AI Writing Tools: What They're Actually Good At (And What They're Not)

AI writing tools are not useless.

They are also not magic interns.

They are extremely good at some writing-adjacent work and suspiciously bland at the work people actually want to publish.

That distinction saves you money, time, and dignity.

OpenAI’s 2026 guide to writing with ChatGPT gets the useful frame right: plan, draft, revise, package. Notice what is missing.

“Press button. Become interesting.”

Because that is not a feature.

The rule

Use AI writing tools for friction.

Do not use them for taste.

Friction is:

• Starting from blank
• Organizing messy notes
• Rewriting for length
• Turning one format into another
• Finding weak logic
• Creating boring first drafts

Taste is:

• Knowing the point
• Having a real opinion
• Choosing what to cut
• Sounding like a person
• Understanding your reader
• Saying the thing with nerve

AI can help with the first list.

You still own the second.

If you do not, your writing turns into competent fog.

Good use 1: rough outlines

The move: Give the tool your actual points and ask for structure.

Do not ask:

“Write a blog post about productivity.”

Please do not summon another oatmeal article into the world.

Ask:

“Here are my messy notes. Turn them into three possible outlines. Keep the strongest claim first. Make the flow practical, not academic.”

Then judge the options.

The tool is not deciding.

It is arranging.

That is a real use.

This is especially helpful if you think in piles. Some people do not think linearly. They think in scraps, gripes, examples, screenshots, and half-sentences.

Good.

Dump the pile.

Make the tool sort it.

Then you decide what survives.

Good use 2: boring first drafts

Some writing does not need your soul.

It needs to exist.

Examples:

• Meeting follow-up emails
• Internal updates
• Basic FAQs
• Product descriptions
• Outreach variations
• Summaries of a known process

The move: Use AI to make the ugly first version faster.

Then fix it.

Do not publish the first output unless the stakes are tiny.

And if the stakes are tiny, fine. Let the machine do the tiny thing.

Your life is short.

But for anything that carries your reputation, the AI draft is clay.

Not sculpture.

Good use 3: clarity edits

AI is solid at noticing clutter.

It can shorten.

It can point out repetition.

It can ask where the argument gets vague.

It can convert a dense paragraph into bullets.

The move: Ask for a diagnosis before a rewrite.

Use prompts like:

• “Where does this section lose clarity?”
• “Which sentences repeat the same idea?”
• “What does this paragraph promise but not prove?”
• “Make this easier to scan without changing the claim.”

That keeps you in charge.

If you ask it to “make this better,” you deserve the beige output you get.

Be specific.

Good use 4: repackaging

This is one of the best uses.

You already did the thinking.

Now you need the same idea in five shapes.

The move: Feed it the finished thing and ask for format shifts.

Examples:

• Article into newsletter summary
• Long post into social thread
• Notes into checklist
• Draft into FAQ
• Client update into executive summary

That is not cheating.

That is not outsourcing your voice.

That is using the dishwasher instead of hand-washing every fork.

If you want the non-AI version of speed, tools that help you write faster with AI is useful, but do not confuse speed with thought.

Bad use 1: final voice

AI can imitate style.

It cannot care.

That is the crack.

It can add contractions, short lines, spicy phrases, and fake certainty.

But if there is no actual belief underneath, readers feel it.

The problem is not that AI writes badly.

The problem is that it writes plausibly.

Plausible is dangerous because it lets weak thinking pass as polished.

Do not let it.

Bad use 2: unsupported claims

AI can sound confident while being wrong.

Everyone knows this.

Then everyone forgets it the second the sentence sounds smooth.

The move: Treat every factual claim as untrusted until checked.

Especially:

• Numbers
• Dates
• Legal claims
• Medical claims
• Financial claims
• Quotes
• Tool features
• Pricing

Do not publish mystery facts because the paragraph had rhythm.

That is how you build a beautiful little liability.

Bad use 3: opinions you have not earned

AI can argue both sides of anything.

That is useful for testing.

It is terrible for conviction.

If you ask it for a bold take, it will often produce the costume of a bold take:

• Strong opening
• Neat contrast
• Confident bullets
• Zero scar tissue

That is not voice.

That is theater.

A real opinion usually costs something. You tried the thing. Lost time. Changed your mind. Annoyed someone. Noticed a pattern other people keep politely avoiding.

AI can help you phrase that.

It cannot give you the scar.

The workflow that works

Use this.

1. Brain dump first

Write your messy opinion before you ask the tool anything.

Even ten ugly bullets are enough.

Your raw material matters.

No raw material, no voice.

2. Ask for structure

Ask for outlines, missing points, reader objections, and simpler flow.

Keep what helps.

Ignore the rest.

3. Draft ugly

Let the tool draft boring sections if needed.

But rewrite the opening, transitions, and final claim yourself.

Those are where the human shows up.

4. Edit for clarity

Ask it to find clutter.

Then you cut.

Do not let it polish every weird line into office furniture.

5. Final human pass

Read it aloud.

Ask:

• Do I believe this?
• Does this sound like me?
• Is the useful action clear?
• Did I verify anything factual?
• Would I be mildly embarrassed if this went viral?

Good.

That last one keeps you honest.

The short version

AI writing tools are good at:

• Starting
• Sorting
• Summarizing
• Repackaging
• Clarifying
• Drafting low-stakes formats

They are bad at:

• Taste
• Lived opinion
• Specific voice
• Original stakes
• Knowing your reader
• Caring when the claim is wrong

Use them.

But do not hide behind them.

If you are trying to build an actual writing practice, the freelance writing starter pack is the slower, better foundation. If your worry is whether clients can tell, read will clients know you used AI.

The answer is simple:

They may not know you used AI.

They will know if the work got generic.