How to choose when you work on your own
When you run your own practice, every hour on admin is an hour you can't bill or sell. That's why AI tools are tempting, and why it's easy to end up paying for six of them and using two.
Start from the job. Look at last week and find the one task that took the most unbillable time. For most independent consultants and coaches it's one of five: writing proposals, writing up meetings, research, chasing follow-ups, or replying to new enquiries. Pick a tool for that one job, use it for two weeks, and time it against how long the job took before.
The tools below are sorted by those five jobs. Each is described from its own website, read on 28 September 2026. Features change, so check each site before you sign up. AI sales tools covers more tools for the wider sales job.
Five jobs and the tools for each
1. Proposals and writing
A general AI assistant is the tool here. You give it your notes from the discovery call and your usual proposal structure, and it gives you a first draft to rewrite.
- ChatGPT describes itself as help with drafting messaging, explaining concepts, searching the web and thinking through a decision, and lists ChatGPT Work, which creates and edits documents, slide decks and spreadsheets.
- Claude lists writing, editing and creating content, and research, and says you can hand it a task rather than only a question.
Keep a proposal outline you trust and paste it in each time. The consulting proposal template is one to start from.
Here are my notes from a discovery call: [paste].
Here's my proposal structure: [paste].
Write a first draft. Use the client's own words for the problem. Don't add any fact, number or promise that isn't in my notes. Mark anything you're unsure of with [CHECK].
2. Meeting notes
An AI notetaker records or listens to a call and writes a summary with the actions. For a consultant, it's the difference between writing up at 9pm and not writing up at all.
- Otter transcribes meetings, which you can send it to or record from your device, and has an AI chat that searches across your meetings.
- Fathom summarises meetings, with searchable transcripts and follow-ups, can run with or without a meeting bot, and offers a free plan.
- Granola is an AI notepad that uses your computer's audio instead of adding a bot to the call, and works with Zoom, Google Meet and Teams.
- ActivityTracker's notes let you talk your meeting notes in, snap paper forms and turn the promises into tasks. It's built for writing up after a call, and it doesn't record the call itself.
Always tell the client you're recording, and ask first.
3. Research
Before a first call, you want to know the client's business, recent news and the questions worth asking, and a general assistant will pull that together. ChatGPT lists Deep Research and web search, and Claude lists research among its tasks. Ask for sources and open the ones your advice rests on. A confident summary can still be wrong.
I have a first call with [role] at [company] about [topic].
From their website and recent news, tell me: what they do, what seems to have changed this year, and three questions worth asking. Give me the link for each fact.
4. Follow-ups and admin
This is the job that quietly costs you the most. A proposal sent with no chase date is a proposal that goes quiet.
- Motion lists an AI calendar that plans your day, an AI task manager and a meeting notetaker, with a free trial.
- ActivityTracker's tasks let you type a follow-up in one line and it's dated, with a reminder. Tasks from your WhatsApp chats and meeting notes land on the same list.
Whatever you use, the rule is the same: every proposal and every "let's talk next month" gets a date before you close the laptop. Follow-up email templates has messages for each stage.
5. Answering new enquiries
Enquiries arrive when you're with a client or asleep, and a slow reply often means they've asked someone else. An auto-reply that says when you'll answer is the simplest fix. If your enquiries come in on WhatsApp, ActivityTracker's AI agent answers new enquiries on your own number in your words, books the meeting and passes you the chat when a person should take it. The answer leads at night page covers replying at 10pm or mid-meeting.
The tools write and summarise, and someone still has to chase. ActivityTracker keeps every proposal you owe a chase on today's list, and keeps the outreach going while a project fills your week.
What not to put into AI tools
Your clients trust you with things they'd never post online. Before you paste anything into an AI tool, check three things:
- Your contract. Many client agreements and NDAs limit who can see the client's information. A third-party tool may count.
- The tool's data settings. Read what the plan you're on does with what you type or upload, and turn off anything you're not comfortable with.
- How much you really need to paste. You can usually get the same help with the client's name, figures and identifying details taken out.
Never paste passwords, bank details, health information or anything a client marked confidential. When in doubt, describe the problem in general terms instead.
A made-up week
Made-up example: Hana is an independent HR consultant. She bills about 25 hours a week and spends another 12 on proposals, notes, admin and enquiries, which she times for one week.
She picks one job first: meeting notes, which took 4 hours. She turns on an AI notetaker, with each client's permission, and reads and edits each summary for five minutes after the call.
Week two: notes take 1.5 hours instead of 4. One summary gets a client's budget wrong, and she catches it because she reads every summary before filing it.
Week three: she adds a general assistant for proposal first drafts. Proposals drop from 3 hours to about 2, because she still rewrites the problem section herself.
She doesn't add a third tool yet. Her follow-ups still slip, so she tries a simpler habit first: every proposal gets a chase date in her task list before she sends it.
The numbers are invented. Two tools, used properly, saved her more than the six she'd bookmarked.
How to test a tool in a week
- Time the job for one normal week before you start.
- Use the tool for that job only, every time, for a week.
- Check every output before it reaches a client, and count the mistakes you catch.
- Time the job again. If it didn't save at least an hour, or the mistakes cost you more than the time saved, cancel it.
Keep reading
Questions people ask
Sources
- ChatGPT, overview: help with drafting messaging, explaining concepts, searching the web and thinking through decisions; Work creates and edits docs, slide decks and spreadsheets; Deep Research listed, read 28 September 2026
- Claude, product overview: write, edit and create content; research; hand Claude a task, not just a question, read 28 September 2026
- Otter, homepage: transcribes every meeting, send it to meetings or record from your device, AI chat searches across your meetings, read 28 September 2026
- Fathom, homepage: meeting summaries, searchable transcripts and follow-ups, bot or no bot, free plan, read 28 September 2026
- Granola, homepage: AI notepad that uses your computer audio without a meeting bot, works with Zoom, Google Meet and Teams, read 28 September 2026
- Motion, homepage: AI calendar that plans your day, AI task manager, AI meeting notetaker, free trial, read 28 September 2026