I’m a heavy AI user — partly because my work involves a lot of writing and research, partly because keeping pace with AI tools is part of how I justify my fees to coaching clients. In the last 6 months I’ve tested 15 AI productivity tools in real workflows. Here’s what actually earned its spot in my stack.
Claude — best for writing and serious analysis
Cost: Free tier; Pro at $20/month. Best for: writing, long-form analysis, document work.
For long-form writing and structured analysis, Claude is genuinely the best in class as of mid-2026. The 200K-token context window means I can paste in a 50-page document and ask specific questions. Voice and tone tracking across long conversations is noticeably better than ChatGPT. I draft about 60% of my client communications and 80% of my long-form content with Claude as a thinking partner.
ChatGPT — best general-purpose, biggest ecosystem
Cost: Free tier; Plus at $20/month; Team at $30/user/month. Best for: general-purpose tasks, image generation, voice mode, custom GPTs.
Still the most widely useful generalist AI. The image generation (DALL-E inside ChatGPT) is good enough for casual use. Custom GPTs let me bundle prompts I use often (e.g., “interview question generator”). Voice mode is genuinely useful for brainstorming on walks.
Perplexity — best for research with sources
Cost: Free tier; Pro at $20/month. Best for: research, citing sources, current events.
Perplexity has earned a permanent spot in my workflow because every answer comes with sources I can verify. For coaching research (“what’s the current UK graduate scheme application deadline for PwC?”), it’s much more reliable than asking ChatGPT or Claude, which can hallucinate.
Granola — best AI meeting notes
Cost: Free trial; £14/month. Best for: anyone in 5+ meetings/week.
I covered this in my productivity apps post — Granola has changed how I work post-meetings. Runs in the background, captures audio, generates structured notes with decisions and action items. Otter is the more mature alternative; I find Granola’s summaries cleaner.
Reclaim.ai — best for AI-driven calendar scheduling
Cost: Free tier; Pro at $10/month. Best for: people drowning in meetings who want auto-scheduling.
Reclaim looks at your calendar, your tasks, and your habits, then auto-schedules focus time, exercise, and 1:1s into the gaps. Genuinely useful if your week is meeting-heavy. Less useful if you already plan deliberately.
Notion AI — best for AI inside an existing knowledge base
Cost: £8/month add-on. Best for: Notion users who want AI summarisation and drafting inline.
The Notion AI add-on shines when you’re working inside Notion already — summarise a long page, generate a meeting prep from notes, translate inline. Less compelling as a standalone tool. Worth it if you live in Notion.
Tools I tested and dropped
- Gemini — improved a lot in 2025 but didn’t surpass Claude or ChatGPT for my use cases.
- Mem — interesting AI-first notes app but I never trusted retrieval enough to commit knowledge to it.
- Superhuman AI — solid AI email features but I prefer composing email myself.
- Lex — AI writing tool, redundant once Claude landed long context.
- Otter.ai — solid meeting notes; replaced by Granola in my stack.
- Copy.ai / Jasper — marketing-copy AI, too specialised and lower quality than general models now.
- Tactiq — meeting transcription, fine but no advantage over Granola.
- Magical (Magic AI) — text expander with AI, too narrow a use case.
- Rytr — generalist writing AI, weaker than Claude/ChatGPT.
How to actually use AI without becoming dependent
My coaching clients fall into two camps with AI: those who use it to think faster, and those who use it to skip thinking. The second group’s work starts to feel hollow — and ironically performs worse over time as their own skills atrophy.
The principles I now teach:
- Brainstorm with AI, but write the final pass yourself
- Use AI for first drafts of structured documents (emails, briefs, summaries) where the structure is more important than the voice
- Never paste AI output into your own work without rewriting it — your voice is the differentiator
- Verify any factual claims AI makes — hallucination rates are still 5–15% for specific facts
- Keep AI out of your one core craft — for me that’s coaching conversations. The shortcut destroys the skill.
Privacy considerations
Free tiers of all major AI tools may use your inputs to train future models. For sensitive content (client work, confidential strategy, legal), use the paid tiers which typically opt out of training. UK + EU users — check GDPR provisions on AI tools handling personal data.
What about AI agents?
AI agents (Devin, Auto-GPT, OpenAI’s Operator) had a hype peak in 2024-2025 and are stabilising in 2026. For repetitive structured workflows (data extraction, simple research), they’re useful. For complex multi-step reasoning, current agents are still error-prone enough that I find them more time-consuming to manage than to do the work myself. Worth re-evaluating in 12 months.
Related reading
For the broader productivity stack, see the best productivity apps for knowledge workers. For deep work specifically, the best morning routine. And to build habits around new tools, how to build a habit that sticks.
FAQ
Q: Should I pay for Claude AND ChatGPT?
For heavy users, yes — they’re genuinely different. For light users, pick one.
Q: Will AI replace my job in 2026?
For most knowledge-work roles, AI is augmenting not replacing. Roles most at risk: pure summarisation, basic data entry, low-skill writing. Roles least at risk: anything requiring judgement, relationship, or accountability.
Q: How much time does AI actually save?
For writing-heavy roles, 5–10 hours/week in my experience. For meeting-heavy roles, 2–4 hours/week from notes alone.
Q: Should I disclose AI use at work?
Yes, for any final deliverables. Increasingly, employers expect transparency.
Q: Best free AI tool?
Claude.ai (free tier) — most usage allowance for thoughtful work without paying.
