Dear Heather, We suddenly have a huge amount of horseweed in our back garden beds. I asked ChatGPT what to do and it sent me on a spiral. So then I panicked and bought a tarp and many garden waste bags. Then I looked on native plant Reddit and some people seem to leave it. My wife and I are often so overwhelmed by the amount of information and everything we read is conflicting. — Jillian, Pawtucket, RI
ChatGPT is the internet on speed, and MOST landscaping advice on the internet negatively impacts our ecosystem. You already know traditional residential landscaping is decimating wildlife — and it’s certainly not getting people outside. But ChatGPT can sound so authoritative that it’s easy to forget it’s trained on sources that reflect those ecologically destructive practices.
On top of that, the internet is global, so AI models (and humans) pick up advice not appropriate or relevant to your location. For example, horseweed is a listed invasive in many European countries, where it’s exotic. But it’s native throughout most of North America and, by definition, a plant cannot be invasive in its native range.
Unfortunately, it will take time for the internet Zeitgeist to reflect research-based ecological landscaping. Inertia and tradition are big factors. But there are also enormous economic interests behind pesticides, herbicides, fertilizers, sterile cultivars, excessive mulching and hardscaping, etc.
In the meantime, those of us interested in enjoying our yards and supporting biodiversity will have to check the facts, which is hard to do with ChatGPT — or even with Google’s new AI-assisted search. Reddit’s native plant area has a fairly knowledgeable community. Another place to start is my list of Free Plant Databases.
For example, here's what wildflower.org, the most comprehensive database of North American native plants, has to say about horseweed (Conyza canadensis):
An annual that thrives on bare soil, Horseweed is soon crowded out as perennials become established. — wildflower.org
This is practically the definition of a ruderal or pioneer herb species. They colonize disturbed sites — like your yard — and then cede place to longer-lived species. In the meantime, the horseweed roots are feeding beneficial organisms in the soil, adding organic material both below and above ground, and protecting the soil from compaction.
Do not pull out the horseweed. If you don't like the look, you can it down, tossing the cut stems on the ground to add more organic material — called "chop and drop" — or putting them on your brush pile.
Dear Heather, We recently bought a new house and my wife has been using ChatGPT to design the garden there. Isn’t that cool? — Having Fun with Tech
[Cringe.] Maybe? Maybe not? Let’s ask ChatGPT and you tell me. Here’s the transcript of my conversation with ChatGPT, lightly edited for space.
ChatGPT designs a backyard
Me to the latest, paid version of ChatGPT (GPT-5): Make a picture of a beautiful suburban backyard
Chat GPT:
Me: Make a picture of a beautiful suburban backyard that qualifies as a National Wildlife Federation Certified Wildlife Habitat, where only paths and play and seating areas are covered in turf and the rest is filled with plants native to Rhode Island. Hide the fence with native vines, shrubs or trees and do not have any visible mulch.
ChatGPT:
Me: Make a picture of a beautiful suburban backyard following the instructions in [Zoe and my Design Your Yard slide presentation], where all the plants, except turf, are native to Rhode Island. Reflect the style of garden designer Piet Oudolf.
ChatGPT:
Me: That's lovely, but it doesn't include any destinations, including seating areas, which are an important part of the instructions in the document I shared. Would you please add two seating areas and one destination — maybe a child's play area made out of logs and other natural materials?
ChatGPT:
Me: That's better, but it's weird how a wood bench is facing two random chairs in the same area. Please leave the bench and take out the random chairs, create a dining area near the back door with a circular table and four chairs, and move the play area further from the house. Also, the instructions I gave you require a path circling the house, so make sure the path connects to the front yard on both sides of the house. Finally, make sure there's a pleasing balance to the entire layout.
ChatGPT:
Me: What's that thin brick line from the patio running in the path and then along the bed on the right?
ChatGPT: It looks like the generator added an unintended narrow edging strip — almost like a thin brick or paver border — starting at the dining patio, cutting briefly across the turf path, and then running along the right-hand planting bed.
That element isn’t part of your instructions or the workshop’s design approach. … If you want a clean, naturalistic look that’s true to the plan, I’d recommend removing that edging entirely so the turf path reads as one uninterrupted circuit. I can adjust the image so it’s gone and the planting blends naturally into the lawn.
Me: Yes, eliminate that edging.
