenge assumptions embedded in most Apple Search Ads campaign strategies. According to the analysis, the exact relevance order matched the Apple auction order on only 7.6% of keywords. More strikingly, the most semantically relevant app took the top auction position in only 43.9% of keywords.

That 44% figure is the critical one. If semantic relevance were the primary ranking driver, the most relevant app would win substantially more often. Instead, the data strongly suggests that once an app clears Apple's relevance threshold, bid strength and predicted performance do most of the ordering work.

What the data shows

The analysis includes four key datasets that together make the case.

The relevance-by-position data is the most striking. It plots the distribution of semantic relevance scores across all five auction positions. The distributions for positions one through five overlap heavily - a finding that directly contradicts any assumption that Apple systematically places the most relevant app first. Crucially, the average relevance score at position two is slightly higher than at position one in this dataset. That single data point, according to Rhodes, is a strong signal that bid and expected performance are reshuffling the final order after eligibility is determined.

The match-type mix data breaks down the 627 observations by inferred targeting pathway using the relevance thresholds established in the model. Apps scoring 75 or above were classified as likely exact match placements. Scores between 35 and 74 were categorised as probable broad match. Scores below 35 were classified as likely Search Match placements. The result: the auction is not dominated by irrelevant results - most ads in the stack still show strong relevance - but nearly one in five observations displayed Search Match-style loose expansion. That proportion, according to the analysis, is large enough to create significant noise, distort cost-per-tap rates, and generate competitive groupings that have no clear semantic logic.

The conquest patterns data visualises the first strategic cluster: apps that repeatedly appear in high positions despite weak or moderate relevance. These are the likely overbidders and conquest advertisers. The Search Match-heavy appsdata shows the flipside - apps appearing across wide keyword sets with consistently weak semantic fit, the strategy fingerprint of large-scale Search Match usage across at least five distinct keywords, with 50% or more of appearances classified as Search Match.

How Apple's auction actually works

Apple's public documentation tells advertisers that Search Results ads are determined by a combination of keyword targeting, relevance, and bid. Broad match expands to close variants, misspellings, synonyms, related searches, and phrases containing the keyword. Search Match automatically matches ads using app metadata, similar apps, and other search data.

Apple also explicitly recommends using more aggressive bids for important exact match keywords, strong bids for broad match, and more moderate bids for Search Match discovery traffic. According to Rhodes, the dataset is consistent with that published framework - but fills in what the documentation leaves unsaid.

The analysis proposes a three-stage model. First, candidate generation: exact, broad, or Search Match pathways each produce a set of potentially eligible apps. Second, eligibility gating: Apple applies a relevance threshold, but that threshold varies by pathway. Exact match applies the strongest semantic gate. Broad match relaxes it considerably. Search Match applies a very weak semantic gate, relying instead on behavioural and similarity signals. Third, auction ranking: once an app clears the gate, bid and predicted engagement determine the final order.

The formula suggested in the analysis is: Auction rank ≈ bid × predicted engagement × relevance. Apple does not publish that formula directly, but according to Rhodes, the behaviour of the dataset is consistent with it.

Apple Ads introduced Maximize Conversions as a general-availability bid strategy on February 26, 2026, replacing the older CPA cap mechanism. That automated bidder sets bids in real time for every relevant search query. Understanding the underlying relevance-gate structure is therefore more consequential than before: advertisers relying on automation without grasping how the gate works may misattribute poor placement outcomes to metadata problems when the actual cause is commercial.

Six real examples from the auction stack

The research presents six worked examples drawn directly from the UK auction data, each illustrating a distinct pattern.

"vpn" represents a clean, relevance-led market. The keyword is explicit, intent is narrow, the category is mature, and the entire auction stack is semantically on-topic, with all five positions occupied by apps directly in the VPN category. According to the analysis, this is what Apple Ads looks like when working as advertisers expect - and why Apple recommends exact match as the primary source of efficient traffic.

"flight tracker" demonstrates the opposite. Gemini occupied position one despite the dataset recording it as the least relevant app in the entire stack by a wide margin. Four actual flight-tracking apps occupied the remaining positions. The analysis concludes this result makes sense only if Search Match or broad expansion surfaced Gemini as a candidate, the relevance gate was permissive enough to admit it, and bid or predicted performance pushed it to the top slot.

"pdf scanner" is the most extreme case. Audible held position one. Audible is not a scanner, not a document capture tool, and not functionally related to pdf scanning in any practical App Store sense. According to Rhodes, this is one of the strongest pieces of evidence in the dataset that Search Match can be extremely permissive. Apple's expansion of ad positions across App Store search results began rolling out on March 3, 2026 in the United Kingdom and Japan, making this kind of mismatch more costly: there are now more positions to misallocate budget across.

"ai photo enhancer" shows the reverse. PhotoBoost sat at position five despite recording the highest semantic relevance score of any app in the entire set. According to the analysis, this is a commercial problem, not a metadata problem. The app looks undervalued in the auction rather than irrelevant to the keyword.

"digital catalog" illustrates a market almost entirely detached from semantic fit. There is no clear exact-match leader - the results appear to have drifted from the core intent of the keyword through Search Match-style loose expansion. In such a market, isolating exact match keywords and aggressively sculpting negatives is, according to the analysis, the only route back to quality traffic.

