How to build an AI identifier app: method, numbers and risks

As of 21 October 2026, an AI identifier app (the kind that names a plant, a coin, a rock or an antique from a photo) is one of the few App Store niches where an unknown developer can still reach $10,000 a month, but it is not an easy one. Of the 678 identifier apps that have at least 20 ratings on the US App Store, 60 (8.8%) earned $10,000 or more in September 2026, according to Victoire's estimates of worldwide gross revenue. 605 (89%) earned less than $5,000.

The money does not sit where you would expect. Plants carry the most revenue, but one app takes almost all of it. Collectibles (coins, antiques, banknotes, stamps, trading cards) are where a new app has the best odds. And a large part of the market belongs to a handful of studios that build one identifier per subject, on the same engine, again and again. This article counts all of it, explains the economics, and gives a method to launch one without walking into the traps.

What is an AI identifier app?

An identifier app answers one question from a photo: "what is this?". The user points the camera at a leaf, a coin, a mushroom or a vase, and the app returns a name, a description and, very often, a value or a care guide. The recognition itself is done by an image model, either trained by the developer or bought as an API; the rest of the app is the listing, the paywall and the content around the answer.

The niche is older than the current wave of AI. PictureThis, the plant identifier that still dominates the category, was released in 2017. What changed with multimodal models is the cost of entry: identifying a subject no longer requires years of labelled photos, so new identifiers appear every week, for subjects nobody covered five years ago.

The apps fall into four families, and they do not behave the same way:

• Nature identifiers. Plants, mushrooms, insects, birds, fish, snakes, dog and cat breeds. This is the largest family, with 300 apps in our count. Users are gardeners, hikers and curious parents, and the question is mostly "what is it and is it dangerous?". Demand is seasonal, and the leaders are old and very well rated, which makes the top of search hard to reach.

• Collectible identifiers. Coins, antiques, banknotes, stamps, trading cards and comics: 173 apps. Here the question is "what is it worth?", and the answer has a price tag, which changes everything about willingness to pay. A user who thinks a coin might be worth $200 is far more likely to accept a $39.99 annual plan than one who just wants the name of a flower.

• Rocks and crystals. 71 apps sitting between the two: part nature, part collection. Revenue is concentrated in one app, Rock Identifier: Stone ID, which makes about $600,000 of the subject's $640,000.

• "Identify anything" apps. 51 general-purpose scanners that promise to recognise any object. They compete directly with Google Lens, which is free and built into Google's apps, yet six of them make $10,000 or more a month, several with weekly plans between $5 and $10.

What are the key figures?

• 8.8% of identifier apps make $10,000 a month or more. That is 60 of 678 apps with at least 20 US ratings. 21 of them (3.1%) make $100,000 or more. At the other end, 605 apps (89%) are under $5,000.

• One app takes almost half of the money. The apps above $5,000 make about $13.4 million a month together. PictureThis alone makes about $6 million of it, and the top five apps take 64%.

• Collectibles have the best odds. 13.9% of collectible identifiers (coins, antiques, banknotes, stamps, cards) make $10,000 a month, against 8.0% of nature identifiers. Nature makes more money in total ($9.4 million against $2.4 million), but PictureThis alone accounts for $6 million of it.

• New apps still get in. Of the 326 identifier apps released since January 2025, 23 (7.1%) reached $10,000 a month, and three reached $100,000: AntiqSnap, PlantAID and Lens AI.

• Identifiers earn more per download than the average app. Among the winners with a download estimate, the median is $4.00 of revenue per download in September 2026. In our study of all apps launched between October 2025 and March 2026, the same median was $1.50.

• Studios own most of the market. 101 publishers have two identifier apps or more. They publish 286 of the 678 apps and make 83% of the revenue. One of them, Next Vision Limited, has eight apps above $10,000 a month.

• A rating under 4.0 is fatal. None of the 99 identifier apps rated under 4.0 in the US reaches $10,000 a month. Between 4.5 and 4.79, 14.9% do.

• $39.99 a year is the default price. 40 of the 56 winners whose price list we could read sell an item at $39.99, and 29 of them offer a weekly plan.

