Used well, AI is an extraordinary writing partner. It can draft emails, structure arguments, brainstorm ideas, rewrite clumsy sentences, summarise long documents, and help you think through problems.

You are not replacing your own thinking. You are giving yourself a sounding board that is available 24 hours a day and never gets tired of your questions.

Which AI Tool for Writing

Claude
What it isAI assistant built by Anthropic. Current models (May 2026): Sonnet 4.6 (daily driver), Opus 4.7 (most powerful, 1M-token context, released April 2026), Opus 4.6 (still available), Haiku 4.5 (fast and cheap)
Best atLong-form writing, nuanced analysis, following complex instructions, working with large documents, coding, Projects for persistent context, Cowork for task automation, inline charts and infographics
Free tierFree tier with limited daily messages using Sonnet
First paid tierClaude Pro - US$20/month (~A$32/month)
Ken's takeThis is the tool I use most. Claude produces cleaner, more natural writing than ChatGPT. It follows complex instructions better, handles long documents well, and is less likely to give you that generic AI voice. The free tier is enough to try it. Pro at around A$32/month is worth every cent if you use it daily. For heavy use, Max plans run A$160-320/month. I use Max only when a project justifies the cost.
Sign uphttps://claude.ai
ChatGPT
What it isAI assistant built by OpenAI. Current models (May 2026): GPT-5.5 (newest, April 2026, plus a Pro variant), GPT-5.4 and GPT-5.2 Instant still in rotation, plus dedicated reasoning models
Best atGeneral-purpose conversations, quick answers, image generation (built in via GPT Image 1.5), wide plugin ecosystem, Custom GPTs, voice conversations
Free tierFree tier with access to GPT-5.2 Instant and limited GPT-5.5 usage
First paid tierChatGPT Plus - US$20/month (~A$32/month)
Ken's takeChatGPT was first to market and has the largest user base. It is very capable and moves fast. OpenAI has pushed aggressively through GPT-5 and beyond, with releases every couple of months. I personally find Claude writes more naturally and follows instructions with more precision, but ChatGPT has a broader feature set including built-in image generation and voice mode. Many people use both.
Sign uphttps://chat.openai.com
Google Gemini
What it isGoogle's AI assistant. Current models (May 2026): Gemini 3.1 Pro (most powerful, April 2026), Gemini 3.1 Flash (fast and efficient), Gemini 3.1 Flash-Lite (most cost-efficient)
Best atDeep integration with Google Workspace (Gmail, Docs, Sheets, Slides, Drive), 1-million-token context window across the line, video generation via Veo, image generation, Deep Research, Google Search grounding
Free tierFree tier with daily limits using Gemini Flash. Includes basic chat, image input, and limited reasoning
First paid tierGoogle AI Pro - US$20/month (~A$32/month)
Ken's takeIf you live in the Google ecosystem (Gmail, Docs, Drive, Calendar), Gemini has a unique advantage. It integrates directly into those apps, meaning you can ask Gemini to draft an email in Gmail, analyse a spreadsheet in Sheets, or summarise your Drive documents without leaving them. The AI itself is competitive with Claude and ChatGPT. The free tier is generous. My honest view: for pure writing quality and instruction-following, Claude still wins. For the broadest feature set, ChatGPT wins. But for people who spend their day in Google apps, Gemini's integration advantage is significant.
Sign uphttps://gemini.google.com

The Google Ecosystem Advantage

If you use Gmail, Google Docs, and Google Drive daily, Gemini is worth trying alongside Claude or ChatGPT. Google AI Pro (US$20/month, ~A$32) gives you Gemini embedded in all your Google apps, plus upgraded NotebookLM with 5x more audio overviews and enhanced features. No other AI tool has this level of integration with Google's suite. However, the Gemini integration in Workspace apps requires the paid tier.

Power Features That Change Your Workflow

Claude Projects

Once you start using Claude regularly, Projects will change how you work. A Project is a persistent workspace where you can upload reference files (documents, spreadsheets, images), write custom instructions that apply to every conversation in that project, and have multiple chat sessions that all share the same context.

