A landscape architect can use an assistant (ChatGPT, Claude or Gemini) to draft and cross-check plant lists, schedules and proposals, and a render tool such as Adobe Firefly, Rendair or Veras for concept images. The drawing stays in your CAD software, and every plant, quantity and fee still needs your own check.
Much of what ranks for this search is aimed at homeowners: upload a photo of the yard, get a pretty redesign. Those apps skip what a practice has to defend, which is why each plant is on the drawing, whether it will survive the site and what it costs to establish. Practice work runs from the first enquiry to the plant schedule, and the useful AI sits at several points along the way.
A conditions sheet that shows what was measured and what was assumed
The palette is only as good as the conditions behind it, and most planting failures trace back to a condition that was assumed and never checked. Before any plant list, turn your visit notes, photos and the client's comments into a sheet with a source column: measured, told or assumed. An assistant is good at sorting messy notes into that shape, and at leaving gaps empty if you tell it to.
From the visit notes below, build a conditions sheet with these columns: condition, value, source (measured, told by client, or assumed), and "verify before design stage 3?" (yes or no).
Rules: never infer soil type, pH, drainage or wind from a photograph. If a condition is not in the notes, write "not recorded". Mark anything not measured or told as "assumed" and set verify to yes.
Notes: [paste visit notes and client comments]A filled-in sheet for a courtyard behind a dental practice, an invented example used from here on, might look like this:
| Condition | Value | Source | Verify first? |
|---|---|---|---|
| Aspect | Courtyard faces north-east | Measured with a compass | No |
| Winter light | Shaded from November to February by the neighbouring building | Told by client | Yes (sun-path check) |
| Soil | Heavy clay | Two test holes dug | Yes (pH not measured) |
| Drainage | Water standing 24 hours after rain at the low corner | Seen on the visit | Yes (drainage test) |
| Wind | Exposed from the west | Assumed from the layout | Yes |
| Underground services | One line along the south edge | Told by client | Yes (service search) |
| Water supply | Outside tap on the east wall | Measured | No |
Five of those seven rows need verifying, and three of them are "told" or "assumed", which is exactly what would otherwise have slid into the palette prompt as fact. The sheet also tells the client what you need from them: a pH test, a drainage test and a services search. The instruction not to infer soil from a photograph is deliberate. Assistants will describe a soil as a sandy loam from a photo with complete confidence, and they cannot know.
A planting palette from site conditions, then four checks per plant
Start the assistant where a designer starts: with the site, not with a plant. Give it the conditions you have measured or been told, the constraints in the brief and the limits you will not cross. A prompt that works as a template:
Suggest a planting palette for the area described below. Return a table with: botanical name (genus, species, cultivar), common name, mature height and spread, light needs, soil and moisture needs, the hardiness rating in the system I name, season of interest, maintenance level, and a column called "verify" listing anything you are not certain of.
Site: courtyard garden, 180 sq m, shaded for much of the winter, full sun from midday in summer. Heavy clay, wet in winter, dry in late summer. Exposed to wind from the west side.
Hardiness: the site's rating in [name the rating system and number].
Brief: low maintenance, pollinator-friendly, no plants toxic to people or pets, no species listed as invasive in the region, evergreen structure plus seasonal colour.
Return 20 candidates. Do not state a hardiness rating you cannot support; write "unsure".The "verify" column and the permission to say "unsure" are the part that matters. An assistant asked for a confident table will give one, and a confident table of plants is how a wrong plant reaches a drawing.
Then every plant goes through four checks before it is drawn:
- Name. Confirm the botanical name and cultivar in a plant database you trust. Names change, cultivar names get misspelt, and a plan that cites a plant the nursery cannot find is a problem on tender day.
- Climate. Check the hardiness rating for the site's region in the rating system your clients and nurseries use. A hardiness rating describes winter cold. It says nothing about heat, wind, wet feet or drought, and it describes a region rather than a frost pocket behind a wall.
- Conditions. Match light, soil, moisture and exposure to the site, in that order of importance.