How did we count these apps?

We took every non-game app on the US App Store whose name contains "identifier", "identify" or "identification" and that had at least 20 US ratings on 21 October 2026. That gives 693 apps. We removed 15 that identify songs or phone callers, which are a different business, and kept 678. We then sorted each app into a subject from its name: plants, coins, rocks, antiques, insects and so on.

For each app, we read Victoire's estimate of its worldwide gross revenue and downloads for September 2026, and for August 2026 when it exists. These are rounded estimates: under $5,000 a month they are only given as "under $5,000", above that they move in steps. They give an order of magnitude, not an exact figure, and they are gross: before Apple's commission and before costs.

The method has two blind spots. Apps that do not have "identifier" in their name are missed: Google Lens, Merlin Bird ID, Seek by iNaturalist and Pl@ntNet, for example, are not in the count, which is acceptable here because none of them is a subscription identifier of the kind this article studies. And the 20-rating bar excludes the many identifiers that never got any traction, so the real failure rate among all launches is higher than 89%.

How much does an identifier app make?

Most make almost nothing: 89.2% of identifier apps are under $5,000 a month, and 8.8% reach $10,000 or more. The distribution looks like this:

The shape is the usual App Store shape, with a long tail of apps that make almost nothing and a short head that makes almost everything. What stands out is the size of the head: 21 apps above $100,000 a month in a niche of 678 is a high proportion. In our study of all apps launched between October 2025 and March 2026, only 1.1% reached that level. Identifiers that work tend to work at scale, probably because the same funnel (search, scan, paywall) can absorb a lot of paid traffic.

Here are the twelve identifier apps with the highest estimated revenue in September 2026:

Three things in this table matter for anyone planning to launch. First, the leader is in a league of its own: PictureThis makes more than the next eleven apps combined. Second, seven of the top twelve come from two publishers: Next Vision Limited (CoinSnap, Rock Identifier, AntiqSnap, Picture Insect, Picture Mushroom) and Vortemol Limited (PlantIn, CoinIn). Third, one of them is barely a year old: AntiqSnap, released in October 2025, already made about $400,000 in September 2026.

Which subjects make money?

The subject decides most of the outcome. Some subjects have one giant and many failures, others have no winner at all, and a few still share their revenue among several apps. Most identifiers sit in the Education and Reference categories of the App Store (501 of the 678); in our study of how much apps make in their first year, Reference was the third-best category for new apps, with 11.7% of them at $10,000 a month. For the ranking of every category, see which app categories make the most money.

The table reads in three groups.

The crowded subjects with a giant: plants and insects. Plants have the most apps (179) and the most money, but PictureThis makes about 70% of the $8.6 million that plant identifiers earn, with more than a million US ratings. Still, 18 other plant apps clear $10,000 a month: Plantum, PlantIn, Plantiary, Plantaria, Botan's Plant Identifier & Care App. They share the long tail of searches the leader does not cover, most likely with paid traffic on top. Insects are worse: 38 apps and a single winner.

The value subjects: coins, antiques, banknotes, mushrooms, birds. Here the share of winners reaches 12% or more. Coins and antiques combine a large audience (collectors, people clearing out a relative's house, thrift-store resellers) with a question that has a monetary answer. Banknotes and birds have a high rate on very few apps: two winners out of ten or eleven is encouraging, but far too few to read as a rule. Mushrooms are a special case, covered in the risks section, because a wrong answer can lead to a poisoning.

The empty subjects: dog breeds, fish, snakes, jewelry, animals in general, wood, cars, pills. 97 apps across these subjects and not one reaches $10,000 a month. The likely reasons differ. Dog breeds and generic animals are curiosity questions that people ask once, then never again. Jewelry and watches sound like value subjects, but an app cannot weigh gold or check a hallmark from a photo, which limits what it can honestly promise. Pills and snakes carry a safety risk that makes them harder to sell and to get through review.

The lesson is not "pick the subject with the highest rate". It is to pick a subject where the question has a price and comes back often. A collector photographs dozens of coins; a gardener asks about one plant a month. That frequency is what turns a $39.99 annual plan into a habit rather than a one-off purchase that gets cancelled or refunded.