Why this matters: without Projects, every new conversation starts from scratch. With Projects, Claude already knows your background, your preferences, your reference material, and your specific requirements. I use Projects for everything from book writing (where Claude has the full manuscript and style guide) to job applications (where it has my resume and target role descriptions). The difference in output quality is dramatic.

Projects are available on the free tier with limitations, and fully on Claude Pro. If you use Claude for any ongoing piece of work, set up a Project for it.

Claude Cowork

Cowork is Claude's newest and most ambitious feature. It turns Claude from a chat assistant into something closer to an actual digital coworker. You point it at a folder on your computer, give it a task, and it gets to work: reading files, creating documents, organising content, extracting data, and completing multi-step workflows without you guiding every action.

Think of the difference this way: regular Claude shows you how to do something. Cowork does it. Need to sort a chaotic downloads folder? Compile expense data from a pile of receipts? Pull key dates and obligations from a stack of contracts? Cowork handles it.

It also connects to external tools like Gmail, Google Drive, and Slack, meaning it can pull information from your real work environment, not just files you manually upload. The Dispatch feature lets you control Cowork from your phone while the desktop app runs, so you can delegate tasks while you are away from your computer.

Cowork is available on Pro and Max plans. It is still in research preview, which means it is powerful but has rough edges. Complex spreadsheets can confuse it, and some tasks run slower than you would expect. But the trajectory is clear: this is where AI assistants are heading.

Persistent Instructions and Memory

This is one of the most underused features across all the major AI tools, and one of the most transformative once you set it up.

Both Claude and ChatGPT allow you to provide standing instructions that apply to every conversation, not just one. You tell the tool who you are, how you like to work, and how you want it to respond, and it remembers this across every new chat.

Claude has two mechanisms. First, user preferences (found in Settings) where you can write instructions about your preferred tone, formatting, and how you want Claude to behave. Second, memory: Claude learns from your conversations over time and builds up knowledge about you, your name, your work, your preferences, your ongoing projects, that it draws on in future chats. You can view, edit, and delete what Claude remembers. You can also explicitly tell Claude to remember something: "Remember that I work in IT procurement and I prefer concise responses with no jargon."

ChatGPT has a similar setup. Custom Instructions (found in Settings > Personalization) lets you provide two blocks of text: one describing who you are and what you do, and another describing how you want ChatGPT to respond. Memory works automatically. ChatGPT picks up facts from your conversations and stores them for future reference. You can review and manage what it has stored in Settings > Personalization > Memory.

Gemini has memory capabilities as well. It learns preferences from your conversations and, on paid plans, integrates with your Google account context (your emails, calendar, documents) to provide more personalised responses.

Why this matters: Without persistent instructions, every new conversation starts from zero. The AI does not know your name, your role, your industry, or your preferences. You end up repeating the same context every time. With instructions and memory set up, the AI already knows that you are an IT professional based in Melbourne who prefers direct communication and works in procurement risk. The quality and relevance of every response improves immediately.

What to put in your instructions: Your name and location. Your profession and industry. Your communication preferences (concise vs detailed, formal vs casual). Any recurring context ("I frequently ask about Australian law," "I work with government procurement," "I prefer metric units"). Formatting preferences ("Use short paragraphs," "Avoid bullet points unless I ask for them," "Never use emojis"). The more specific you are, the better the results.

Try this right now

Open Claude or ChatGPT settings. Find the preferences or custom instructions section. Write 3-5 sentences about who you are and how you want the AI to respond. Then start a new conversation and notice how the responses are already tailored to you without having to explain your context.

Managing Your AI Workflow

Here is something nobody tells you when you start using AI tools: if you do not develop good habits early, you will drown. I have thousands of conversations across Claude, ChatGPT, and other tools. Without a system, finding that brilliant prompt you wrote three months ago, or the document Claude produced for a specific project, becomes a needle-in-a-haystack problem. Good admin in this space saves an enormous amount of time.