- Safety and supply. Check toxicity against the brief, whether the species is listed as invasive or restricted where the site is, and whether nurseries can supply the size you specify.
What the plant audit finds in a typical first draft
An illustrative audit of five suggestions from a draft palette shows the kind of problem to expect. None of these is exotic. Each is something a tired designer could miss at six in the evening.
| Assistant's suggestion | Its claim | What the check found | Action |
|---|---|---|---|
| Lavandula angustifolia 'Hidcote' | Suits a sunny bed | The bed is heavy clay that stays wet in winter, and lavender needs sharp drainage | Replace with a clay-tolerant perennial such as Hemerocallis |
| Buddleja davidii | Pollinator magnet, low maintenance | Classed as invasive in some regions, which is why the regional list has to be checked | Remove, or use a sterile cultivar only if the regional list allows |
| Digitalis purpurea | Cottage-garden accent | Toxic if eaten, and the brief says no toxic plants | Remove and check the whole list against the brief |
| Acer palmatum cultivar | Specimen by the west fence | Exposed to wind from the west, so leaf scorch is likely | Move to the sheltered corner or choose a tougher tree |
| A shrub with a stated mature spread of 1.2 m | Spacing at 1.2 m centres | The nursery lists the spread as 2 m | Respace and recount the quantity |
The last row shows how a small data error spreads. In a 24 square metre bed, a square grid at 1.2 m centres holds about 17 shrubs, and at 2 m centres it holds 6. That is 11 plants of difference; at $38 a plant the schedule is out by $418, and the bed would have been crowded within three seasons. The audit took minutes. The correction after planting would take a replanting.
Run the audit as a rule, not a mood. The designer checks every plant. A second person re-checks a sample, perhaps one plant in ten, and any error found sends the whole palette back for a full pass. If the project involves a sensitive habitat or protected species, put a specialist in charge of selection. An assistant can then draft the table, but it should not choose. For notes and emails around the planting, the tutorial on ChatGPT for landscape architects: planting notes, reports and emails covers the writing side.
A second prompt that argues against the list
A useful second pass reverses the assistant's job. Paste the approved palette and the conditions sheet, and ask it to act as a sceptical nursery manager: for each plant, give the most likely reason it fails on this site and what evidence would settle the question. A typical result (illustrative):
- Geranium (hardy cranesbill): copes with clay but will sulk in deep winter shade, so check how much light reaches the bed from November to February.
- Panicum virgatum: needs sun, and the north-east corner is too dark; move it to the sunnier end.
- Cornus sericea: coloured stems depend on hard pruning each spring; confirm the maintenance contract includes it.
Each point is either a reason to move a plant or a question for the nursery. None of them is an answer, and the assistant can be wrong in this direction too, by inventing a problem that is not there. The geranium point is the weak one here: many hardy cranesbills tolerate shade and most die back over winter, so winter shade is a reason to check the species you chose, not to move it. Verify before acting, and note the result on the schedule.
From palette to schedule: let a spreadsheet count
Once the palette is approved, the plant schedule is a table: symbol, botanical name, common name, size at planting, spacing, quantity. Assistants are fast at laying that table out from your notes. They are unreliable at counting in their heads, so ask for a formula or a spreadsheet and check a few cells yourself.
Spacing is where mental arithmetic fails most often. Take a 36 square metre bed of groundcover planted at 0.3 m centres. On a square grid that is 400 plants. Planted in staggered rows, which is how most groundcover is actually laid out, the same bed takes about 462, because a staggered layout packs roughly 15% more plants into the same area. A draft that used the square grid would leave you 62 plants short on a $9-plant line, which is $558. Ask the assistant which layout it assumed, and state the layout in the schedule notes.