Who builds these apps?

Mostly studios, and the most successful ones industrialise the format. The 101 publishers with several identifiers account for 42% of the apps and 83% of the money. Of the 60 apps above $10,000 a month, 38 come from publishers with four apps or more on the US store, and only 10 from publishers with a single app.

The clearest example is Next Vision Limited, which has 23 apps on the US App Store. Ten of them are identifiers in our count, eight make $10,000 a month or more, and together they make about $2.4 million a month. The pattern is visible in the names: Picture Insect, Picture Mushroom, Picture Bird, Rock Identifier, then a "Snap" series: CoinSnap, AntiqSnap, NoteSnap for banknotes, ComicSnap for comic books. Same interface, same paywall logic, a new subject each time. Vortemol Limited does the same with PlantIn, CoinIn, RockIn and SportDex, and AIBY with Plantum, Coin ID and Rock ID.

This matters for an independent developer in two ways. The good news is that these studios have proved that a new subject can be opened from scratch: AntiqSnap went from zero to about $400,000 a month in under a year. The bad news is that if your subject works, they can enter it quickly with an engine that is already tuned. Your advantage has to come from what these factories do badly: depth in one subject, a real community, content written by someone who knows coins or minerals, and a brand that collectors recognise.

The ten single-app publishers above $10,000 a month show it is possible. Botan's Plant Identifier & Care App makes about $300,000 a month, Plantaria about $200,000, Seapptic's AI Scanner about $80,000. Sporecast, a mushroom identifier, made about $30,000 in September with only 23 US ratings, which suggests most of its users are outside the United States.

How did AntiqSnap reach $400,000 a month in under a year?

AntiqSnap is the best recent case because it shows the whole playbook on a subject that was not crowded. It was released on 15 October 2025 by Next Vision Limited. In September 2026, eleven months later, Victoire estimates it at about $400,000 of worldwide gross revenue and 200,000 downloads, with 31,451 US ratings.

The last two lines deserve attention. AntiqSnap collected its 31,000 ratings early and now gains a few per day, while its revenue stayed flat at about $400,000 between August and September. Reviews are not what keeps it alive; search visibility, a working paywall and probably paid acquisition do. That is consistent with what we measured in apps with few reviews but high revenue: ratings are a poor proxy for money.

What could break the model? First, competition: since AntiqSnap's launch, at least three new antique identifiers have reached $20,000 a month or more (AntiqueValue in March 2026, AntiqueID in April, Worth It AI in April), so the subject is filling up. Second, the quality of the answer: valuing an antique from a photo is guesswork, and a user who feels misled leaves a one-star review or asks Apple for a refund. Third, Apple: a studio that publishes ten near-identical apps is exactly what guideline 4.3 on spam targets, and a rejection or removal would hit the whole series.

Can an independent developer still win?

Yes, and the clearest proof is ThriftAI: Profit Identifier. It was released in June 2025 by smallstack ApS, a small Danish publisher with five apps. It made about $60,000 in September 2026, with 20,673 US ratings, and it was #5 among top-grossing Shopping apps in the US on 21 October.

ThriftAI did not copy the format; it shifted the question. A classic identifier asks "what is this object?". ThriftAI asks "how much profit can I make if I buy this at a thrift store and resell it?", and answers with a value, the demand and a buy-or-skip recommendation. The target is a precise group (resellers who buy at thrift stores, flea markets and estate sales) who can scan dozens of items in a single trip and treat the subscription as a work tool. Its keyword, "thrift scanner", has a difficulty of 19, and ThriftAI ranks first on it.

This is the pattern to look for among the 326 identifiers launched since January 2025. The ones that reached $10,000 a month show three routes:

• AntiqSnap: Antique IdentifierAntiques, released Oct. 2025$400k
• PlantAID:identifier&care guidePlants, released Feb. 2025$100k
• Lens AI: Identify AnythingAny object, released Apr. 2025$100k
• ThriftAI: Profit IdentifierResale profit, released June 2025$60k
• Bird Identifier: BirdScopeBirds and songs, released Apr. 2025$40k
• Sporecast: Mushroom IdentifierMushrooms, released July 2025$30k

• Open a subject the studios had not covered yet. Antiques in late 2025, comic books, sports cards, banknotes, grass species. The first credible app on a new subject gets the search traffic before the factories arrive. TCG Card Identifier Value, published under an individual's name, makes about $20,000 a month with collectible card games.

