A landscape architect can use ChatGPT to tidy planting-schedule wording, draft maintenance notes, turn site-visit notes into report text and write client emails, provided the species, sizes, quantities and technical figures come from the architect. Check every plant name against a plant-names database, and check every species against the site yourself.
The line to hold is simple: ChatGPT chooses words, you choose plants. It has no soil test, no survey and no knowledge of the drainage on this site, and it will write a confident planting note around a species that fails in this ground. The routine below gives it a narrow job, then checks the output in the same order every time.
One invented courtyard scheme runs through the examples. Bed A is 24 square metres of shade on moist soil, taken from the survey. Bed B is 42 square metres of sun on free-draining soil. A yew hedge runs 18 metres along the boundary. The planting densities are made up for illustration; yours come from your design.
None of this needs a paid plan to try. For client work, two things matter more than the plan price. The first is training: ChatGPT's business plans keep your practice's content out of model training by default, while on a personal plan a setting in the data controls decides it, so check yours. The second is your client contract and your insurer, which may have views on AI tools, so look before you start. Specialist software for the profession is a separate question, and AI tools for landscape architects covers its planting plans, renders and proposals; the routine here needs nothing beyond plain ChatGPT.
The dividing line: ChatGPT words it, you decide and verify
Every task below fits the same pattern: the architect supplies the content, ChatGPT shapes the language, and a check follows. The table shows where the boundary sits for each document, and it is worth pinning above the desk.
| Document | ChatGPT may | ChatGPT must not |
|---|---|---|
| Species list | Nothing; the list is yours | Suggest, add or swap species |
| Planting schedule | Tidy the layout, keep headings consistent, flag names that look wrong | Change sizes, densities or quantities |
| Planting notes | Draft the wording, leaving a placeholder for every figure | Supply depths, volumes or intervals |
| Maintenance notes | Organise tasks by season | Set how often or how much |
| Site report | Turn your notes into readable sections | Add observations, test results or causes |
| Client email | Draft and adjust the tone | Commit to dates, costs or substitutions |
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowStep 1: a planting schedule that only gets tidied
Start with the schedule, because it is where a small error becomes a purchase order. Give ChatGPT a strictly cosmetic job: consistent columns, consistent headings and a list of names that look wrong. The prompt forbids changes to anything a plant order depends on, and it asks for suspect names to be listed, not corrected, because only a database check can say which name is right.
Tidy this planting schedule for a drawing. Do NOT add, remove or substitute any plant. Do NOT change any size, density or quantity. Output a table with columns: Ref, Botanical name, Common name, Size, Density per square metre, Qty, Notes. Put only what I give in Notes. If a botanical name looks like a synonym or a misspelling, leave it as written and list it under "Names to check" below the table. If a figure is needed that I have not given, write [OFFICE STD].
Bed A (shade, moist soil per survey, 24 square metres)
A1 Dryopteris filix-mas, male fern, 3 litre, 4 per square metre over 14 square metres = 56
A2 Hosta, mixed, 2 litre, 2 per square metre over 10 square metres = 20
Bed B (sun, free-draining soil per survey, 42 square metres)
B1 Lavandula angustifolia 'Hidcote', lavender, 2 litre, 5 per square metre over 14 square metres = 70
B2 Calamagrostis x acutiflora 'Karl Foerster', feather reed grass, 3 litre, 3 per square metre over 20 square metres = 60
B3 Sedum spectabile, showy stonecrop, 2 litre, 6 per square metre over 8 square metres = 48
B4 Rosmarinus officinalis, rosemary, 5 litre, accent plants = 6
Boundary hedge (18 metres)
H1 Taxus baccata, yew, 60-80 cm height, 5 litre, 3 per metre = 54Illustrative output. The table comes back tidy, with two faults hiding in it.