The better check is a reconciliation. Export the plant counts from your drawing as a CSV, paste both lists and ask for every mismatch (illustrative):
| Symbol | Drawing export | Schedule | Difference |
|---|---|---|---|
| CS (Cornus) | 11 | 9 | 2 on the drawing, not in the schedule |
| GR (Geranium) | 58 | 58 | None |
| PV (Panicum) | 24 | 26 | 2 in the schedule, not on the drawing |
| AM (Amelanchier) | 3 | 3 | None |
Two mismatches in four lines would be a normal result late in a project, when a bed has been redrawn and nobody has updated the schedule. A tender that goes out with those two differences produces two variations later. This is a good use of an assistant because the task is comparing, and the answer is checkable in seconds against the two sources.
Concept renders: what each tool needs and what it costs
The prices below are what the vendors' own pages showed on 29 October 2026, in USD, except where noted.
| Tool | What it takes in | List price | Points to know |
|---|---|---|---|
| Adobe Firefly | Photos and prompts; Generative Fill for editing a site photo | Standard $9.99 a month (2,000 credits); Pro $19.99 (4,000); Pro Plus $49.99 (10,000); Premium $199.99 (50,000) | Adobe says outputs from its own Firefly models are safe for commercial use. For partner models in Firefly you judge whether the output suits the project. IP indemnity is for eligible enterprise customers only. |
| Rendair | A sketch, a floor plan, a 3D model or a photo | Starter $19 a month (500 credits); Pro $49 (1,500); Studio Pro $199 (7,500); yearly billing gives two months free | Every plan includes a commercial licence (Pro and Studio Pro say "full"). Free start with 20 credits and no card. Its examples include a site plan and a garden path plan, and its site links an "AI Backyard Design Generator" use case, but nothing describes practice drawings, so test it on your own plan. |
| Veras (EvolveLAB) | A view from your 3D model, as a plug-in for SketchUp, Revit, Rhino, Archicad or Vectorworks, or an image on the web | $59 a month on monthly billing; $29 a month billed yearly ($348); floating licence $51 a month billed yearly. The page shows a dollar sign without naming the currency | Free start of 30 renders and a 15-day trial. Its examples are interiors and building elements, with no mention of landscape. |
| Midjourney | Prompts and reference images | $10, $30, $60 and $120 a month across its four plans, per its docs as shown in search results (the page itself did not load for me) | Per the same search summaries, companies above $1,000,000 gross annual revenue must use Pro or Mega, and private "Stealth" mode is on Pro and Mega only. Check the current page before buying. |
One compatibility note matters for a CAD-first practice. Vectorworks Landmark's own page describes landscape tools such as sun and shade analysis, drainage, slope, material takeoffs and converting sketches into drawing objects. Vectorworks also has a built-in AI image tool that makes images from a prompt or a Vectorworks view; Vectorworks presents it as an ideation and quick-sketch aid rather than a rendering engine, and its output is a picture, not drawing objects. So a Landmark user can try concept images without leaving the software (check that your licence includes it), and Veras, which plugs into Vectorworks, or a web tool is the alternative. Either route works, as long as the render starts from the approved plan and not from a screenshot someone changed.
Rules that keep a render honest
A render is persuasive, which is the problem. Clients remember the image and forget the caption. Five rules protect you:
- Specify the growth stage. Write "three years after planting, not mature" in the prompt. Otherwise the tool paints year-twenty trees on day one.
- Specify the season. A garden shown in full bloom in a winter presentation is a promise.
- Lock the hard landscape. Say what must not change: paving size and colour, wall heights, the position of the pergola, the shape of the raised bed. Then compare the output to the plan.
- Show only plants that passed the check. If the render includes a plant you did not specify, it has no place in the presentation.
- Caption every image. "Concept image only. Planting and proportions are indicative." Keep the prompt, the tool and the date in the project file.
A prompt that applies the rules:
Concept render of a courtyard garden for a dental practice, viewed from the waiting-room doorway at eye height, late afternoon in early summer.
Keep exactly as drawn: the light grey large-format paving terrace, the curved raised bed on the left, the timber pergola at the far end, the boundary wall at 1.8 m.
Planting: three years after planting, not mature. One multi-stem serviceberry as the focal tree, about 3 m tall; ornamental grasses; low perennials in blue and white.
