How to read an App Store listing as evidence.

What one listing actually contains

The captured Apple listing fields include the name, developer, category, release date, listed download price, star average, rating count and developer-written description. This snapshot contains no download count, revenue figure, retention curve, review text or search-volume data. The catalog's separate Google keyword reports have their own source and capture date; they are not Apple listing facts. How to distinguish the two sources.

The description is the richest field and the least reliable one. It is written to persuade; it describes an intention, not a behaviour. When three apps describe the same first five minutes, that tells you the demo is obvious — not that the workflow is solved.

The five numbers, and what each can carry

Scroll horizontally to see all columns ↔

Ratings are the field most often over-read. In our whole snapshot, 61.5% of the 307,938 listings have zero ratings, and only 1,623 listings — 0.5% — have more than a thousand. The distribution is not a bell curve; it is a cliff, and most launches sit at the bottom of it. That shape is the honest baseline against which any niche count should be read.

Build the pool, then distrust it

Define the job you want to compare: who needs to do what, and which part of the workflow matters? Use category and keyword filters to form a candidate pool, count what comes back, and analyze the selected full descriptions. The selection rule and sample scope should be explicit.

In the invoicing review, rule estimates-summary-v1 selected Business / Productivity / Finance listings whose first 180 description characters contain invoice, estimate, quote or billing, using a case-insensitive substring OR match. It returned 1,474 candidates. We then read the full descriptions of the sixteen with 50+ ratings: twelve were relevant and four were adjacent or unrelated. Inspect the exact candidate manifest.

Scope correction, 26 October 2026: an earlier version described that selection as a full-description search. Full-description substring matching produces 7,089 candidates and 82 with 50+ ratings; that broader set has not been reviewed here.

Count the pool with a filter. Judge the niche with your eyes. Never publish the first number without the second step.

In this reviewed sample, four keyword matches were not competitors. That demonstrates a precision problem in this filter; it does not establish why false matches occur or how often they occur across the rest of the catalog. Summary screening can also miss capabilities mentioned later in a description.

Read the distribution, not the total

A pool of 1,474 sounds like a crowded market and a pool of 16 sounds like an empty one. Both numbers are true and neither is decisive. What matters is the shape:

• 961 candidates with zero ratings — no rating signal in this snapshot; this does not show whether they have users.
• 497 candidates with 1–49 ratings — a public signal below our chosen reading threshold.
• 16 with 50+ ratings — the selected source-analysis sample. Their descriptions do not represent the whole market.
• 1 with 1,000+ ratings in this pool — and it is a mortgage app the filter caught by accident, not a product in the niche. Among the twelve real listings the ceiling is 632 ratings.

The only thousand-plus listing in this candidate pool is a false positive, and the reviewed relevant products top out in the low hundreds. This does not establish who leads the market. Older incumbents are excluded because the catalog only holds apps first released inside the window; a market that looks leaderless in this pool may have a leader outside it.

Three misreadings to avoid

1. Reading monthly launch counts as a trend

Monthly counts in this snapshot rise toward the capture date: September 2026 has 47,174 first releases and March has 23,047. A currently available listing snapshot can be affected by removal and index coverage, so these counts cannot isolate a change in launch activity or demand. Do not present this monthly distribution as measured market growth. See the captured monthly counts.

2. Reading “AI” in the copy as an AI product

“AI” appears across the invoicing pool as a marketing vocabulary, not a capability. Invoice makers and photo estimators use the same word for different promises. The sentence to look for is not “powered by AI” but what the AI is claimed to decide — the price, the scope, the material list — because that is where the product's real risk and differentiation sit.

3. Reading a free download as a free product

97.6% of listings in the full snapshot are free to download. That figure does not reveal whether later steps are paid. In-app pricing and subscription terms are not captured here, so the listed download price cannot establish the business model.

A worked example, start to finish

The estimating source analysis applies this method: pool definition, the sixteen selected listings, four exclusions, workflow patterns and explicit limits. To use that evidence in a decision, follow the SaaS direction screening guide: define a premise, compare counterexamples and write the next research task.