Save your outputs. This is the most important habit to develop. Documents, reports, code, and other outputs you create in AI conversations do not persist forever. Conversations can be deleted, platforms change, and finding something buried in a six-month-old chat thread is painful. When an AI tool produces something you want to keep, save it immediately: download the file, copy the text to a document, or export it. Do not assume you can find it later.

Build a prompt library. When you write a prompt that works well, save it. Over time, you will develop prompts for recurring tasks: a specific way you ask for email rewrites, a template for competitive analysis, a format for meeting summaries, a prompt that gets the AI to review your writing in a particular way. Store these somewhere accessible.

NotebookLM is excellent for this. Create a notebook called "My AI Prompts" and add your best prompts as sources, organised by category. You can then ask NotebookLM to help you find and refine prompts, or generate variations for new use cases. Alternatively, a simple Google Doc, Notion page, or even a text file works. The tool matters less than the habit.

Name your conversations. Most AI tools let you rename conversations. Instead of accepting the auto-generated title (which is usually vague), rename important conversations with something descriptive: "Q1 Budget Analysis - March 2026" or "Website Copy - Assure Advantage" or "Cover Letter - DXC Role." This takes two seconds and saves ten minutes of searching later.

Use Projects and folders. Claude's Projects feature lets you group related conversations with shared files and instructions. ChatGPT has a similar Projects feature. Use them. A project for each client, each major piece of work, or each ongoing task area means your conversations are pre-organised rather than dumped into a single chronological stream.

Search works, but only if you remember keywords. Both Claude and ChatGPT now offer conversation search. Claude also has memory that draws on past conversations. These features help, but they work best when you can remember roughly what you were working on. Good naming and project organisation make search much more effective.

Periodic cleanup. Every few weeks, spend ten minutes archiving or deleting conversations you no longer need. This reduces clutter and makes the important stuff easier to find. Think of it like cleaning your email inbox. Nobody enjoys it, but everyone benefits from doing it.

A Simple System to Start With

Create three things today: (1) A document or notebook for your best prompts, organised by task type. (2) A folder on your computer called "AI Outputs" where you save anything worth keeping. (3) A habit of renaming every important AI conversation with a descriptive title as soon as you start it. These three habits will save you hours over the coming months.

ChatGPT Custom GPTs

ChatGPT has a feature called GPTs (sometimes called Custom GPTs) that lets you build specialised mini-apps. You give a GPT a name, a set of instructions, and optionally a knowledge base of files. It then behaves like a purpose-built tool for a specific task.

A real example: when I had a dispute with a former landlord who was being unreasonable, I built a negotiation GPT. I used Perplexity to research negotiation tactics and compiled them into a knowledge base. I then trained the GPT with instructions on how to apply those tactics to tenancy disputes under Australian law. Every time the landlord sent a message, I ran it through my GPT and got back a strategically crafted response. It worked. The landlord backed down.

You can build GPTs for anything: a writing style guide checker, a meal planner that knows your dietary restrictions, a tutor for a subject you are studying, or a customer service response drafter. GPTs are available to ChatGPT Plus subscribers and can be shared with others.

Choosing the Right Model

Both Claude and ChatGPT offer multiple AI models, and choosing the right one matters more than most people realise. The models change frequently. The summary below was verified in May 2026 and you should expect names and version numbers to drift every six to twelve weeks.

Claude models (as of May 2026): Claude Sonnet 4.6 is the default and daily workhorse. It has improved so dramatically that, for the first time, developers often prefer it over the previous generation's top-tier model. For most tasks, Sonnet is all you need. Claude Opus 4.7 (released April 2026) is the most powerful generally available model, with a step-change improvement in agentic coding over Opus 4.6. Opus 4.6 is still around for users who prefer it. The 1-million-token context window is generally available across the line under unified pricing. Claude Haiku 4.5 fills the speed-and-cost tier. Use Sonnet for everyday tasks. Switch to Opus when you need the AI to process something very large or think very deeply.

