AI vs Human Garden Design: The 2026 Retest
Can AI help you redesign your own garden? These fresh examples suggest it can make a useful starting point: the planting is more coherent, the garden visual follows the photo, and the overhead plan largely follows the visual. But convincing presentation still hides mistakes and assumptions that matter when you start buying plants or laying paving.
This September 2026 update revisits our original AI vs human garden design comparison. We follow the same three tests, in the same order: planting plans, 3D garden visuals and 2D plans. All the human design examples below are unchanged; the AI outputs are new.
We have in mind someone redesigning their own garden with ChatGPT or a similar assistant, perhaps using a free account or a paid plan, without specialist design software.
About this retest: we generated one fresh output for each test using OpenAI’s image-generation tool in this workspace. This is not a verified ChatGPT Free or Plus test, and results in your account may differ. We reused the published planting prompt. The original full garden brief is unavailable, so we reconstructed a shorter brief from the article and used the original garden photo. The human designer had the original brief: this is a practical comparison, not a controlled like-for-like trial.
Planting Plans
A planting plan needs to do more than show attractive flowers. It should help you understand what to buy, how many plants you need and where they go, while allowing for how they grow. For a beginner, a confident-looking key can make an unverified answer especially persuasive.
AI Vs Human Test: Planting
The prompt
Create a detailed 4m x 2m planting plan design with an image. The plan should: - Be designed for full sun - Include a mix of evergreen and deciduous plants - Provide year-round interest (flowers, foliage, structure) - Incorporate pollinator-friendly species where possible - Specify planting positions in a simple layout grid
AI Output
AI Summary
There is a clear improvement over the original example. Plants repeat across the border, the letters relate to a readable key, and the composition mixes upright flowers, grasses and rounded shrubs.
However, the picture still lacks any artistic finesse, laying the plants out in a simple grid of individual plants rather than staggered, asymmetrical groupings. The output is actually symmetrical and laid out using overly simplified front, middle and back layers with matching plants on each side. It's clear by looking that the output would look fairly bland and uniform in a garden setting and with little consideration for balance or flow. Some of the plants are also oddly positioned e.g. the Heuchera in the middle will get swallowed by the Rudbeckia in-front of it. Last of all there's no coherent stylistic vision. The colours jar in places and there is little aesthetic cohesion between the plants.
There is also a specific size problem: the key gives Choisya ternata a mature height and spread of only 1m. The RHS choisya guide describes most choisyas reaching around 2.5m, with the compact ‘White Dazzler’ around 1.5m high. A 1m allowance for this shrub needs revisiting in a border only 2m deep.
Human Output
The original palette includes Miscanthus ‘Morning Light’, Pittosporum ‘Golf Ball’, Potentilla ‘Primrose Beauty’, Stipa tenuissima, Agapanthus ‘Northern Star’, Geranium ‘Mavis Simpson’, Geum ‘Totally Tangerine’, Salvia ‘Caradonna’, Silene coronaria and Verbena bonariensis.
Human Summary
The human example links the planting layout to a numbered key and quantities. In Urban Plot’s view, its repeated groups, contrasting leaf shapes and purple-and-orange accents create a more deliberate rhythm. Those are design judgements; the practical advantage is that you can trace the plants from the schedule to their positions.
These are the same human drawings as before, not a new design made after seeing the AI result. They illustrate the output being compared, rather than a universal planting recipe for every sunny garden.
Verdict
AI is still no able to recreate the human eye for aesthetic, the imitations have improved since we last tested this but they are still far behind a skilled designer.
3D Garden Design Visuals
A garden visual helps you imagine changes before committing to them. It can be particularly helpful if you struggle to picture a patio or border from a plan. Here, the useful test is whether AI keeps the real garden recognisable and responds to what you want.
AI Vs Human Test: 3D
The prompt
We supplied the original garden photograph below. Because the full original brief was not published, this wording reconstructs only the requirements described in the article:
Redesign my UK back garden in this photo and show me a realistic 3D-style garden visual from the same viewpoint. Keep the garden recognisable, including the existing boundaries and shed. I would like attractive planting, a seating patio, a small wildflower meadow area and a good view from a window seat in the house behind the camera. Please deal with the tree stumps at the back and keep the manhole cover in the lawn accessible. Make paths lead somewhere useful. If you include a pergola, consider the light coming into the house. I do not have measured dimensions, soil details or orientation to give you.
AI Output
AI Summary
This result deserves credit. The shed, neighbouring buildings and broad garden shape remain recognisable. There is a seating patio, a meadow-like area at the back and a path leading to the shed. The inspection cover remains visible in the lawn. Several weaknesses highlighted in the old AI image are less apparent here.