• Change the question. ThriftAI turned identification into a resale calculator. Worth It AI does the same for clothing, sneakers, handbags and electronics, with an estimated market value for each item. Bird Identifier: BirdScope identifies birds by their song as well as by photo. The scan stays the same, the promise changes, and so does the audience.

• Out-execute on an old subject. PlantAID makes about $100,000 a month in plants, the most crowded subject of all, with only 434 US ratings. Its price list is built around weekly plans between $3.99 and $5.99, next to annual plans at $49.99 and $59.99. Plant Identifier: Bloom, released in February 2026 by a publisher with a single app, already makes about $30,000. With so few ratings, these apps are not winning on reviews; their paywall and, most likely, paid acquisition do the work.

The base rate is sobering, though: 7.1% of identifier apps released since 2025 reached $10,000 a month, against 10.5% of older ones. The younger apps have had less time, but the niche is also more crowded than it was. A new identifier needs a reason to exist beyond "the same, with AI".

How much should an identifier app cost?

The category has converged on $39.99 a year, and the safest choice is to use it. We read the in-app purchase list on the App Store page of 56 of the 60 apps above $10,000 a month:

• 40 sell an item at $39.99, and in 24 of them that item is explicitly named as a yearly plan. PictureThis Pro, Rock Identifier Premium, AntiqSnap Premium, Picture Insect Premium, Picture Bird Premium, ChatCoin, CoinValue, AntiqueID: all at $39.99.

• 29 have an item explicitly named weekly, usually between $3.99 and $9.99. Weekly plans are the main lever of the "identify anything" scanners: Lens Scan sells $8.99 a week next to a $79.99 year, LensAI between $5.99 and $9.99 a week.

• 8 sell a lifetime purchase, between $19.99 and $199.99. PlantIn's lifetime access is listed at $49.99.

• None charges for the download. All 60 winners are free to install.

Two structures dominate. The "annual first" structure (PictureThis, Next Vision's apps) is built around a $39.99 year, with cheaper short plans at $3.99 or $7.99. The "weekly first" structure (Lens Scan, PlantAID, many plant apps) puts forward a weekly plan that looks cheap and adds up to roughly $200 to $520 a year. The second can earn more per paying user, and it carries a higher risk of refund requests and angry reviews. If your subject has a real recurring use (collectors, resellers), the annual plan fits; if the use is a one-off curiosity, weekly plans are how the category monetises it, with the reputational risk that goes with them.

For more on price levels across all categories, see our study of app subscription prices.

What does an identification cost you?

Each scan costs money, which is the main difference between an identifier and a classic utility app. You have three options:

1. A specialised identification API. Kindwise (plant.id, insect.id, mushroom.id) charges one credit per identification, from €0.05 per credit when you buy 1,000 credits down to €0.01 at 1.5 million credits, with 100 free credits to start. Pl@ntNet, run by a consortium of French public research institutes (Cirad, INRAE, Inria, IRD), offers a free plan of 500 identifications per day and a paid plan billed by volume, from €5 per 1,000 identifications down to €2 at very high volumes.

2. A general multimodal model. You send the photo to a large model with a prompt that asks for the species, a confidence level and a description. It covers any subject, including the ones no specialised API handles (antiques, comics, banknotes), but its answers on fine distinctions are less reliable and the cost varies a lot by model.

3. Your own model. Only worth it at scale, or if your subject is so specific that no API and no general model handles it well. It needs thousands of labelled photos per class and a way to keep them up to date.

