Apple probably isn’t reading your screenshot captions.
Since the summer of 2025, a growing pile of ASO articles has said the same thing: Apple now runs OCR on your App Store screenshots and ranks you for the words in your captions. Some present it as settled fact and some sell tools built around it. The evidence behind the claim is thinner than the confidence. The only published test we could find points the other way. Here is what is actually known, and how to write captions that work either way.
Where the claim came from
Something did change. ASO trackers saw a large reshuffle in App Store search rankings in the first week of June 2025, and ConsultMyApp dates the update to June 6. Within days, Appfigures published a guide calling it the biggest algorithm change in years and stating that “Apple is now extracting text from your app’s screenshot captions and treating that text as part of your keyword metadata.”
The evidence offered was the author’s own analysis of thousands of keywords, plus the observation that some apps rank for keywords that “only appear in their screenshots.” No app list, keyword list or method was published, and the same article flags parts of its advice, such as where on the frame captions should sit, as speculation. That is a reasonable hypothesis from someone who stares at rankings all day. It isn’t a finding anyone else can check. Many later articles repeat the claim with more certainty, and most of them cite each other.
What Apple says it indexes
Apple’s own App Store search page names two groups of inputs. The first is text relevance, meaning your app name, subtitle, keyword field and primary category. The second is user behaviour, such as downloads and ratings. Screenshots aren’t on that list. The same page goes out of its way to say that promotional text does not affect search ranking, so Apple is willing to tell developers which fields don’t count. It has said nothing either way about screenshot text, and none of the articles making the claim cite an Apple statement or document.
Silence isn’t proof. Apple doesn’t publish its ranking weights and changes them without notice. But it means the claim can only be settled by testing, and very little testing has been published.
The one test we could find
ConsultMyApp ran it with its APPlyzer tool shortly after the June update. It took the top two apps in four US categories (Books, Business, Education and Finance), which gave it Audible, Libby, Indeed, DoorDash, Duolingo, PictureThis, Cash App and PayPal. It pulled 64 phrases straight from their screenshot captions and checked two things for each one: whether the app ranked for the phrase, and whether the words already appeared in an indexed field.
The result:
- 36 phrases didn’t rank at all. DoorDash’s “see your pay for every delivery” and Duolingo’s “bite-sized lessons that work” were both clearly visible on screen and both returned nothing.
- 27 phrases ranked, but the ranking was explained by words already in the title or subtitle. PictureThis ranked #1 for “identify any plant”, which its title and subtitle already cover.
- 1 phrase couldn’t be explained. Audible ranked #2 for “choose 1 title every month bestsellers originals”, lifted from its third screenshot and absent from its title and subtitle.
The Audible case is odd. It ranked for three-word chunks of the phrase in the exact order they appear on the screenshot, but not for “bestsellers” or “originals” on their own. That pattern is what you’d expect if the caption were indexed. It also fits a keyword field nobody outside Audible can see, or text Apple indexes from somewhere else. The authors’ own verdict was “no strong evidence Apple is broadly indexing screenshot titles.”
Why it’s easy to fool yourself
If you test this on your own app, the 27 explained phrases are the trap. App Store search combines words across your name, subtitle and keyword field, so a caption phrase can rank because its words are scattered through your metadata. You see your app at #4 for the caption’s exact wording and conclude that the screenshot did it, when the words were already indexed and the screenshot played no part.
Timing is the other trap. On the App Store, screenshots are tied to a version, so new captions usually ship with an update that may also change the subtitle or keywords, or shift downloads. Rankings also move on their own from week to week. One before-and-after comparison around a release can’t separate those effects.
A test that could actually tell you something looks like this:
- Pick a phrase of three or more words where no single word appears in your name, subtitle or keyword field. Partial overlap is how most false positives happen.
- Check that you don’t rank for it today, and pick a second phrase you leave out of the captions as a control.
- Ship a version where the only metadata change is the caption, then check both phrases daily for two to three weeks.
- Only call it if the caption phrase starts ranking and the control doesn’t. One app and one phrase is still an anecdote, so do it more than once before you rely on it.
What to write in your captions
None of this makes captions less important. Your first frames show up in search results, and the caption is often the only line a scrolling user reads before they decide whether to tap. That conversion effect is real and well understood. The ranking effect is a maybe. So write for the effect you know exists:
- Keep search terms in the fields Apple names. Your name and subtitle get 30 characters each and the keyword field gets 100. Those are where ranking is documented, and caption text doesn’t replace them. We covered how to spend those 100 characters in an earlier post.
- Use the words people search for when they fit the sentence. “Track your sleep” is a better caption than “Rest, reimagined” whether or not Apple reads it, because it tells a scrolling stranger what the app does in plain words. If the caption does turn out to be indexed, you get that for free.
- Don’t stuff them. “Budget tracker, expense tracker, money manager” across a frame reads as spam to the human it’s meant to convert. It gives up a known conversion gain for a possible ranking gain. Apple’s Guideline 2.3.7 also warns against packing metadata with irrelevant phrases “just to game the system.”
- Make them legible at thumbnail size. This is common advice for OCR, but it matters more for people: high contrast, a plain face, short lines and no text over a busy photo. A caption a phone user can’t read in the search grid doesn’t help you under either theory.
The general lesson
ASO has a lot of folklore. A ranking jump coincides with an update, someone offers an explanation, and a year later dozens of articles repeat it as fact. Before you redesign a screenshot set around a ranking claim, ask who tested it, how, and whether they published the results. In this case the answer was one vendor with unpublished data and one vendor with a 64-phrase test that mostly came back empty. The captions that convert will serve you no matter how that debate ends, so write those.
ShotCanvas writes caption drafts from your actual screens and keeps each one short enough to read in the search grid, so you can spend your effort on the words.