| Ref | Botanical name | Common name | Size | Per sq m | Qty | Notes |
|---|---|---|---|---|---|---|
| A1 | Dryopteris filix-mas | Male fern | 3 litre | 4 | 56 | |
| A2 | Hosta (mixed) | Hosta | 2 litre | 2 | 20 | |
| B1 | Lavandula angustifolia 'Hidcote' | Lavender | 2 litre | 5 | 70 | Drought tolerant |
| B2 | Calamagrostis x acutiflora 'Karl Foerster' | Feather reed grass | 3 litre | 3 | 66 | |
| B3 | Sedum spectabile | Showy stonecrop | 2 litre | 6 | 48 | |
| B4 | Rosmarinus officinalis | Rosemary | 5 litre | accent | 6 | |
| H1 | Taxus baccata | Yew | 60-80 cm, 5 litre | 3 per m | 54 |
Names to check: B3 Sedum spectabile and B4 Rosmarinus officinalis may have been reclassified.
The first fault is B2: the quantity became 66, but 20 square metres at 3 a square metre is 60. The second is the note "Drought tolerant" on B1, which nobody supplied; it may even be true, and it still doesn't belong on a drawing unless you put it there. Recompute every quantity yourself, since each is area times density: A1 is 14 x 4 = 56, A2 is 10 x 2 = 20, B1 is 14 x 5 = 70, B2 is 20 x 3 = 60, B3 is 8 x 6 = 48 and H1 is 18 x 3 = 54. The model did one useful thing, which was to list two names for checking, and the species check further down shows what that check involves.
Step 2: planting notes with the numbers left blank
General planting notes are where ChatGPT is most tempted to be helpful, and where helpfulness does the most damage. Nearly every note wants a figure: a depth, a volume, an interval. Left alone, the model supplies one that looks right. The workaround is to ban numbers outright and have it write [OFFICE STD] wherever one belongs, so you fill every gap from your own specification and the draft can't smuggle in a value you didn't choose.
Draft general planting notes for drawing [DWG NO] as short numbered imperative sentences. Cover: stock quality and inspection on delivery, planting depth, staking and ties, mulching, watering in, aftercare, and replacement of failures. Write [OFFICE STD] in place of every number: depths, sizes, volumes, intervals, dates. Do not refer to any standard, code or law by name. Do not add notes about soil, drainage or species suitability.Illustrative output.
1. All plants to be nursery-grown, healthy and true to name, and inspected on delivery.
2. Reject stock that is damaged, pot-bound or dry; notify the landscape architect within [OFFICE STD] of delivery.
3. Plant at the same level as in the container, in pits [OFFICE STD] wider than the root ball.
4. Stake and tie trees as detailed, checking ties at [OFFICE STD] intervals.
5. Mulch all beds to a depth of 75 mm, keeping mulch clear of stems.
6. Water in thoroughly at planting and twice weekly thereafter for the first 12 months.
7. Replace failures in the next planting season at the contractor's cost.
Three slips got through: 75 mm, the watering rhythm and "at the contractor's cost". The first two are figures the prompt banned. The third is a contract term, which belongs to your appointment and the building contract, not to a planting note. Whether 75 mm is right or wrong isn't the point, because it isn't ChatGPT's call. Replace each with wording from your master specification, and keep one rule for the office: no number reaches a drawing unless a person typed it. The rest of the draft is still only wording, so read it against the specification line by line.
Step 3: maintenance notes without invented frequencies
A first-year maintenance schedule has the same trap in a different form. Frequency depends on climate, soil and the contract, and the model knows none of the three. Ask for structure only: tasks grouped by season, a placeholder for how often and how much, and a column that says what each task depends on. For contract renewals and visit notes after the first year, AI in landscape maintenance goes further.
From this planting schedule, draft a first-year maintenance task list grouped by season (spring, summer, autumn, winter). Columns: Task, Applies to (Ref), How often, Depends on. Write [FREQ] for every frequency and [EXTENT] for every quantity. In Depends on, say whether the task depends on site, climate or contract. Do not give species-specific pruning advice or dates.