ChatGPT models (as of May 2026): OpenAI has moved fast. GPT-5.5 (released April 2026) is the newest model, deploying to Plus, Pro, Business and Enterprise tiers, with a GPT-5.5 Pro variant available on Pro and above. GPT-5.4 (the previous flagship) and GPT-5.2 Instant remain in rotation; GPT-5.1 was retired in March 2026. For reasoning-heavy tasks (maths, logic, coding, complex analysis), there are dedicated reasoning models that think step by step before answering and produce significantly better results on hard problems. The pace of releases is relentless; expect this paragraph to be stale within a couple of months.

Gemini models (as of May 2026): Google's current line is Gemini 3.1. Gemini 3.1 Pro (released April 2026) is the most capable and is optimised for complex multi-step agentic workflows. Gemini 3.1 Flash is the fast/cheap tier and now outperforms older Pro models on benchmarks. Gemini 3.1 Flash-Lite (released March 2026) is the most cost-efficient, designed for high-volume traffic. The 1-million-token context window is available across the line. Free tier users get Flash. Google AI Pro subscribers get access to 3.1 Pro and integration with Google Workspace apps.

The practical rule: Start with the default model. If the answer feels shallow, incomplete, or wrong, switch to the more powerful model and try again. You will quickly learn when the upgrade matters and when it does not.

Try this right now (free)

Open Claude or ChatGPT. Paste a work email you have been putting off and type: "Rewrite this to be more concise and direct, but keep it friendly. Keep it under 100 words." See what comes back. Then say "a bit more formal" or "add a line thanking them for their patience." That back-and-forth is how you get the most from these tools.

Specialist Writing Tools

The big chat tools above are the right starting point for most writing tasks. But there is a small set of tools built specifically for one kind of writing where they genuinely beat a general AI. If you are writing fiction, editing a manuscript, producing marketing copy at volume, or paraphrasing as an ESL writer, these earn their place.