No people. Two chairs only. Natural light, no lens effects.Realistic slips to look for in the result: the focal tree drawn at ten metres, the paving pattern "improved" into a different layout, the pergola gaining a second bay, and flowers that bloom in a different month. Overlay the render on the plan and measure two or three elements before it goes anywhere near a client. For the commercial question, an image from a tool that gives you a licence is not the same as an image you own outright, so read the plan's terms. The copyright and client data risks in architecture practices tutorial goes into it.
Proposals: from a discovery call to a staged fee letter
Proposals are where an assistant can save a lot of time at modest risk, provided you own the numbers. Use a call transcript or your notes, and ask for the structure: a cover paragraph, a scope per stage, assumptions, exclusions and next steps. You supply the hours, the rates and the terms.
Here is a fee table for a 180 square metre courtyard behind a dental practice, at $120 an hour:
| Stage | Hours | Fee |
|---|---|---|
| 1. Brief and site appraisal | 15 | $1,800 |
| 2. Concept design | 30 | $3,600 |
| 3. Developed design and planting plan | 40 | $4,800 |
| 4. Construction information, specification and plant schedule | 45 | $5,400 |
| 5. Tender support | 10 | $1,200 |
| 6. Site visits (6 visits of 3 hours) | 18 | $2,160 |
| Total | 158 | $18,960 |
Compute the total in a spreadsheet, not in the chat window: 158 hours at $120 is $18,960. Then ask the assistant to write the stage descriptions around your table, with these assumptions and exclusions stated plainly:
- The client provides a topographical survey, or one is quoted separately.
- Two rounds of revisions per design stage; further rounds are charged at the hourly rate.
- Planning or permit fees and application preparation are excluded unless added in writing.
- Irrigation and lighting design are excluded.
- Plant quantities at concept stage are indicative only.
- The fee is valid for 30 days.
Here is one stage description as the assistant first drafted it, and as it went out:
| First draft | As sent |
|---|---|
| Stage 2, Concept design: we will create a stunning, low-maintenance courtyard that thrives all year round, with unlimited design options until you are completely happy. | Stage 2, Concept design (30 hours, $3,600): two concept options as plans, with a mood board and up to three captioned concept images; up to two rounds of revisions; an indicative planting palette, with quantities to follow at stage 4. |
The first draft promised an outcome nobody controls ("thrives all year round") and contradicted the revision limit in your own assumptions. The second sells hours, deliverables and a limit the client can hold you to, and it matches the fee table line for line.
Read the draft for promises you never made. Typical ones are "unlimited revisions", "guaranteed approval", "plants guaranteed to thrive" (an establishment warranty usually belongs to the contractor, not the designer) and "completion in six weeks" when nobody has seen the programme. Strike them, check the total and the stage hours against your sheet, and read the exclusions list twice.
If you want proposal, contract and invoice in one place, HoneyBook's pricing page lists Starter at $29, Essentials at $49 and Premium at $109 a month on yearly billing. Starter lists proposals and contracts plus invoices and payments, and the higher plans say they include everything below them. Each plan lists HoneyBook AI, and the comparison table names features such as email drafts, an AI chat and a meeting notetaker without making clear which plan gets which. There is a 30-day free trial with no card needed. It is a business-admin tool, not a design tool, so judge it on whether you will actually send proposals through it. When you tell a client how you use AI, the guide to mentioning AI use in client contracts gives wording.
A two-person practice's monthly stack
A plain version, at list prices on annual billing where offered: two Claude Pro seats at $17 each is $34, Firefly Standard at $9.99, and HoneyBook Starter at $29, for $72.99 a month. Swap Firefly for Rendair Starter, $19 on monthly billing or $190 a year, if you want sketch-to-render, and the total becomes $82 a month (about $79 with Rendair billed yearly). One recovered chargeable hour at $120 pays for either, but only if the hour is spent on work that bills. Before you add a fourth tool, spend a month finding out which of the three you actually opened each week.
