GardenTrack
Join the beta

Point the camera, and it answers from your own notes

Live · This works today.

Three doors, one camera. The round microphone button on every screen takes a photo and sends it to the chat, naming the plant you are standing on. On the Add a plant screen, the camera beside the What is it? box photographs a plant you cannot name and opens a conversation about what it is — then offers to add it, and waits for you to confirm before anything is written. On the Ask screen, a photo of a plant that looks wrong goes with that plant's own journal and its sourced crop record to an AI that answers in plain words: what it sees, how sure it is, and which of your own notes it used. When it cannot tell, it says so and says what would settle it.

The question box, and when to use it

Above the camera button there is a box for your own question. It belongs to the photo you are about to take — type it, then take the picture, and the two go together. Press return and the camera opens, so you do not have to reach for the button afterwards.

Leave it blank and the question is “what is wrong with this?”, which is the right one most of the time. The box is there for when it is not: “is this ready to pick?”, “what is eating this?”, “should I prune this now?”

Since last time

If you have photographed this plant before, the answer opens with how it compares. Better, worse, or no real change — and what is different, in the words you would use: “the lower leaves were green in the earlier one and are yellow now”.

This is the part nothing else can do. Any app can look at a leaf. Only this one is holding the photograph you took three weeks ago of the same plant.

And sometimes it will tell you it cannot compare them. If last time you photographed the whole plant and this time you photographed the one leaf that worries you, those are two different amounts of a plant — and a plant does not get worse because the camera moved closer. Rather than guess, it says so, and tells you which picture would answer the question. That is the most useful thing it can say, because you can act on it in ten seconds.

It compares against a picture at least ten days old rather than your most recent one. Two photos three days apart show the light and the angle changing, and that is the commonest pair there is, because a worried gardener photographs a plant again the next morning. And “no real change” is a real answer, not an empty one — a plant that looks the same three weeks on is usually a plant that is fine.

What it actually does

You take a photo of something that looks wrong. GardenTrack sends three things together: the picture, that plant's own journal, and the sourced record for that crop. It comes back with

  • what it can see — in the words a gardener would use, not a pathologist's
  • what it thinks it is, and how sure it is
  • which of your own notes it used, with their dates
  • what to do, or an honest refusal to say

The middle one is the part nothing else does. Your notes are the reason the answer can be about your plant rather than about that species in general.

The photo also fills in what you cannot tell us

Most plants in most gardens are half-described, and not because anyone is lazy. Nobody knows how many gallons their pot is. A new gardener has no idea how tall a tree ought to be at that age. You would never think to come and tell an app there is mulch on the soil.

A picture knows all of that at a glance. So alongside the written answer, GardenTrack reads off what it can see and offers it to you:

  • What the roots are in — a tray, a pot, a raised bed, open ground. This is the single thing that decides how often it needs water. A greenhouse is not one of the four, because a greenhouse is a roof and the plant under it is still in a pot or a bed; that is a fact about the spot, and the spot has its own place to say it.
  • How big the pot is, roughly, in gallons.
  • Whether there is mulch on it — which changes how fast the soil dries.
  • How tall it is and how far along — seedling, flowering, fruiting.
  • What plant it is, when you are not sure yourself.

None of it is saved until you say so. You get one sentence and one button. If something it worked out is already on the plant, we do not ask you again — a fact you have already given us is not a question. If it disagrees with something on record, that one we do show you on its own, because one of the two is wrong and only you know which.

When it recognizes something you have not added

A line appears under the answer offering to add it, and tapping it opens the ordinary add-a-plant screen with the crop already chosen. It does not show up for a plant that is already in your garden — you are looking at it, we know.

It does not guess when you planted it. That question is asked, exactly as it always is.

We did look at whether the picture could narrow it down, because a plant in flower is plainly further along than a seedling. Every crop we hold records how long each stage lasts and what it counts from, so the information is there. The problem is how wide the windows are. A tomato flowers anywhere from 35 to 65 days after it goes in the ground — and that is the tightest link; working back from there to when it was sown widens it further. A date offered with a month of slack in it is this question with more confidence attached, and the planting date is the one that quietly decides every other date we work out for that plant. So it asks.

The camera sits next to Add a plant now, everywhere

Anywhere the app offers to add a plant — your daily list, your garden, a bed, a plant you already have — there is a camera beside the button, the same size and the same color. Photographing a plant is a way of adding one, not a separate feature buried on another screen.

