AI Writing Tools: When AI Should Help You Think, Not Just Write
AI writing tools have made it extremely easy to produce a draft.
Give an AI a topic, a few instructions, and a minute later you may have an introduction, an outline, or an entire article.
But writing does not always begin with a blank page.
Sometimes you already have:
- a vague idea
- a few notes
- screenshots you want to understand
- research material
- several possible arguments
- an earlier draft
- feedback from other people
- an AI response that is useful, but not quite right
In these situations, the hardest part is not generating text.
It is developing the thinking behind the text.
That distinction changes what an AI writing tool needs to do.
The short answer
There are two broad ways to use AI for writing:
Generation-first: ask AI to produce the text.
Development-first: use AI to help develop the ideas, material, structure, and arguments before writing the final text.
Generation-first tools are useful when speed matters and the cost of a generic draft is low.
Development-first workflows are more useful when the writer's own reasoning, voice, research, or judgment matters.
The second approach does not mean that AI should never write.
It means writing is the end of the process, rather than the entire process.
Generation-first AI writing
The simplest AI writing workflow looks like this:
Prompt → AI → Draft
You provide a topic, audience, tone, and perhaps a few key points.
The model produces something that looks like writing.
This is genuinely useful.
For example, you might need:
- five alternative headlines
- a rough outline
- a short product description
- a first-pass email
- several variations of an advertisement
- a deliberately disposable draft
When the main constraint is time, generation-first tools can be extremely effective.
The problem appears when the generated text is mistaken for the thinking process itself.
A polished paragraph can hide unresolved questions.
An elegant outline can contain the wrong argument.
A fluent article can quietly flatten the writer's original idea.
And once the model has produced a large amount of text, it can become tempting to edit the output instead of reconsidering the underlying idea.
Development-first AI writing
A development-first workflow starts somewhere else.
Instead of asking:
“Can AI write this for me?”
you ask:
“Can AI help me figure out what I actually want to say?”
The process might look like:
Idea → material → questions → connections → structure → revision → writing
At the beginning, the material does not have to be organized.
You might have a paragraph in your notes, three screenshots, a quotation from a book, an interesting contradiction, and an AI response that you don't completely agree with.
The point is not to turn these into polished prose immediately.
The point is to work with them.
You might ask AI to:
- challenge an assumption
- find a missing argument
- compare two interpretations
- connect two apparently unrelated ideas
- identify contradictions
- suggest questions worth exploring
- reorganize material
- rewrite only after the underlying argument is clear
Here, AI becomes less like a ghostwriter and more like a set of thinking tools.
Which approach is better?
Neither is universally better.
It depends on what kind of writing you are doing.
| Situation | Generation-first | Development-first |
|---|---|---|
| Need ten headline ideas | Excellent | Unnecessary |
| Need a disposable first draft | Excellent | Probably excessive |
| Writing from substantial research | Limited | Strong |
| Developing an original argument | Limited | Strong |
| Maintaining a distinctive voice | Requires careful editing | Stronger control |
| Exploring an unfinished idea | Useful | Particularly useful |
| Writing something where your judgment matters | Needs supervision | Better fit |
The important distinction is not AI writing vs. human writing.
It is whether the AI is generating the thinking or participating in the thinking.
What happens to the material?
This is where the choice of workspace starts to matter.
In a conventional chat, your material often gets converted into messages.
You send a note.
The AI replies.
You ask another question.
The AI replies again.
Eventually, useful ideas may be spread across dozens of messages.
The conversation is still there, but the individual pieces of thinking become harder to manipulate.
A development-first workflow benefits from treating those pieces as objects.
A research note can remain a note.
A screenshot can remain a reference.
An AI response can remain an AI response.
An argument can remain an argument.
You can put them next to each other and decide what they mean later.
This is one reason a canvas can be useful for writing even when the canvas itself is not a writing editor.
Where common AI writing tools fit
ChatGPT, Claude, and Gemini can all be used for development-first writing if you deliberately structure the conversation around questioning, comparison, and revision.
Dedicated AI writing products often put more emphasis on producing or editing prose directly.
Note-taking tools such as Notion and Obsidian can provide a persistent place for research and drafts, while their AI features can help transform or summarize that material.
A canvas adds another possibility:
keep the material visible before deciding what it becomes.
That is a different workflow rather than simply another AI model.
Where Spoor fits
Spoor is not primarily an AI writing generator.
It is a canvas for developing ideas.
You can bring notes, screenshots, documents, references, and AI responses onto the same board. Each piece remains editable, and you can arrange the material spatially instead of forcing everything into one linear document.
Then AI can work on the material that you have selected.
You might place:
an unfinished idea
next to
three research references
next to
a screenshot
next to
an AI objection
and then ask an AI persona to examine the group.
The response becomes another card.
You can keep it.
Move it.
Challenge it.
Ignore it.
Or use it as the starting point for another round of thinking.
Eventually, some of those cards may become an outline.
The outline may become a draft.
And the draft may become something else entirely.
But that transformation happens after you have developed the idea, not because an AI decided what the finished piece should look like.
The writer still makes the important decisions
A development-first AI workflow is not about pretending that AI has no role in writing.
It has a very important role.
AI can generate alternatives quickly. It can expose assumptions you missed. It can approach the same material from different perspectives. It can help you see connections that were difficult to notice.
But the writer still decides:
- which idea is worth pursuing
- which source is trustworthy
- which objection matters
- which argument is convincing
- which suggestion is irrelevant
- what should be removed
- what the final piece should actually say
That distinction matters because a good AI writing workflow should preserve the writer's ability to make those decisions.
So, what is the best AI writing tool?
If you simply need text, a conversational AI or dedicated AI writer may be enough.
If you are developing a complicated argument, researching a topic, building a long-form project, or trying to preserve your own thinking throughout the process, the more useful question is different:
Where can I keep all the material while the idea is still changing?
That might be a document.
It might be a knowledge base.
It might be a whiteboard.
Or it might be an infinite canvas.
Spoor is built around the last option.
Not because every piece of writing needs a canvas.
But because some writing starts long before there is anything worth calling a draft.
And when the idea is still unfinished, giving it room to develop can be more useful than asking AI to finish it.