"translator" closes the set with the underbidding paradox. iTranslate, a direct semantic match scoring 100 out of 100, sat in position five. The keyword is relevant, the app is relevant, but the auction slot says the commercial strength is not there. Relevance alone does not buy the top position.

The aggregate match-type picture

The full auction breakdown across all 627 observations, using the 75/35 relevance thresholds, confirms the auction is not mostly noise. The majority of the stack falls into the likely exact or broad match buckets. But the Search Match-classified tier - just under 20% of total observations - is large enough to distort competitive landscapes meaningfully across many keyword categories.

Bid Pressure Index and the two strategic clusters

To make the findings actionable, Rhodes constructed a Bid Pressure Index calculated as: Relevance Rank minus Auction Position. Positive values indicate an app winning a higher position than its relevance rank would predict. Negative values indicate an app is more semantically relevant than its position suggests.

The conquest cluster shows apps repeatedly appearing high despite modest or weak relevance - advertisers likely leaning on Search Match heavily, bidding aggressively outside their core intent, or running broad conquest-style acquisition strategies.

The undervalued opportunity cluster shows the opposite: apps among the best semantic matches for keywords, consistently finishing low in the stack. According to the analysis, these represent the best candidates for exact-match isolation, bid increases, or budget protection - they signal genuine commercial upside rather than a content or metadata problem.

App-level strategy fingerprints

Section six of the research reveals distinct advertiser strategy patterns at the app level.

The Search Match-heavy fingerprint shows apps appearing across broad keyword sets with weak semantic fit across at least five distinct keywords, with more than half of appearances classified as Search Match. These are not apps drifting into adjacent intent on one or two terms - they are systematically appearing across wide keyword sets.

The exact-heavy fingerprint shows advertisers whose appearances are consistently semantically tight - either prioritising exact match or operating in very clean intent markets where broad and Search Match are not distorting the competitive set.

The broad-heavy fingerprint sits between those poles: apps exploiting adjacent intent through broad match rather than full-blown Search Match drift.

Apple rebranded from Apple Search Ads to Apple Ads in April 2025, signalling broader advertising ambitions beyond search-only placements. The platform now operates across 175 storefronts and 44 currencies, and the App Store averaged 850 million weekly users globally throughout 2025. Apple expanded geographically in October 2024, adding 21 new countries including Türkiye, Cyprus, and Morocco. Scale makes the gap between published documentation and observed auction behaviour progressively more expensive to ignore.

What the data means for campaign managers

The practical implications of the ConsultMyApp analysis are significant. Semantic relevance matters, but it is not the dominant factor in Apple's auction. Bid and predicted performance appear to drive most of the final ordering once an app clears the relevance gate. That distinction has direct consequences for how campaign managers diagnose poor placement outcomes.

If an app is highly relevant but consistently under-positioned, the problem is likely commercial rather than a matter of App Store metadata quality. If an irrelevant or weakly relevant competitor is repeatedly outranking a more relevant app, the likely explanations are Search Match surfacing that competitor, broad match expanding into the query set, the competitor outbidding the incumbent, or Apple predicting stronger engagement or conversion rates from the competitor.

According to the analysis, the best Apple Search Ads teams do not manage solely by cost-per-tap and cost-per-acquisition. They manage by auction structure - exact versus broad versus Search Match, bid pressure versus relevance, and discovery noise versus true intent. APPlyzer's Apple Auction Stack data provides the observational layer that makes that kind of structural analysis possible at scale.

Timeline

Summary

Who: Mike Rhodes of ConsultMyApp (Rhodes Consulting Limited, UK Company Number: 10490546, registered at Bushbury House, 435 Wilmslow Road, Manchester, M20 4AF), published the analysis using auction data from APPlyzer.

What: An analysis of 132 keywords and 627 keyword-app auction observations from a real UK Apple Search Ads dataset, scored with a semantic LLM relevance model, found that the most relevant app takes the top auction position in only 43.9% of keywords, and that the exact relevance order matches Apple's auction order on just 7.6% of keywords. Data covering relevance-by-position distributions, match-type mix, conquest patterns, and Search Match-heavy app profiles supports the conclusion that Apple's auction functions as a relevance-gated, bid-and-performance-weighted system with varying eligibility thresholds depending on match pathway.

When: The analysis was published on March 10, 2026, drawing on auction data collected throughout the research project using APPlyzer's UK App Store dataset.

Where: The auction data reflects the UK App Store, with five auction positions (ASA #1 through ASA #5) observed per keyword. The findings concern Apple Ads, which operates across 175 storefronts and 44 currencies globally.

Why: The research matters because Apple Ads is one of the fastest-growing mobile advertising platforms, now processing search queries from 850 million weekly App Store users. As Apple expands ad positions - with multiple placements rolling out globally through March 2026 - and migrates advertisers toward automated bidding via Maximize Conversions, understanding what actually determines auction outcomes is central to campaign management. The gap between Apple's published documentation and the auction's observed behaviour has direct financial consequences for app developers who diagnose placement problems incorrectly.

Share this article
The link has been copied!