Tap it and you get the same thing you would get from typing a name: a suggestion, and a card asking you to confirm before anything is written down. It behaves identically wherever you tap it, because it is one button used in several places rather than five that drifted apart.

If you already know what the plant is, typing is faster. The camera is for the pot you were given, the seedling somebody dropped off, and the thing in the corner that came with the house.

The things it will never guess

Some answers are worse than no answer, and these are the ones we have decided in advance rather than case by case.

When a plant went in, or how old a tree is. A photograph cannot know, and a confident wrong planting date does not stay wrong in one place — every stage, every reminder, every "will it ripen before frost" hangs off it, and a first-year tree and a third-year tree need different work. So the app asks you roughly how long instead. "A couple of months ago" is a real answer and a guess dressed up as one is not.

Which canes a bramble fruits on. Cut the wrong ones in winter and you remove next summer's entire crop. That one has a question you can actually answer, in late summer, when you can see it.

Where the answers live

Answers are kept in your account, not only on the phone you took the photo with. That means they survive a lost phone, show up on another device, and are still there next spring when you want to know what was wrong with that tree in August.

What it worked out is kept as dated readings rather than as a paragraph — the pot was five gallons and the tree was forty centimeters in April, ninety centimeters by August. Kept that way, a plant's record becomes something you can look back through rather than a pile of old text.

The rules it answers under

Six of them, and they are the same six the chat follows — one rulebook, so anything either side learns applies to both. It will never name a chemical, a brand or a dose; never tell you when something was planted or how old it is unless you told it; never say how a plant is doing without seeing it; never contradict you about a plant you are standing in front of; never claim to have saved something it has not; and never talk about how the app works instead of about your plant.

The one that matters most is the one that is not a rule at all: refusing is a correct answer. If the photo does not show enough, it says so and tells you which picture would settle it. A confident wrong answer is worse than no answer, because you act on it — you dig something up, or you put something on food you are going to eat.

When we have written the long version

An answer on a phone is four or five sentences. Some problems need four or five pages — how to actually find a hornworm at nine in the morning, why spraying the squash bugs you can see does not work, which of the six causes of a yellow tomato leaf you are looking at.

When the answer names one of those, a single line appears under it pointing at our page on it. Four problems have one today, and that number is the honest one: tomato hornworm, squash bugs, yellow tomato leaves and yellow cucumber leaves. Everything else gets no link.

The order matters more than it sounds. The app never shows the model a list of problems and asks which one this is — hand anything a list of diseases and a healthy plant, and it will pick one. It answers first, on its own, and only then does the app go and look up whether we happen to have written about what it said. If the answer is “nothing is wrong”, there is nothing to look up and no link appears.

When the answer is “nothing is wrong”

That is the most common answer there is, and for a while it was the one the app did least with. We ran fourteen real garden photos through it and about half came back as some version of this is normal — spent flower heads, a seed head, new growth coming in pale, a plant that is small because it is growing under a tree.

Those are the photos people take. You do not photograph a plant because it looks fine; you photograph it because something looks off and you want to know whether it matters.

So it now offers to write that down too. One tap and “spent blooms, not a disease” goes into that plant's story with the date on it. Next August, when the same brown heads look alarming again, the answer is already there and you do not have to ask twice.

How carefully we read the answer

Changed on 18 August 2026, and nothing about it is visible on screen. When you ask about a photo, the model's reply gets written into that plant's record. The app used to find the useful part of that reply by searching it for something that looked like a list — which worked, and was a guess.

The reply is now required to come back in an exact shape, so there is nothing to search for, and anything outside that shape is refused before it reaches your journal. We deleted the old searching code rather than keeping it as a backup: two ways of reading one answer is worse than one, because the second only ever runs when the first has already been surprised.

If an answer cannot be read at all, it costs you the look and nothing else. That has always been true and has not changed — a reply we do not understand never becomes a record about your plant.

