We’ve already admitted that we’re using AI for more than just designing the garden. It’s helping us write about the process too.

But we’ve discovered something important: the quality of the advice depends heavily on the quality of the information we give it.

What worked well?

The biggest improvement came when we stopped asking general questions and started giving AI the specifics.

Photographs helped it understand what the garden actually looks like. A measured plan helped it understand the proportions and awkward boundaries. And mapping the existing planting made an even bigger difference.

Once we explained that we already have a Red Robin, Choisya, Buddleia, Ceanothus, roses, Geums, Sea Hollies, Bottlebrush and Hostas, the planting recommendations changed dramatically.

Instead of suggesting lots of new shrubs, the advice became:

Use what we’ve already got and add only what is missing.

That’s much more useful.

What hasn’t worked?

Questions such as:

‘What drought-tolerant plants would work in my garden?’

or:

‘What should I plant in this border?’

produce perfectly reasonable answers – but they’re generic.

Without knowing our soil, aspect, existing plants, available space and priorities, AI can’t really design our garden.

We’ve also found that simply asking for ‘ideas’ tends to produce too many ideas!

Sometimes what we actually need is for AI to tell us what not to buy.

What makes a good prompt?

We’re now finding that the best prompts include:

The facts:
Garden dimensions, aspect, soil, drainage and existing structures.

The existing planting:
What’s already there, what we want to keep and what we’re removing.

Our priorities:
For example, drought tolerance, low maintenance, more shade, child and pet friendliness – and retaining plenty of lawn.

Our preferences:
The style we’re aiming for and the things we definitely don’t want.

The constraints:
For example, the pergola only covering the patio and the summer house needing to fit a particular space.

And finally, perhaps the most useful instruction:

‘Challenge the plan. Tell me what you’d change, what isn’t necessary and what I shouldn’t buy.’

That last part is important.

We’re not looking for an AI that enthusiastically fills every empty space.

We’re looking for one that understands that an empty space can be useful too.

Our new approach

So going forward, our prompts will look less like:

‘What would look good here?’

and more like:

‘Here’s my garden, its measurements, conditions, existing plants, priorities and constraints. Revise the design around what is already there. Be realistic about mature plant sizes and spacing, and tell me what I should add, move, remove – and what I shouldn’t buy.’

That seems to produce much better advice.

And perhaps that’s the biggest lesson we’ve learnt so far:

AI doesn’t need us to ask better questions. It needs us to give it a better picture of the problem.

Which, as it turns out, means we have to understand our own garden rather better too.

Blooming Lovely Chaos, but with a plan.