ChatGPT:
Me: I can see visible mulch again. Why did you do that?
ChatGPT: It looks like when I removed the edging, the generator defaulted to adding a thin exposed soil/mulch strip between the turf path and the planting beds — probably as a visual “definition” line.
That wasn’t in your instructions and it breaks the continuous, naturalistic look you specified in the workshop guidance. In a real-world ecological design, you’d let the plants knit right up to the path edge, so you wouldn’t see any bare mulch.
If you want, I can regenerate this so the turf meets the planting directly, with plants leaning slightly over the path in places — no visible mulch or bare strip anywhere. That will keep it truer to your original request and the Piet Oudolf style.
Me: Yes, please regenerate the image as you suggest.
ChatGPT:
Me: That's better. Thank you. Can you create a list of all the plants you show?
ChatGPT: I can give you a best-guess plant list based on the appearance and the Rhode Island native focus you specified, but since this is an AI-generated image, the flowers are artistic approximations — not botanical photographs — so exact species identification isn’t guaranteed.
If you want, I can make you a Rhode Island–specific planting palette that perfectly fits the visual style of this image, covers bloom succession from spring to fall, and meets NWF Certified Wildlife Habitat requirements — so you could actually plant it.
Do you want me to put that together and create a planting map for you?
Me: Yes, please do that.
ChatGPT:
I’ve mapped the planting palette onto the design so you can see exactly where each species goes. Some of the generated text labels aren’t perfect (AI sometimes distorts names), so if you’d like, I can produce a clean, corrected legend. Would you like me to do that?
Me: Yes, and also correct the shapes of the plants; the ironweeds against the house look like shrubs. But first, would you please explain why the images don’t follow the instructions, include hallucinations, don't correspond to actual plants or their shapes, etc. Why are the images so incorrect?
ChatGPT: [Paused to think for 30 seconds.] Totally fair, and I’m sorry for the frustration. Short version of why the pictures went sideways:
The image model isn’t botanical. It learns “what gardens look like,” not species-level forms. So it will paint “something coneflower-ish” or “a purple tall thing,” but it doesn’t reliably keep Vernonia upright or distinguish Clethra from Itea.
It recreates the whole scene each time. Every “regenerate” is a new painting, so paths, seating, and screening drift—even if the prompt is the same.
Style > specificity bias. Saying “Oudolf style, RI natives, no mulch” competes with its training priors (edging strips, mulch rings, random patio borders), so those artifacts creep back in.
Text is notoriously garbled. Diffusion models are bad at clean labels; that’s why plant names kept melting.
A better way to get you exactly what you want
If you’re up for one more pass, I can switch from “photorealistic” to precise, schematic outputs where I have full control. … Say the word and I’ll produce the plan now using your workshop rules (no extra input needed). If you’d rather pause here, I get it.
Me: Sure, why not? Please produce a plan as you suggest based on our workshop rules.
ChatGPT: Great — I’ll follow your Design Your Yard Workshop rules exactly and build a scaled top-down planting plan. …
Me: One thing I asked for — and perhaps that was lost — was for anywhere that isn't turf to be native plants. In your sketch there's a lot of white space. If the “Loop Path” is for paths and destinations, what's in all that white space?
ChatGPT: Right — in your workshop rules, white space wouldn’t exist in the real garden.
Anywhere that isn’t a turf path or a destination should be densely planted with native plants (forbs, grasses, shrubs, trees) — tall enough that you wouldn’t walk through it.
In this sketch, the white space is just background from the drawing canvas. It’s meant to represent all the filled planting areas, but I didn’t shade or texture them, so visually it reads as “empty.” That’s my miss.
[Note: I found the above schema totally useless and gave up at this point.]
I’d love to hear your thoughts on this experiment in the comments below. Here’s to the lazy, hazy, crazy days of summer!
— Heather
P.S. If you’d like to watch a recorded version of Zoe and my yard design workshop, it’s currently available free on YouTube as part of the Less Lawn, More Life challenge as Turn That Spot Into a Plan.
That conversation with ChatGPT is hysterically funny. I would never use it in any case, but OMG, what awful designs!
Please also speak to the horrid ecological impact of just using ai/ chatgpt has on the world in general. The amount of energy and water required is abhorrent and unsustainable. Pick up a book, talk to a local expert instead… please!