Common questions

Can a photo tell you how tall my tree is, or how old it is?
**How tall, yes — how old, no, and that one is on purpose.** From 25 August 2026, if a picture shows us roughly how tall a plant is, we offer it as something you can tap to confirm, and confirming it writes the height onto the plant. It reads in feet and inches, and it says *about* — it is an estimate from a photograph, not a tape measure. **Age we will not guess, ever.** A photograph of a tree with nothing beside it for scale cannot tell you whether it is three years old or six, and every date we work out for that plant hangs off its age — when it should fruit, whether it is old enough to prune hard, what its first season should look like. A wrong age there is wrong in a way you would have no way to spot. So it stays a question we ask you, and "I bought it at about this size" is a perfectly good answer. The same rule covers when it was planted, and whether a cane fruits on old wood or new. Those three we ask; everything else a picture can honestly see, we offer.
Does it know when I planted the thing in the photo?
**Only if you told us, and from 24 August 2026 it is told which of those it is.** When we hold a planting date, the photo answer gets it — that is what lets it check the stage it can see against your own calendar. When we do not, it used to get **nothing at all**: not a date, and no mention that there was no date. A gap where a fact should be is the easiest thing in the world to fill in, and filling it in is precisely what we do not want. It is now handed *date not recorded* in plain words, which is a fact rather than a hole. **Knowing when it went in is still not knowing how old it is**, and it will not tell you that either way — a tree bought at three years old went in the ground last month, and nothing we hold can tell those two apart. **One thing came back with this fix.** How a plant was started — sown where it stands, or raised somewhere and moved — was only being passed along when a planting date came with it. That is one of the more useful facts there is when something looks wrong at the base of a young plant, because transplant shock and damping off look alike and start differently. It now goes with the photo whether or not we know the date.
Do I have to go to the Ask screen to photograph something?
**Not any more.** The round microphone button in the corner of every screen has a camera in it. Standing at a plant, tap it, tap **Take a photo**, and send — the box already knows which plant you are looking at and says so, so the answer comes back about that plant. **A picture with no words is a complete question.** The Ask screen is still there and unchanged; this is a shorter way in from wherever you are standing. See [Talking to your garden](/help/talking-to-your-garden/).
There are two little buttons by the message box. Which is which?
**The left one opens your camera; the right one opens the photos already on your phone.** Until 18 August 2026 neither of them said so — not a short name or a confusing one, but no name at all, so anyone using a screen reader heard "button" twice and had to press one to find out. **Nothing looked wrong**, which is exactly why it lasted: the message box rendered precisely as designed and survived every visual review we did. It was only found by pressing the buttons rather than looking at them. They now say what they do, and they grew from 36 to 44 pixels across — the same minimum every other control in the app already used, which matters most here because they are the closest-together pair we have.
Can it tell me what a plant IS, not just what is wrong with it?
**Yes, as of 14 August 2026, and it starts on the Add a plant screen.** There is a camera beside the *What is it?* box: photograph the plant and it opens a conversation with your picture already in it, asking what it is. You do not have to type a name you do not know. **It answers in the same voice as everything else here** — it will say what it thinks and what it read to get there, and if it genuinely cannot tell, it says which two it is choosing between and what would settle it. A leaf laid flat, the underside, a fruit cut through the middle. **This is a bigger deal than it looks, which is why it took a decision to ship it.** A wrong diagnosis gets checked against the plant standing in front of you and corrected in a minute. **A wrong identification becomes the record** — every date, task and frost warning after it is worked out for a plant you do not own, and nothing about that looks broken from the outside. It is quietly wrong all season.
Will it add the plant to my garden by itself?
**No, and it is built so that it cannot.** The camera opens a conversation; it has no way of its own to write anything. When the app has an answer it offers to add the plant, and that offer arrives as a card you read and confirm — the same card that already appears when you tell the app about a plant in words, showing the crop, which bed, and when. **Nothing reaches your garden until you tap it.** That is deliberate rather than cautious. We would rather ask you a question you can answer in a second than write a confident guess into a record you will be reading in October. **One photo, one plant, for now.** If you photograph a bed with four things growing in it, expect it to talk about the most obvious one. Handing back four plantings from one picture is a different screen and a different set of questions, and it is worth doing properly rather than quickly.
It told me my plant looks fine. Is that a real answer, or did it just fail to work out what is wrong?