A worked example. Suppose a user scans 30 items a month and you pay €0.02 per identification: that is €0.60 a month, about €7 a year. Against a $39.99 annual plan (about $34 after Apple's 15% commission under the Small Business Program, open to developers with up to $1 million of proceeds in the previous calendar year), the margin is comfortable. The real cost is the free users: if 95% of your users never pay and each makes ten free scans, their scans are paid out of the 5% who do. That is why so many identifiers limit free scans or put the paywall right after the first result. Antique Identifier - Relic even names its plan "Pro: Unlimited Scans".

The other economic fact is revenue per download. Among the winners with a download estimate, the median is $4.00 per download in September 2026, with a quarter of them at $8.00 or more. That is far above the $1.50 median we measured for all apps launched between October 2025 and March 2026. An identifier with a working paywall can afford to pay for installs: if each download is worth $4 in its first month and an Apple Search Ads install on your keyword costs less than that, the funnel finances itself. Measure both numbers before spending, because the $4 median comes from apps that already work.

Do identifier apps depend on the season?

Some subjects do, sharply. Comparing August and September 2026 for the apps that have both months:

• Plants fell by 28%, from about $11.8 million to $8.5 million across 61 apps. PictureThis alone went from about $9 million to $6 million. The end of the gardening season in the northern hemisphere is the likely explanation.

• Birds halved, from $290,000 to $140,000 across five apps, and rocks lost about a quarter.

• Mushrooms rose by 60%, from $206,000 to $330,000 across eight apps. September and October are mushroom season in Europe and North America. Sporecast went from about $6,000 to $30,000 in a month.

• Coins and antiques barely moved (about −2% and −3%). Collecting does not follow the weather.

Two months are not a full year, so these figures describe a direction, not a precise seasonal curve. But the direction matters for planning: a plant app launched in October will measure its first weeks while plant revenue is falling, and a mushroom app launched in November misses its season entirely. Collectibles give you a steadier base, which is another reason they suit a small developer who needs to read results quickly.

How to build an AI identifier app in 6 steps?

1. Choose a subject where the answer has a price or comes back often. Start from the subject table above: coins, antiques, banknotes, stamps, trading cards, or a question that changes the use, like ThriftAI's resale profit. Avoid subjects with no winner at all (dog breeds, jewelry, fish) unless you have a precise reason to think the market is wrong. Check the markets page to see which countries leave room for a small app.

2. Check the keyword before writing code. Look up the difficulty of your subject followed by "identifier" ("coin identifier", "stamp identifier") on the keywords page. Below 30, a new app with a good title can reach the first results ("stamp identifier" is at 26, "banknote identifier" at 24, "thrift scanner" at 19). Above 55, you are fighting a giant ("plant identifier" 64, "card scanner" 78, held by vertical companies like Collectr and CollX). Then see who is in the top three, how old they are, and how many ratings they have.

3. Buy recognition, build the answer. Plug a specialised API if your subject is covered (plants, insects, mushrooms), or a multimodal model otherwise. Spend your effort on what comes after the name: the value range with its sources, the care guide, the history of the object, the next action (sell, keep, insure). That is where a small team can beat a factory, because it needs knowledge of the subject, not just an engine.

4. Put the paywall where the value is. Show it right after the first result, when the user has seen the answer and wants the rest (value, details, history), with a short free trial, a $39.99 annual plan and a weekly or short plan. ThriftAI, for example, announces a three-day free trial in its description. Keep one or two free scans so the user sees the answer before paying, and cap free use so free users do not eat your API budget. Read our subscription pricing study before testing other prices.

5. Build the listing for search and for other countries. Put the subject and "identifier" in the title, show the scan in the first screenshot, and translate the title and screenshots early. In our count, 13.9% of identifiers in ten languages or more reach $10,000 a month, against 6.2% of single-language apps. AntiqSnap, listed in France as "AntiqSnap: Identifiant antique", has more than a thousand French ratings.

6. Protect the rating and measure for 30 days. Ask for a review only after a successful identification, never after a failure or a paywall. Answer one-star reviews about wrong answers. After 30 days, compare your revenue per download with the $4 median of the winners, and follow your app on Victoire's trends page next to its competitors. If you are far below after a month, change the subject or the question rather than polishing the same app.

What are the advantages of an identifier app?