[paste the tidied schedule]Illustrative output.
Spring: check stakes and ties (H1; [FREQ]; depends on site). Weed all beds ([FREQ]; depends on site and contract). Top up mulch where thin ([EXTENT]; depends on site).
Summer: water in dry periods (all beds; [FREQ]; depends on climate and contract). Clip the yew hedge in late summer.
Autumn: clear fallen leaves from beds ([FREQ]; depends on site).
Winter: inspect for frost damage and replace failed plants at the next planting opportunity ([EXTENT]; depends on contract).
The slip here is "clip the yew hedge in late summer", which breaks the no-species-advice rule. Timing varies with climate and with the contractor's programme, so either cut the line or replace it from your own maintenance specification. The Depends on column is the reason to keep the placeholders. It shows which blanks you can fill alone and which need the contract open beside you.
The six-gate check for every species line
This is the part of the routine that matters most. Run each species line through six gates, in order, and record the result. The gates cost little for a plant that is fine and a great deal to skip for one that isn't. None of them is a question for ChatGPT, which can speed up the paperwork around the check but can't pass a gate on your behalf.
| Gate | Question | Where to check | A pass looks like |
|---|---|---|---|
| 1. Name | Is it spelled correctly, and is it the accepted name? | IPNI for spelling, author and first publication; Plants of the World Online for accepted names and synonyms; then your nursery's list | Accepted name on the drawing, trade name in brackets in the order list |
| 2. Cultivar | Does the cultivar exist, and is it still sold? | Nursery catalogues and the breeder's own listing | At least one nursery lists it at the specified size |
| 3. Site match | Do soil, drainage, light, exposure and hardiness fit the survey? | Your site analysis and a regional reference you trust | Each factor ticked against survey evidence, not memory |
| 4. Ecology and rules | Is it invasive or restricted? Does the brief demand native or wildlife planting? | The regional invasive-species list, the brief, any conditions attached to the consent | On no list that applies, and the brief's requirements met |
| 5. People and place | Toxic, thorny, allergenic, slippery underfoot, or blocking a sightline? | The brief and how the site will be used | Acceptable for the users, or the client has been told |
| 6. Supply | Is it available at that size and quantity, and when? | Two nurseries' stock lists | Written availability, with a fallback species agreed with the client |
On gate 1, the two databases do different jobs. IPNI says it provides nomenclatural data, meaning spelling, author, types and first publication, for vascular plant names. It doesn't say which name is accepted, and its own page sends you to Plants of the World Online for taxonomic data. So a name can pass in IPNI because it was validly published and still be a synonym today. That is why the nursery's list is the last step: catalogues can carry an older name for years after a reclassification.
A filled log: six lines, four corrections and one query
Here is what the check produces for the courtyard schedule above. The log is short on purpose, because the point is a record you can hand to a colleague or an insurer, not an essay about each plant.
| Line | Gate | Finding | Action |
|---|---|---|---|
| B4 rosemary | 1. Name | Rosmarinus officinalis is now treated as Salvia rosmarinus; rosemary was moved into Salvia in 2017 | Use Salvia rosmarinus on the drawing and give the old name in the order notes |
| B3 showy stonecrop | 1. Name | Sedum spectabile is now Hylotelephium spectabile | Same treatment |
| B2 feather reed grass | Quantity | Draft said 66; 20 square metres at 3 a square metre is 60 | Corrected to 60 |
| A2 hosta | 6. Supply | "Mixed" leaves the nursery to choose the cultivars | Name the cultivars so the drawing and the delivery match |
| H1 yew | 5. People and place | Yew is toxic if eaten, and the brief says children use the courtyard | Raise with the client: keep with a note, or substitute |
| A1 male fern | 3. Site match | Shade and moist soil match the survey | Pass |
A name change isn't always an instruction to change the drawing. Reclassifications are sometimes disputed, and they reach nurseries slowly, so the order list may need both names to avoid a delivery of the wrong plant. The rule I'd use: the accepted name goes on the drawing, the trade name sits beside it in the schedule, and the log says which source you relied on and on what date.