**It is a real answer, and it is the commonest true one.** Tested against fourteen photographs from a real garden, about half of the honest replies were a version of *this is normal* — spent blooms, a seed head, new growth coming in pale, the oldest leaves yellowing with age. **People photograph a plant because it looks odd to them, not because it is sick**, and those are two different things. **This needed fixing, and it was fixed on 13 August 2026.** Before that the answer was framed, in effect, as find what is wrong — so on healthy plants it found something anyway. A perfectly good sunflower came back with a fungal leaf spot. A healthy corn plant was said to be losing out to a nearby tree's roots. **Fluent, specific and wrong**, which is the worst shape a wrong answer takes, because nothing distinguishes it from the ones that are right. **There are three answers now, not two.** It can tell you what is wrong; it can tell you nothing is wrong and say what you are actually looking at; or — when it genuinely is not sure — **it can name what it might be and tell you what to look for.** **That middle answer is the one we care most about getting right.** *"I cannot tell, send me a better photo"* is the app asking you to do work for its benefit. So an unsure answer now has to carry three things: **the likeliest candidate named**, in *might be* language; **the signs on your own plant that would confirm it or rule it out**; and **what to do if it turns out to be that**. Something you can act on this evening, rather than a shrug. In practice that reads like: *this might be peach leaf curl — watch the new growth as it comes in, puckered and thickened and often reddish rather than just pale, worst after a cool wet spring.* **None of it makes the app slower to spot a real problem**, and it is not permission to hedge something obvious. It was never being too cautious; it simply had no way to say a plant was fine, and no shape for saying it was unsure. **The same damage is not the same problem on two different plants**, and what decides it is what you are growing the plant for. A leafminer tunnels between the two surfaces of a leaf, where nothing sprayed on the outside can reach it. On spinach or lettuce that is the crop ruined — you cannot cut around a tunnel and serve the rest. On a sunflower or a tomato it is a few marked leaves on a plant carrying hundreds, and it changes nothing about what you pick. So the first gets told to you as a problem and the second gets mentioned as something worth knowing, with nothing to do about it. **And "fine" comes with its reason.** *The lower leaves are yellowing, which is what a tomato does to its oldest leaves all season* is worth more than the word healthy on its own, because next August it will look just as alarming.
I typed what the nursery tag said and it made the whole thing the variety name.
**Fixed 20 August 2026, and the reason it matters is not tidiness.** A tag reading *apricot Mormon semi dwarf* is two facts: the variety is **Mormon**, and **semi-dwarf** is the rootstock — the root system the tree was grafted onto. The app used to store all of it as the variety name. **The rootstock is a field we already had, and it decides real things:** how big the tree finally gets, how far apart to space it, and whether it will throw up shoots from the base at all. Buried inside a variety name, none of that can be read, so the advice you get about spacing and suckers would have been wrong for the life of the tree. It now splits the tag and files each part where it belongs. **Two things it deliberately will not do.** It will not guess: a tag that names no rootstock leaves the field empty rather than assuming a standard one, because we would rather know nothing than record something the app will act on. And it only splits for plants that actually have a rootstock — the eleven grafted tree fruits and citrus we hold. **Your lavender is called Munstead Dwarf and that is its real name**, so nothing touches it.
I photographed a plant I do not own and it asked which of my plants to log it against.
**It should not have, and it no longer does — fixed 20 August 2026.** The app had just told you what the plant was; offering to file it against your unrelated plants is the app ignoring its own answer. Now there are three cases and each behaves. **If the photo matches one plant you own**, the note goes there and you are not asked. **If it matches a plant you have several of** — two sowings of the same bean — you are offered *only those two*, which is a question worth asking rather than a guess. **If it matches nothing you grow**, no plants are offered at all: the only sensible thing to do with a plant that is not in your garden is add it, and that button is already on screen. **We did not simply shorten the list.** Five unrelated plants would be the same mistake, only quieter — the question is whether a plant could be the one you mean, never how many fit on a row. **And a second fault came out with it.** With two plantings of the same crop, the app used to pick the earlier one and write to it without asking. A record filed against the wrong plant looks exactly like one filed correctly, so nobody would ever have noticed.
I picked a plant to log the note on and afterwards there was no note. What happened?
**You did everything right and the app was waiting for a second tap it never told you about — fixed 20 August 2026.** When the app could not tell which plant the note was for, it showed your plants as a row of buttons. Tapping one did not save anything: it swapped that row for another button, in the same place and the same shape, and waited. **So answering the question looked exactly like finishing it.** Choosing the plant now writes the note there and then, and if it cannot save, it says so on that same row instead of leaving you to find out later.
I photographed my seedlings and the app was still asking whether they had come up.