Sudowrite
What it isAn AI writing tool built specifically for fiction. The Muse model is fiction-trained, so prose comes out sounding like fiction rather than generic AI
Best atScene generation, character development, plot brainstorming, the Story Bible feature for keeping continuity across a long manuscript
Free tierNo free tier (free trial available)
First paid tierHobby - US$10/month annual (~A$16/month)
Ken's takeIf you write fiction, Sudowrite is the one I would try first. The output reads like fiction, not the bland Claude-or-ChatGPT prose, and the scene-level tools match how novelists actually work. It is credit-metered, so heavy users will hit caps. If you want continuity across a long series with hundreds of characters, NovelCrafter (below) handles that better. For short or standalone fiction, Sudowrite is enough.
Sign uphttps://sudowrite.com
NovelCrafter
What it isA writer's IDE for novel-length fiction, with a structured Codex for characters, locations, factions, magic systems and items
Best atMaintaining continuity across hundreds of thousands of words. The AI uses your Codex when generating, so consistency holds across a series rather than drifting
Free tierNo free tier (free trial available)
First paid tierHobby - US$4/month (~A$6/month)
Ken's takeIf you are writing a fantasy series with a complex world, NovelCrafter is built for the job. The Codex is the killer feature. Sudowrite produces more naturally readable prose; NovelCrafter wins on long-form continuity. The bring-your-own-key support means you can plug in your own OpenAI or Anthropic account and avoid paying for credits twice. Steeper learning curve than Sudowrite. Not the right pick for short fiction or non-fiction.
Sign uphttps://novelcrafter.com
ProWritingAid
What it isA manuscript-level editor for longer writing projects. Where Grammarly fixes commas, ProWritingAid runs reports on pacing, repetition, dialogue tags and consistency across a whole document
Best atDiagnostic editing on long-form writing. Authors and editors working on books or theses use it to spot the patterns a sentence-level checker misses
Free tierFree tier with limited daily checks
First paid tierPremium - US$30/month (~A$48/month) or US$120/year. One-time lifetime licence at US$399 avoids the subscription
Ken's takeThis is what you reach for when Grammarly is not enough. The reports on sticky sentences, overused words and pacing genuinely improve a manuscript. The interface feels dated next to newer competitors and it is slower than real-time tools, but no other tool does the deep diagnostic pass at this price. The lifetime licence is worth it if you write seriously, since you avoid the subscription forever.
Sign uphttps://prowritingaid.com
Hemingway Editor
What it isA plain-English readability checker. Not actually AI, completely rule-based: highlights long sentences, passive voice and adverbs
Best atFast, predictable readability checking when you do not want anything trained on your text. Web version is free and requires no account
Free tierWeb version is free and unlimited
First paid tierDesktop app - US$19.99 one-time payment (~A$32 once)
Ken's takeNot really an AI tool, but it earns its place by being simple, free and useful. I run business writing through it before sending. It suggests aggressive cuts, sometimes too aggressive (Hemingway himself would not pass it), so use judgement. The desktop app at $20 once is the right move if you want it without an internet connection. For anything beyond plain-English checking, Claude or ChatGPT will do more.
Sign uphttps://hemingwayapp.com
QuillBot
What it isA paraphraser and grammar checker. Paste something you have written and QuillBot rewrites it in a different tone or fixes the grammar
Best atESL writers, students, and anyone who needs to rephrase or summarise existing text quickly. The free tier is genuinely useful
Free tierFree tier with paraphrasing, grammar checking and a summariser
First paid tierPremium - US$8.33/month annual (~A$13/month) or US$19.95/month
Ken's takeUseful when you have something to say but want help saying it differently, particularly for non-native English speakers. The paraphrase modes (Standard, Creative, Formal) feel like a workaround for what Claude or ChatGPT can do directly with a better prompt, but for someone who just wants a button rather than a prompt, QuillBot is friendlier. Output can read mechanical compared to a chat tool with good instructions. Try it free; only upgrade if you use it daily.
Sign uphttps://quillbot.com
Jasper
What it isA marketing-focused AI writing platform with templates, brand voice training and team workflows
Best atMarketing teams that need to produce a lot of on-brand content quickly across blog, ads and email. Brand voice training is the best in the category
Free tierNo free tier (free trial available)
First paid tierCreator - US$39/month annual (~A$62/month)
Ken's takeIf you run a marketing team and need consistent brand voice across many writers, Jasper does that one job better than a general AI. For a solo writer or small team, the price is hard to justify when Claude or ChatGPT with a custom system prompt produces output that is just as good. Jasper's value is the workflow layer (templates, brand voice memory, team accounts), not the underlying AI quality. Worth the price for a marketing team. Probably not worth it for individuals.
Sign uphttps://jasper.ai
Writesonic
What it isOriginally a Jasper competitor, Writesonic has pivoted to AI search visibility tracking. It monitors how often your brand or business is cited in ChatGPT, Perplexity, Gemini and Claude answers
Best atSEO and brand teams who want to see whether AI chatbots mention their company when people ask related questions. Tracks share-of-voice and sentiment across the major AI search platforms
Free tierFree trial
First paid tierLite - US$39/month (~A$62/month). Entry pricing tripled in 2026
Ken's takeAn interesting pivot. As more people search by asking a chatbot rather than typing into Google, brands need to know whether they are showing up in those answers. Writesonic is the most established tool for this. The pricing has climbed sharply in 2026, which makes it hard to justify unless you genuinely depend on AI search traffic. The "AI writing" angle is now secondary, so do not go to Writesonic if you just want to write content. Use it for visibility tracking only.
Sign uphttps://writesonic.com
Get better results. The quality of your AI writing depends heavily on how you prompt it. If you have not already, read How to Talk to AI for techniques that make a real difference. For turning your writing into polished slide decks, see Presentations.