• The demand is explicit. People type "coin identifier" or "what is this plant" into the App Store because they have the object in hand. You do not need to create the need, only to be found when it appears, which makes App Store search optimisation worth more than in most categories.

• The core feature is visible in one screenshot. A photo, a name, a value: the promise is understood in two seconds, both on the listing and in a short video ad. That makes paid acquisition and organic videos (an "I found this at a garage sale, the app says it's worth…" format) easier to produce than for an abstract app.

• The technology is now a commodity. Recognition APIs and multimodal models give an independent developer, for a few cents per scan, what used to require a research team. Your effort goes into the subject, the content and the funnel, which are things a small team can do well.

• Revenue per download is high. A median of $4.00 per download among the winners, against $1.50 for all apps launched between October 2025 and March 2026, means each install is worth more, and paid acquisition can pay for itself sooner. That gives you a way to grow that does not depend only on luck with reviews or virality.

• The subjects are almost unlimited. Coins, stamps, comics, sneakers, vinyl records, tools, fossils, seashells: every collection and every hobby can have its identifier. The studios cover the big subjects; dozens of smaller ones have no credible app yet, and some of them have communities that would pay.

What are the risks of an identifier app?

Apple's spam rule (guideline 4.3)

Apple's App Review Guidelines refuse apps that are "indistinguishable from what's already widely available" and warn against "opportunistically creating variants of existing app categories". An identifier with a generic name, a template interface and a recognition API behind it is exactly what a reviewer may flag as spam, especially on crowded subjects like plants. The same rule forbids publishing several Bundle IDs of the same app, which is a risk for anyone tempted to copy the studio model with one app per subject. The defence is to make the difference visible in the screenshots and the review notes: a specific subject, a specific answer (value, care, resale), real content.

Wrong answers that can hurt someone

For some subjects, a wrong answer is not just a refund. A 2023 study in Clinical Toxicology tested three mushroom identification apps on photos of 78 specimens sent to the Victorian Poisons Information Centre and Royal Botanic Gardens Victoria in Australia in 2020 and 2021. The best app, Picture Mushroom, identified 49% of them correctly, and recognised 44% of the poisonous ones; Amanita phalloides, the death cap, was falsely identified twice by that app. The authors conclude that these apps are not reliable enough to rule out a poisonous mushroom when used alone. In France, Anses explicitly advises against relying on recognition apps, because of their high risk of error, and recommends having a harvest checked by an expert.

The same logic applies to snakes, pills, skin rashes and edible plants. Apple's guideline 1.4.1, written for medical apps, says those that could give inaccurate information "may be reviewed with greater scrutiny" and expects them to remind users to consult a doctor; a rash, pill or snake identifier is in the same position. If you choose one of these subjects, the app must never present an identification as a safety guarantee, and the warning must be in the product, not hidden in the terms.

Valuations that disappoint

Collectibles have the best odds partly because they promise a value, and that promise is fragile. A model can recognise a 1921 Morgan dollar, but not its grade, its condition or whether it has been cleaned, and those details can multiply or divide the price. A user who tries to sell at the price the app gave and fails will feel misled. Present ranges, show sources (recent auction or marketplace sales), and say clearly what the photo cannot tell. That is both more honest and better for your rating: under 4.0, it is fatal in this category.

Weekly plans and refunds

Weekly plans earn a lot and anger a lot of users. An app that pushes $8.99 a week to someone who wanted to identify one plant will collect refund requests and angry reviews. Apple's guideline 3.1.2 asks subscription apps to describe clearly what the user gets for the price before asking them to subscribe. A weekly plan works best when it is offered next to a clear annual plan, not instead of it, and when the free trial is not designed to be forgotten.

Seasonality and the studios

A plant or bird app can lose a quarter to half of its revenue between two months without doing anything wrong, as the August-to-September 2026 comparison showed. Plan your launch at the start of the season, and do not judge the app on its worst month. The studios are the other structural risk: if your subject turns out to be profitable, a company with twenty identifiers can enter it within weeks with a better-tuned funnel. Your protection is depth in the subject and a community they cannot copy quickly.