It also avoids adding a pergola, so the earlier concern about a new structure blocking light does not arise in this version. That is a sensible response to an optional feature, though it does not establish how sunlight falls across the garden.
Other questions remain unanswered. The rear stumps are no longer visible, but the image does not say how they would be dealt with. The house and proposed window seat are behind the camera, so their precise relationship to the garden cannot be checked. Patio size, levels and drainage are unknown.
For someone exploring their own garden, this is useful visual inspiration. It could help you say “I like that seating position” or “I want more lawn”. It cannot establish that the proposed arrangement fits your measured space or budget.
Again though the artistic balance and flow is missing from the design. AI can obviously only treat the image at face value, which means it can't consider elements that sit outside the image e.g. location of the house and doors. Also, though it hasn't been able to faithfully recreate how a human would use the space. The output design is fairly typically and unimaginative. The journey through the garden is not really considered. Balance and proportion don't align to that of a skilled designer. What's more, as the image is only able to capture part of the garden, the treatment elsewhere is to be assumed as AI can't knit images together and carry through visuals in one space to align with how you design in another - at least not in the tests we've done.
Human Output
Here are the original human views of the same garden, retained without alteration.
Human Summary
The two views explain the same designed space from different angles. The curved lawn, seating areas, arch and rear section work together, with the window-seat view considered as part of the original brief. In Urban Plot’s view, the partially screened rear space also gives the garden a sense of depth and discovery.
The advantage is the connection between the brief, the views and the plans that follow. A human-produced render alone would still need measurements and site information before work could start.
Verdict
The fresh AI visual is still in the same place as it was over a year ago, from our tests; a decent starting point or conversation starter rather than something considered and actionable. The human design remains more developed across the supplied drawings, but this reconstructed-brief test cannot tell us how AI would have answered every detail of the original brief.
AI Vs Human Test: 2D Plans
The prompt
For the final test, we supplied the new AI garden visual as the reference image and used the original follow-up wording:
Create a 2D plan drawing of the 3D design just created.
AI Output
AI Summary
The overhead image broadly follows the visual: patio at the back left, shed at the back right, curved lawn and stepping-stone path. Its labels and legend are readable. This is a much more coherent way to explain the concept than an unrelated layout.
The serious problem is the authority it appears to have. It prints “Scale 1:50 (A3)”, a metre scale bar and a north arrow, although we supplied neither garden measurements nor orientation. Those details cannot be relied on. Do not measure from this image to order paving or mark out your garden.
It also invents details outside the supplied view, including the house frontage, and labels features such as a water butt and plant types without those being established in the brief. The meadow area is no longer clearly identified. A tidy drawing can quietly turn assumptions into apparent facts.
Also, without wanting to repeat myself, the final output from a design perspective is simple, dull and repeated in every garden up and down the country. There's no artistic vision, style or imagination. In short, AI can't put itself in the garden and think how it feels to be there.
Human Output
The original human output includes the measured layout and supporting sheets below.
Human Summary
These sheets connect the layout to dimensions, design explanations and material choices. They give the homeowner more to discuss with a landscaper than the AI image alone. The original material examples should still be checked for current availability and suitability.
The Society of Garden and Landscape Designers’ guide separates presentation drawings from planting plans and construction information. That distinction matters whichever tool creates the picture: a convincing overview is only one part of getting a garden built.
Verdict
AI has produced a clearer overhead illustration, but it has not produced a trustworthy measured plan. The invented scale is the strongest reason in this retest to pause before treating a polished output as something you can build from.
Most importantly though it has not met the core requirement of a garden designer; to create a beautiful space the customer couldn't have thought of or created for themselves.
Looking Ahead
These examples change the balance of our original comparison. AI is more helpful here at organising a planting idea, keeping a photo-based garden recognisable and carrying a concept into a second view. We should acknowledge those improvements rather than repeat criticisms the new images no longer support.
If you are redesigning your own garden, use the conversation to explore preferences, compare ideas and sharpen your brief. Then check plant sizes and growing conditions, measure the space and resolve practical details before spending money. Paying for an AI subscription does not, by itself, validate a drawing.
Ready to turn your ideas into a coordinated garden design? Explore Urban Plot’s garden design packages.
Still deciding what you want? Start with our garden design brief questions. Your own photos, measurements and priorities make that conversation more useful.
Sources
Plant size check: RHS — How to grow choisya.
Design stages and drawings: Society of Garden and Landscape Designers — Working with a designer. The AI assessments above are observations of the three fresh outputs shown, not claims about every AI tool.
The cover image is an AI-generated editorial illustration. The human design drawings are the original Urban Plot examples; the three fresh AI test outputs are labelled above.
