AI for Consultants: How a One-Person Practice Runs on Claude Without Losing the Judgment Clients Pay For (2026)

TL;DR: Most "AI for Consultants" Advice Is Written by People Who Sell Courses to Consultants

I'm a consultant. This is how my practice actually runs on AI, not how a productivity influencer thinks it should. Every day, a model does three of the four jobs in this business: it sells (proposal drafts, follow-ups, qualifying), it delivers (research, first drafts, notes into action lists), and it runs the shop (invoices, scheduling, the newsletter). It is banned from the fourth job, which is deciding what the client should do. That division is the entire method. Get it right and a one-person practice does the work of three. Get it wrong and you become the consultant whose deliverable a client could have generated themselves, and they'll figure that out faster than you think.

The three jobs AI does in my practice every day

1. Selling: the call becomes the proposal

The most expensive hour in consulting is the one between a good discovery call and a proposal that doesn't get written. Mine now goes like this: call notes into a project that already holds my offers, my prices, and my last ten proposals; ask for a draft that recommends one option and explains why the other two are worse for this client; edit for forty minutes; present it on a call instead of emailing it. What used to take a weekend takes an afternoon, and the proposal is better because the model has read every winning one I've ever sent. Follow-ups work the same way: the model drafts, I send. Inbound qualification, same: the inquiry plus my ICP definition produces a "yes, no, or ask these two questions" in seconds.

2. Delivering: research and first drafts

An assessment for a client used to start with a week of reading. Now it starts with their documents, their metrics export, and their call transcripts dropped into a project and a set of questions I've refined over a hundred engagements: what's inconsistent, what's missing, what would a skeptical operator ask first. The synthesis takes an hour. Then every deliverable — the roadmap, the written recommendation, the handoff doc — starts as a first draft in my voice, from my templates, and I rewrite the parts that matter. The parts that matter are always the same parts: the recommendation and the reasons. The model never gets those. It gets everything around them.

3. Running the practice: the boring third

Invoicing questions, scheduling replies, the newsletter draft, the bookkeeping categorization, the "what did I promise this client in March" search. I documented the full solopreneur systems stack elsewhere; the AI layer is the draft-and-approve step that sits on top of all of it. Nothing sends without me. Everything starts without me.

The one job it's banned from

The recommendation. Not because the model can't produce one — it will produce five, confidently. Because the recommendation is the product. When a client pays me, they are paying for a decision made by someone who has watched fifty of these go wrong, who will say "don't build that" when the deck says build it, and who can be blamed. A model can't be blamed. The moment your recommendation is whatever the assistant said, you've become a very expensive prompt, and the client can buy the prompt for $20.

The practical version: I use the model to generate the options and to argue against my own pick. Then I choose, in writing, in my own words, with my name on it. That's the sentence in every engagement letter I sign and it's the one clients remember.

The context is the whole game

The difference between a consultant who gets hours back and one who gets generic mush is not the tool. It's what's loaded. Mine, in one shared project: who I serve and who I refuse, every offer with its price and scope, the last twenty proposals, twenty examples of my own writing, the questions I ask on every discovery call, and a running file of what I actually believe about the work. I wrote up the thinking behind that file as the business brain. An assistant with nothing loaded is a stranger. An assistant with that loaded is an associate who has read everything you've ever done. Building it is an afternoon. Keeping it current is the discipline most consultants skip, and it's why they conclude AI "isn't that useful for our kind of work."

What it costs (and what to buy)

As of September 2026, the stack for a solo practice is one line: Claude Pro at $20 a month, $17 on annual billing, which includes projects and Cowork. If you have an assistant, a subcontractor, or a partner, Claude Team is $20 per seat billed annually ($25 monthly) and is listed for teams of two to 150, and it's the plan you want the moment client material is involved, because the no-training-by-default data terms live there. I run Claude rather than ChatGPT because projects, skills, and connectors let me keep the whole practice's context in one place and reuse it across every engagement; the Claude vs ChatGPT for small business comparison has the honest tradeoffs. Add a call transcriber once you're doing more than a handful of calls a week. That's it. If a vendor is pitching you a "consultant AI platform" at $200 a month, ask which of those two lines it replaces.

Confidentiality: three rules that cover it

1. Business plan, not personal account. The no-training default on Team and Enterprise is the difference between a deliberate setup and a hope.
2. A never-paste list, written down: anything under an NDA that names permitted systems, credentials, anyone's health or financial identifiers, anything you'd hate to see produced from a vendor's logs.
3. One sentence in your engagement terms: what tools you use, that client material stays on a business account with no-training terms, and that every recommendation is yours. Clients assume you use AI. What they want to know is that you handle their information on purpose.

The pricing mistake everyone makes the week they get faster

Here's the trap. AI cuts your proposal time by two-thirds, your research time in half, and your deliverable drafting to a fraction. If you bill hourly, your revenue just fell. If you bill fixed, your margin just doubled, and the temptation is to pass the savings on to "stay competitive." Don't. The client was never buying the hours. They were buying the decision and the person who stands behind it, and that got more valuable, not less, because now you can make it with more evidence, faster. Fixed prices, raised when the outcomes justify it. I wrote the longer argument in how to raise prices without losing customers, and the comparison of how firms and solo consultants compete on this is the same story from the buyer's side.

What good looks like, so you can tell if you're there

• A discovery call becomes a presented proposal within 48 hours, every time.
• No deliverable starts from a blank page, and no deliverable leaves without your rewrite of the recommendation section.
• Your context project is updated after every engagement: what you learned, what you'd refuse next time.
• Your engagement terms have the AI sentence, and you've never had to explain it in a panic.
• Your price went up this year, and your hours went down.

If two of those are false, the fix is a weekend of loading context and one honest conversation with yourself about what your product actually is.

The micro-action for today

Take your last three proposals, your offer list with prices, and ten emails you're proud of. Put them in one project. Then paste the notes from your most recent discovery call and ask for a proposal draft that recommends one option and argues against the other two. Read what comes back. The parts that are wrong are your judgment showing; the parts that are right just saved you Saturday. If you want the install done properly — the context loaded, the workflows built, the never-paste list written — that's what I do for owner-led businesses under the Claude for Small Business work, and the self-serve Playbook is $397 if you'd rather build it yourself. Consultants who want to talk through their own practice first can book the 30-minute strategy call; it's free and there's no pitch in it. And if you're on the other side of this page — an owner deciding whether to hire one of us — the AI consultant guide is the honest buyer's version.