The yew line shows why gate 5 exists. Nothing in the schedule is wrong, and no database would flag it. The problem only appears when the plant meets the brief, and that comparison is a human one. The same goes for gate 3: ChatGPT can be asked to reformat the survey notes, but whether a species suits the ground is a judgement made from evidence on site.
Reports: site-visit notes into text the client can read
Site-visit notes are shorthand written for yourself. A client report needs the same facts in full sentences, sorted into what was seen, what it may mean, what you recommend and what needs a decision. ChatGPT can do the sorting and the sentences if the notes hold every fact. The most useful instruction in the prompt is the last one: a list of what the notes don't cover, so that gaps show as gaps instead of being filled with confident prose. The same split between drafting and professional judgement is argued in can AI help write an arborist report, and what stays with the arborist.
Turn these site-visit notes into a short report section for the client. Use four headings: What we saw, What it may mean, What we recommend, Decisions needed from you. Use only the notes. Do not state causes, soil types or test results unless the notes give them. Where I have written "hand test only" or similar, keep that limit in the wording. End with a list headed "Not yet checked" containing anything a site visit like this normally covers that my notes do not mention.
Notes:
- Bed A: soil heavy and sticky when wet (hand test only), water standing about 20 mm after last night's rain, photo 4
- Bed B: soil drier, stones near the surface, photo 6
- Boundary: existing hedge on the neighbour's side, overhangs about 1 m, photo 8
- Access: 1.2 m side gate, no vehicle accessIllustrative output.
What we saw. Bed A: the soil is clay with poor drainage, with water standing about 20 mm deep after rain (photo 4). Bed B: the soil is drier and stony near the surface (photo 6). An existing hedge on the neighbouring side overhangs the boundary by about 1 m (photo 8). The only access is a 1.2 m side gate.
What it may mean. Bed A may need drainage improvement before planting.
What we recommend. Confirm soil type and drainage with a test before species are fixed.
Decisions needed from you. Whether to agree a drainage investigation.
Not yet checked. Soil pH, underground services, existing tree roots, boundary ownership.
Two claims overreach. "The soil is clay with poor drainage" turns a hand test on one wet day into a classification and a diagnosis. Rewrite it as "heavy and sticky when wet on the day of the visit (hand test only), with water standing about 20 mm after rain". Keep the Not yet checked list, which is the most useful thing the model produced: it is a reminder rather than a claim, and every item on it is yours to confirm. Refer to photos by number only; the photographs themselves don't need to go into the chat.
Client emails for supply problems and plant failures
Emails are the lowest-risk use here, as long as the commitments stay with you. Give the model the facts and the choices you have already made, and tell it not to add options, costs or dates. Two cases come up constantly: a supply problem before planting, and a failure found at a later inspection.
Write an email from [ARCHITECT FIRST NAME] to [CLIENT FIRST NAME]. Fact: the nursery cannot supply line B3 (showy stonecrop, 2 litre, 48 plants) before the planting window opens on [DATE]. I have chosen two options: (1) wait until [DATE 2] and keep the planting as drawn, or (2) substitute [SPECIES X] at the same size and density, with no change to the layout. State the effect of each on the programme in one line each, using only what I give here. Ask for a decision by [DATE 3]. Do not suggest other options, mention costs, or promise anything. Plain, calm, under 150 words.Illustrative output.