**It should not have been, and it no longer is — 21 August 2026.** When the app is not sure whether a sowing has come up, it puts a short question on your Today list. A beta tester photographed his bean pot, and the answer was exact: *the one on the left has its first true leaf open, the one on the right is just breaking the surface.* **The app had answered its own open question and did not notice.** The reason was smaller than it looks. The photo's reading was recorded correctly against the right plant — but it recorded *first true leaf*, and the question was waiting for *first seed leaves*. Nothing joined the two, even though a plant with a true leaf has certainly come up. **Later signs now prove the earlier ones**, so photographing a seedling closes the question about whether it sprouted, and a photo of coloring fruit settles that it flowered and set. **Only where it is certain.** A plant can grow all season and never flower, so being knee-high proves nothing about flowers, and the app will still ask. **And it does not rewrite your record.** Only what the photo actually said is stored; the rest is worked out when it is read, so a year from now you can still tell what was seen from what was inferred.
The note says "Diagnosis" or "Suggested" now. What are those?
**They say what kind of thing the note is, which it used to leave you to guess — 21 August 2026.** A note reading *nitrogen or micronutrient (iron/manganese)* could be three different things: something we spotted, something we advised, or something you already did. **A season later that difference is the whole point of having written it down.** So the note now opens with which it is — **Diagnosis**, **Suggested**, **Normal for now**, or **Looked healthy**. **We also took the app out of the sentence.** Notes used to end with *from what you told the chat*, which is about our plumbing rather than about your plant. Whether a note came from you or from the app is still recorded, and shows as a small **read by the app** label beside the entry — it just is not part of the sentence you read any more.
What does the Note button actually write down?
**What the answer said — and until 20 August 2026 it could literally write the word "note".** The button is offered in two situations: when we tell you something you were worried about is normal, and when the plant simply looks fine. The first one always had words to save. **The second one had none**, so the entry that landed in your plant's story was the four letters n-o-t-e, and the button would not tell you that beforehand either — it just said *Note this on your Wax Bean*. Now the button shows you the sentence it is about to save, and if the answer had no short summary of its own, it saves the useful part of what was actually said: *that is a bean seedling at the cotyledon stage and it looks healthy, the mulch is doing its job.* Next August, that is the entry that settles the question. It is still marked as coming from the chat rather than from you.
I tapped "Have a look" on a photo and never saw an answer. Where did it go?
It was answered, and until 13 August 2026 there was **no way to reach it.** The button on the plant's page sent your photo off and then cleared itself, and the reply landed on a screen nothing in the app linked to. So the answer existed, on your own device, and you could not get to it by any route. **The plant's page now shows a line for it** — *what we made of your photo* — from the moment you ask. It says we are still looking while we are, and turns into the way in once the answer is back, so a reply that arrived while the app was shut is still there when you come back. **This is also the only place you can tell us which parts of an answer are right**, and that matters more than it sounds. Confirming *where it is growing*, *which crop or variety it is*, or *anything about its health* changes the plant's own record — the app stops guessing at those. Confirming what **stage** it has reached is remembered, but does not yet move its schedule; that still runs off your sowing date. Asking in the chat gives you the answer in words; confirming it is what turns some of it into something the app acts on.
How does it work out how far along a plant is?
**By what it can see, not by counting days.** Our own record for sweet corn says the seedling stage runs from about day 7 to day 31 and the next stage from day 14 to day 55 — **they overlap by more than two weeks**, and 64 of our 66 crops overlap the same way. That is honest: on day 20 a corn plant really can be either, depending on the spring. So counting days cannot settle it, and until now the app settled it with a rule in the code rather than anything about corn. **Now it asks what is visible** — first true leaves, fast new growth, stems ankle high — and every crop’s record already says which stage shows which of those. *"Several true leaves"* is past the first and short of ankle-high, which is exactly the thing a date cannot tell you.
What if the sign it sees belongs to two stages?
**Then it says both, and does not pick.** Some signs genuinely belong to more than one stage — sweet corn lists silks under two of them, because silks appearing and silks drying are both worth noting and there is only one word for it in our list. When that happens the app names both stages rather than choosing, and it will not question your photo on the strength of a reading it cannot place. **Picking one would look more confident and be less true.**
Can it tell what stage my plant is at from a picture?