Is there a market for identifier apps in France?

Yes, with French specifics that change the strategy. Keyword difficulty in the French App Store is lower than in the US on most subjects:

• Plants: a free public competitor. On "identifier plante" (difficulty 46), the first result in France is Pl@ntNet, the free app built by French public research institutes, with 9,925 French ratings, ahead of PictureThis, which has 70,394. On "plante" (67), PictureThis comes first. A paid plant identifier in France competes with a free, scientific and well-liked app, which is a hard pitch.

• Mushrooms: big demand, bigger risk. "Champignon" has a difficulty of only 26 and "identifier champignon" 16. Picture Mushroom leads with 6,291 French ratings, followed by Champignouf with 2,955. Mushroom picking is a popular autumn pastime in France, but Anses explicitly advises against relying on an app, which makes the subject a legal and ethical minefield.

• Coins, stones and antiques: open ground. "Identifier pièce" (28), "pièce de monnaie" (19), "pierre" (27), "minéraux" (18) and "antiquité" (13) are all reachable. On "pierre", Rock Identifier has 3,374 French ratings and the next app has 64. On "antiquité", the second result is the auction house Drouot, not an identifier.

• Localised titles win. The apps that do well in France use French titles: "PlantIn・Plantes reconnaissance" (10,538 French ratings), "Rock Identifier: Pierre, Gemme" (3,400), "AntiqSnap: Identifiant antique" (1,022). An identifier with an English-only title is invisible to someone typing "reconnaître une pierre".

The French opportunity is less "launch a French PictureThis" than "be the credible French-language app on a collectible subject": coins, stones, antiques or stamps ("timbre" is at 17), with content written for French collectors, flea markets and auction habits. To check the openness of a subject country by country, use the markets page before building.

Frequently asked questions

How much does an AI identifier app make?

Most make very little. Of 678 identifier apps with at least 20 US ratings, 89% made under $5,000 in September 2026. 60 (8.8%) made $10,000 or more and 21 (3.1%) made $100,000 or more, according to Victoire's rounded estimates of worldwide gross revenue. The leader, PictureThis, makes about $6 million a month on its own.

Which identifier app subject is the most profitable?

In total revenue, plants, because of PictureThis. In odds for a new app, collectibles: 13.9% of coin, antique, banknote, stamp and card identifiers make $10,000 a month, against 8.0% of nature identifiers. Dog breeds, fish, snakes, jewelry and pills have no app above $10,000 a month in our count.

Do I need to train my own AI model?

No. Specialised APIs like Kindwise (plants, insects, mushrooms) or Pl@ntNet (plants) and general multimodal models cover most subjects, at prices from €0.01 to €0.05 per identification for Kindwise, and with a free plan of 500 identifications per day for Pl@ntNet. Your own model only becomes worth it at scale, or for a subject no API handles well.

How much should I charge?

The category standard is $39.99 a year: 40 of 56 winners whose price list we read sell an item at that price. 29 also offer a weekly plan, usually between $3.99 and $9.99. No identifier above $10,000 a month charges for the download.

Is it too late to launch an identifier app?

Not on every subject. Of 326 identifier apps released since January 2025, 23 reached $10,000 a month and three reached $100,000. The ones that succeeded opened a new subject (antiques, comics, cards), changed the question (resale profit) or out-executed on paywall and acquisition. A generic plant identifier launched today starts with low odds.

Is it legal to identify mushrooms or plants with an app?

It is legal, but the responsibility is real. A peer-reviewed study found that mushroom apps misidentify poisonous species too often to be used alone, and Anses advises against relying on them. Apple's rules also scrutinise apps that could give inaccurate health information. Any identifier on a subject that can hurt someone must say clearly that the answer is not a safety guarantee.

Who are the biggest publishers of identifier apps?

Glority Global Group (PictureThis), Next Vision Limited (CoinSnap, Rock Identifier, AntiqSnap, Picture Insect, Picture Mushroom), Vortemol Limited (PlantIn, CoinIn) and AIBY (Plantum, Coin ID). Publishers with two identifiers or more make 83% of the niche's revenue.