Jobs that stay with a specialist, however good the draft looks
Some planting decisions carry consequences a drafting tool cannot weigh. Keep these with a person who is qualified to own them:
- Planting that forms part of a consent or an ecological commitment. The species list is then a legal or contractual promise, and changing it is not a design tweak.
- Slope stabilisation. Roots and drainage do structural work, and the engineer decides what the planting has to achieve.
- Trees near buildings, drains and services. Root spread, water demand and soil type interact, and a wrong choice shows up years later as damage.
- Play areas, schools and healthcare courtyards. Toxicity, thorns and allergenic pollen need a documented human decision. In the dental-practice courtyard, an assistant could draft a list, but you would still decide whether strongly scented plants suit a waiting area.
- Rare or protected habitat. Bring in an ecologist before anything is drafted, and treat the assistant's output only as a note-taker's tidy-up.
Sign-off before anything leaves the studio
A twenty-minute check, done by someone other than the person who produced the work, catches most of the slips described above. Here it is filled in for the courtyard example:
- Plants: 20 of 20 names verified in the plant database; 20 of 20 hardiness ratings checked; toxicity and invasive status checked against the brief; nursery stock list received for the three trees.
- Schedule: counts reconciled with the CAD export, with two mismatches fixed; layout assumption (staggered) stated.
- Renders: three images, each captioned, each overlaid on the plan; one rejected for a second pergola bay.
- Proposal: total recalculated, exclusions read, three over-promises removed.
- Files: prompts, tool names and dates saved in the project folder; no client files uploaded to a tool that trains on them.
Keep a count of what the check finds. If the audit catches nothing for a month, the assistant is probably being given easy work. If it catches something every time, the prompts need tighter conditions, or some of the work belongs with a person. For where this fits in the wider tools picture, the comparison of rendering tools for small practices and AI renders versus outsourced visualisation help decide whether to render in-house at all.
Questions from landscape practices
Can AI choose plants that will survive on my site?
It can draft a shortlist from the conditions you give, but it cannot know your microclimate, your soil test or current nursery stock. Treat every suggestion as unverified. Check each plant's name, hardiness rating, light, soil and moisture needs, mature size, toxicity and invasiveness against a source you trust before it reaches a drawing.
Are AI renders safe to show clients as the design?
Show them as mood and massing, never as the design. Renders invent plant sizes, paving patterns and geometry. Label every image as a concept, state the growth stage you asked for, and make sure each plant shown has passed your plant check. The drawings and schedule remain the contract documents.
Who owns images made with AI render tools?
It depends on the tool and the plan. Adobe says outputs from its own Firefly models are safe for commercial use but leaves partner-model outputs for you to judge, and gives IP indemnity to eligible enterprise customers only. Rendair says its plans include a commercial licence. Read the terms of the plan you buy before using images in client work.
Should I tell clients that I use AI?
Yes, in plain words in your terms or proposal: what it is used for, that you check the output, and that client files are not put into tools that train on them. One sentence at proposal stage is far easier than explaining an unlabelled render after the client has spotted it.
Further reads
- Best Landscape Design Software With AI: What to Compare First — Landscape design software with AI, compared.
- AI Garden Design Visuals in Quotes: Do They Help Close Jobs? — Do AI garden visuals help close jobs?
- How Architects Use AI for Fee Proposals, Briefs and Paperwork — Fee proposals and briefs from the architecture side.
- AI Policy for Architects: A Ready-to-Use Checklist and Clauses — A ready-to-use AI policy checklist for design practices.
- Best AI for Architecture and Interior Design: Concept Image Tools by Cost — Concept image tools for design work, compared by cost.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Adobe Firefly plans page; Rendair pricing and home pages; Veras (EvolveLAB) page; Vectorworks Landmark page and Vectorworks news pages on its built-in AI image tool; HoneyBook pricing page; Midjourney plan details as shown in search results of its own docs, because the pages did not load directly; Anthropic Claude pricing page (read 29 October 2026).