It has a go, and **the more useful half is that it checks the answer against your own dates.** We know when you planted — nothing else looking at that photo does. So there are three things it might say. If the picture and your dates agree, it offers to write the stage down. If they differ but your calendar could plausibly put the plant there around now, it says both and records nothing: *"that looks like fruit set to me — your dates put it around leafy growth."* **And it will sometimes ask whether it is the same plant** — but only when it is on solid ground. That happens when the stage it sees is one your crop genuinely goes through and your dates put it in a completely different part of the year: fruiting on something you sowed eleven days ago. At that point the photo is telling us about the photo — a neighbor's plant, last year's picture, the wrong bed. It never says you are wrong, and it never writes anything about the plant. **What it will NOT do is question your photo because it could not follow its own answer.** If it describes the plant in words we do not recognize — and it often does, because it writes like a person rather than picking from a list — it says what it sees and leaves it there. **Failing to understand ourselves is not evidence about your picture.** We shipped it the other way round for a few hours on 11 August and a photo of a corn plant caught it. **And if you have already told us a stage, the picture does not get a vote** — your words stay the record and the photo becomes a question for you.
I took a photo of my plant. Will it look at it?
It asks first — on a PLANT PAGE. Take a photo on a plant's page and you get one line — *want me to have a look at that photo?* — with a yes and a no. **We do not read it unless you say so**, because most photos people take of their plants are records rather than questions: the tomato getting big, the first flower, what the bed looked like in July. An answer nobody asked for is the app talking when it was not spoken to. **The exception is a photo we asked you for.** If the chat said *photograph the underside of that leaf* and you went and did it, the asking already happened — coming back and asking again would be the app forgetting its own question. Those get looked at, and so is any photo you send straight into the chat — handing a picture to a conversation is itself the question.
Does it know what the weather is doing?
Yes. It is sent tonight's actual forecast low and the next few days from the National Weather Service for your grid square — not a thirty-year average for your date. That matters for anything about watering, frost or heat: the difference between "it can get cold in August" and "41 tonight". If we have no forecast for where you are, it says nothing about the weather rather than guessing.
What is the question box above the camera button for?
Anything you want to ask about the picture you are ABOUT to take. Type it, then take the photo, and the two travel together. Leave it blank and the question is "what is wrong with this?", which is the right one most of the time. Pressing return in the box opens the camera, so you never have to reach for the button.
Why does it tell me a second thing it could be?
Because one answer invites you to act; two invite you to check. A photo is not a lab test, and the failure that matters in this kind of app is not being unsure — it is being confidently wrong. If there genuinely is no plausible second answer it says that instead, rather than inventing one to look thorough.
Does it know what the plant looked like last time?
Yes, when you have photographed it before. It is sent the earlier picture alongside the new one, both labeled with when they were taken, and it tells you whether the plant is better, worse, or not really changed. It picks a photo at least ten days older on purpose — two pictures three days apart show the light moving, not the plant. And "no real change" is a proper answer here rather than a blank one: a plant that looks the same three weeks on is usually a plant that is fine.
What makes this different from a plant identifier app?
An identifier looks at the picture and nothing else. This reads the picture alongside that plant's own history — the day it went in, what you fed it three weeks ago, the note where you said the lower leaves were yellowing, whether it is in its first season or its third. When an entry of yours explains what it is seeing, the answer says so and shows you the entry with its date. If it cannot point at your history, we have built a worse identifier, so we show you when it does.
How fast is it?
A few seconds. You ask, it thinks, it answers — the same as any chat app you already use. Most gardens are well within reach of a phone signal or the house wifi, so that is what it is built for.
What if I am somewhere with no signal?
It still takes the question. The photo and what you asked are saved on your phone the moment you tap, and it sends itself as soon as you are back in range — a community garden plot, a walled garden, a high tunnel with a metal roof. You will not lose the question and you do not have to remember to come back and ask again. This is the exception rather than how it usually goes.
It recognized a plant I have not added yet. Can it add it for me?
There is a line under the answer offering to add it, and tapping it opens the normal add-a-plant screen with the crop already filled in. It does not appear for a plant that is already in your garden. What it will NOT do is guess when you planted it — that question is asked, the same as always. Every date the app works out for a plant hangs off its planting date, so a wrong one is not a wrong box, it is a wrong calendar for the whole season. A date filled in for you gets tapped past; a question gets answered.
What will it refuse to tell me?