Hi [CLIENT FIRST NAME], a quick update on the planting. The nursery can't supply the showy stonecrop in bed B before the planting window opens on [DATE]. There are two ways forward. Option 1: wait until [DATE 2] and plant as drawn, which moves that bed back by the same amount. Option 2: substitute [SPECIES X] at the same size and density, with no change to the layout, so the programme holds. Could you let me know which you prefer by [DATE 3]? There would be no extra cost to you either way. Best wishes, [ARCHITECT FIRST NAME]
"No extra cost to you either way" is a commitment the prompt forbade. Whether a substitution changes the price is a contract matter, and the honest answer may be yes, so cut the sentence or replace it with the true one. Swap the facts line and the same prompt handles a failure. Say that at the 12-month inspection 6 of the 56 plants on line A1 have failed, about 11 per cent, and paste the replacement clause from the contract. Let the model quote only the clause you paste, and check the percentage yourself: 6 divided by 56 is 0.107.
The checking record to keep with each document
A short record turns a good habit into evidence. Keep one with each document that ChatGPT touched, in the job file, and fill it in at the point of sign-off. The version below fits on half a page.
Document: [title, drawing number, revision]
ChatGPT used for: [tidying / drafting wording / report sections / email]
Content supplied by me: [species list, sizes, quantities, figures, notes]
Species names checked (gate 1): [source, date]
Site match checked against the survey (gate 3): [yes / no / not applicable]
Rules and ecology checked (gate 4): [list used, date]
Every figure typed by a person, none by ChatGPT: [yes / no]
Quantities recomputed (area x density): [yes / no]
Client details removed before pasting: [yes / no]
Checked by: [name] Date: [date]If a plant fails or a name is queried a year later, the record shows what the check covered and what it didn't. Whether your professional indemnity insurer asks about AI-assisted work depends on the policy, so ask the insurer instead of assuming. The drafting saves time on the dull part of the job. The checking is the work you were always responsible for, and the record is how you show you did it. For a sense of where ChatGPT is reliable and where it isn't, how accurate ChatGPT is on business tasks sets expectations task by task.
Questions landscape architects ask about ChatGPT
Can ChatGPT choose plants for a site?
It can list plants that sound plausible for a description, but it has no soil test, survey or microclimate data, and it will write confidently about species that fail in your ground. Choose species from your site analysis and references you trust, and use ChatGPT only for wording and formatting afterwards.
Where do I check that a plant name is correct?
IPNI records the spelling, author and first publication of plant names but doesn't say which name is accepted; its own page points to Plants of the World Online for that. Then check the name your nursery lists, because trade catalogues can keep an older name for years after a reclassification.
Can I upload drawings or survey PDFs to ChatGPT?
Strip the title block, addresses and client names first, and treat any dimension or area it reads from a drawing as unverified until you have checked it against your own figures. For planting notes and reports you rarely need to upload the drawing at all, because the text and numbers you type in are enough.
Do I have to tell a client I used ChatGPT?
Check your client contract and your professional indemnity policy, because wording on AI-assisted work varies and no single rule applies everywhere. What you can always do is take full responsibility for the document: you checked every name, figure and recommendation, and your name is on it.
Further reads
- Landscape Design Proposal: Write Concept, Scope and Fees With AI — Write the concept, scope and fees section that comes before planting notes.
- Garden Maintenance Contract: What Each Clause Is For, Drafted With AI — What each clause in a maintenance contract is for.
- Best Landscape Design Software With AI: What to Compare First — Design software with AI features, and what to compare first.
- ChatGPT for Building Design: Briefs, Options and Code Checks — The same drafting rules applied to briefs and options for buildings.
- Claude vs ChatGPT vs Gemini for Architects: Which to Pay For — Which chat assistant a design practice should pay for.
- AI Garden Design Visuals in Quotes: Do They Help Close Jobs? — Whether AI garden visuals help close jobs.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: IPNI (International Plant Names Index) site description; GBIF species records showing Rosmarinus officinalis and Sedum spectabile as synonyms of Salvia rosmarinus and Hylotelephium spectabile, with rosemary's 2017 move into Salvia cross-checked in encyclopaedia entries; OpenAI's ChatGPT pricing and help pages on data controls and business-plan training defaults.