It will not name a chemical, a product or a dose. Timing you can have — there is exactly one week a year to treat peach leaf curl, and that is worth knowing — but what goes on food you are going to eat is your decision, not ours. It also refuses to guess when a planting went in or how old a tree is, at any confidence, because a wrong planting date silently spoils every date the app works out afterwards.
What happens if it cannot tell?
It says so, tells you what would settle it — a closer photo, the underside of a leaf, the base of the stem — and does not use up one of your answers. Refusing well is the point. A confident wrong answer is worse than no answer, because you would act on it.
Why does it sometimes link me to one of your guides?
Because on a handful of problems we have written the long version, and an answer on a phone screen is not the place for it. The link only appears when the answer has already named something we have a page about — it is a lookup on what the answer said, not a menu of diseases shown to the app beforehand. Four problems have one so far: tomato hornworm, squash bugs, and yellowing leaves on tomatoes and on cucumbers. Everything else gets no link, because a link to a page that is not quite about your problem is worse than no link at all — you would trust it and stop looking.
Where does that guide link take me?
To the guide on our website, which opens in your browser. It is the same page anybody can read without the app. Until 7 August it did not — the link quietly reloaded the app instead of leaving it, because the app and the website live at different addresses and the link had only been told half of one. Nothing errored and no page went missing, which is why it took a while to notice: you tapped, something loaded, and it was us. That is fixed for every link out of the app now, not just this one — the list of which addresses belong to the website is built from the website itself, so a page we add tomorrow is covered the day it exists rather than the day somebody remembers it.
It told me to mulch a tree that is already mulched, and the mulch is in the photo.
**That is the complaint that started this, and from 14 August 2026 the photo gets read.** When the picture shows the soil surface, the card comes back with *Mulched* already ticked and a line saying **the soil looked mulched in your photo**. If it looked bare, the line says that instead and the box is clear. **It is not cosmetic.** How often we suggest watering already depends on whether the ground is covered — mulch is one of the four things that decision turns on. So a photo that can see the answer is answering a question the app was otherwise guessing at, and a tree told to mulch a surface that is plainly bark-chipped is the app not looking at what it was sent. **Only when it can actually see the soil.** Leaves over the surface, or a frame that stops at the foliage, means nothing is ticked and nothing is claimed — **not being able to see it is not the same as it being bare**, and that difference is exactly what decides whether you get sent to do a job you have already done.
I sent a photo of plants in pots and it still asked whether they were in the ground.
**Fixed on 14 August 2026.** If the photograph shows where the plant is sitting — a pot rim, a nursery container, a seed tray, bed timber, open soil — the card comes back with that already selected, and it says *from your photo* underneath so you know why. Change it if we read it wrong; it is not written until you press Save. **The odd part is that it was mostly seeing the pot all along.** Measured on a real photo of a potted citrus: it identified the pot in two rolls out of three even before this change. The app simply never read the answer, and asked the question anyway. **It is not that the camera could not tell — it is that nothing was listening.** **When the photo does not show the base of the plant, nothing is pre-selected**, on purpose. *We could not see it* is not the same as *in the ground*, and a guess that looks like an answer is worse than a question.
How many can I ask?
**Ten photo answers, once.** Not ten a month — ten in total, counted from the day you make your account, and they do not refill. The twenty written questions are separate and work the same way. When they are gone they are gone until you upgrade. **Why once rather than a monthly allowance:** a small number each month teaches you to ration and wait, and a gardener with a sick plant in week two does not ask. We would rather you spend all ten on the two weeks that actually matter. Refusals and errors do not count against you. **Everything that costs us nothing stays free forever** — your plants, your dates, your journal, your reminders and the plant database. **And writing in your journal never needs AI at all**, so running out stops you asking questions, not recording your garden.
Why does it call some damage a problem and other damage just worth knowing?
Two questions, in order. First: will it spread or get worse? A disease, or an insect that breeds on the plant, is a problem wherever it lands — peach leaf curl sits on the leaves of a tree you grow for fruit, and it still matters, because a tree that loses its leaves loses the fruit. Only once damage is local and finished does the second question apply: does it hit the part you grow the plant for? A leafminer trail on a sunflower leaf is over — the grub has done its tunneling and the plant is fine, so you are told about it and left to decide. The same trail on spinach is a problem, because there the leaf is the thing you were going to eat.
Is my photo used to train anything?
No. Your photos are sent to the model to answer your question and are not used to train it. They stay in your garden, on your phone and in your own account's storage, and are never on a public address.

More on this

Shipped 2026-08-04. Last checked 2026-08-25. Something here wrong or out of date? Tell us — support docs going stale is a bug.