Most people who message an AI consultant describe a chore, not a technical problem: "I copy every new lead from my web form into a spreadsheet and email them by hand. Can AI do that?" As an AI consultant, you can charge for the answer (which tool, what to automate first) as a Paid Question that they pay for upfront and you answer in writing. You can also charge for the build as Custom Work: they send the steps, the apps they pay for and a few rows of fake sample data, and you return an importable n8n JSON, a Make blueprint, a Zapier share link or a custom GPT kit.
No discovery call and no proposal stand between the person who watched your automation video and the workflow running in their business.
Who pays an AI consultant for one answer or one build?
Mostly people who will never write a line of code. A realtor who wants every showing request summarized and answered with a drafted email. A bookkeeper copying invoice totals out of PDFs by hand. A coach who wants a GPT that answers client questions in her voice. A five-person agency that lost a weekend to a Make scenario that quietly stopped running.
They find you through a workflow breakdown on TikTok, an n8n build on YouTube, or a LinkedIn post about the tools you use. And they describe a workflow, not a technical problem: "When someone books a call, I copy their details into my CRM, send a welcome email and add them to a spreadsheet. Can AI do that?"
What they want back is something they can use on Monday: a clear recommendation, a working prompt, a configured GPT, or a workflow they import and switch on. That is a different job from reviewing someone's code. If your audience is mostly developers sending repos and pull requests, the developers page is a better fit.
Today most AI consultants handle these requests one of three ways: answer in comments and DMs for free, push everyone toward a booked hourly call, or quote each build by hand and send an invoice. Free answers eat your evenings. An hour-long call is too much for someone with one question. And hand-quoting turns every inquiry into a negotiation before any work starts.
What should an AI consultant sell first?
Start with two to four offers. For each one below you get what it includes, what the buyer sends and a delivery time that suits the work. Rename them in your own words.
1. Paid Question: "Which tool, and what should I automate first?"
- Includes: one private written answer, with up to two follow-ups. Good for picking a tool (ChatGPT, Claude or Gemini for a specific task; Zapier, Make or n8n for their budget and volume), for "what should I automate first," and for "is this even possible with AI?"
- Buyer sends: the task in plain steps, the apps they already pay for, how often it happens each week, and what they've tried. A Paid Question is text only, so ask them to paste any error message as text and describe screens in words.
- Delivery: 24β48 hours is common. You pick a reply time from 1β24 hours or 1β5 days.
- A useful answer shape: the recommendation, why it fits their volume and budget, the first three steps, a rough idea of the monthly tool costs they'd take on, and the one trap to avoid (for example, a Zap that uses a task on every spreadsheet row).
2. Custom Work: Prompt or custom GPT setup
- Includes: a finished system prompt or custom GPT configuration for one job: instructions, conversation starters, a structured knowledge file, and a short test log showing three sample inputs and the outputs they produced.
- Buyer sends: who will use it, what goes in, what good output looks like, two or three real examples with names removed, and notes on tone. They can attach up to 5 files, such as an SOP, a style guide or a past email thread.
- Delivery: 2β3 days.
3. Custom Work: One automation built to a brief
- Includes: one workflow with a defined trigger, a set number of steps and named apps. You deliver it as an importable n8n JSON file, a Make blueprint or a Zapier share link, plus a setup note covering which accounts to connect, which fields to map and how to test it.
- Buyer sends: the trigger, each step in order, the apps and plan levels they're on, sample data (a CSV with fake rows is ideal), and what should happen when a step fails.
- Delivery: 3β5 days.
4. Custom Work: Broken workflow diagnosed
- Includes: a written diagnosis of why it fails and the exact fix, or a corrected export.
- Buyer sends: the exported workflow, the error text from the run or execution history, and screenshots of the failing step with any keys blurred.
- Delivery: 1β2 days.
How each offer works on the follower's side is covered on Paid Questions and Custom Work.
What should clients send you, and what should they never send?
Put this line in the "what I need from you" field of every listing, in plain words: never send passwords or API keys in a request. Most non-technical clients don't know what is safe to share. If you tell them before they pay, you won't get an OpenAI key pasted into a brief.
What you actually need can be shared safely once it's checked:
- n8n: a workflow JSON export. It points to credentials by name and doesn't include the secret values.
- Make: an exported blueprint. Connections stay in their account.
- Before exporting: delete any key or token typed straight into an HTTP node or module (headers, URLs, body fields), and clear pinned data so no real customer records go along with the file.
- Zapier: screenshots of each step and the error from the Zap's run history, with anything sensitive blurred.
- Sample data: a few rows of made-up names and emails, not their real customer list.
When you deliver, the client connects their own accounts to your workflow. You never log in as them. That keeps their data with them, and it keeps you out of the business of storing anyone's credentials.
How do I scope an automation build so it fits in one order?
Five days is the longest delivery time you can set, and each order is one delivery: no milestones and no built-in revision rounds. So write the scope tightly enough that you can always finish within the window. For a typical workflow build:
- One trigger (a form submission, a new row, a new email, a webhook)
- Up to a set number of steps or modules, such as eight
- Up to three connected apps
- One AI step (a summary, a classification, a drafted reply)
- Basic error handling: a notification when a run fails
Then list what's excluded: hosting, the client's own subscription and API usage costs, ongoing maintenance, and scraping sites whose terms forbid it.
When a client wants a whole system, like a CRM rebuild with a dozen workflows, don't squeeze it into one order. Sell the first workflow on its own. Or sell an "automation map" as Custom Work: a written plan of what to build, in what order, with which tools. Each later workflow then becomes its own order at your listed price.
Write your correction and refund policy before you publish. Followers see it before they pay. On Custom Work, once you've delivered, the client can open a support conversation on the same order ("the import fails on my n8n version") and you reply there.
Say this in every listing too: no guaranteed results. You build what was scoped. You aren't promising hours saved, extra revenue, or that a model's output will be correct every time. The same prompt can behave differently after a model update.
How much should I charge for AI tool questions and automation builds?
You set the price. Common starting points among independent AI consultants:
- Paid Question on tool choice or "what to automate first": often $30β$75
- Prompt fix: often $40β$90
- Custom GPT configured with instructions and knowledge files: often $120β$300
- Single Zap built to a brief: often $75β$200
- n8n or Make workflow built to a brief: often $150β$450, depending on steps and apps
- One scoped AI agent: often $250β$600
Price on what the build replaces, not on how fast you work. A workflow that takes you ninety minutes can still end three hours of copy-pasting every week for the client. Raise prices once orders come in faster than you can deliver them.
No monthly fee, you only pay a small platform fee when you get paid. 8% platform fee when a follower pays for work, 5% platform fee on tips and goals. Card processing is separate.
The client pays their own tool subscriptions and API usage. Say so in the listing so nobody expects you to cover their monthly OpenAI or Zapier bill. The full breakdown is on Pricing.
Building AI agents: how do you sell an agent in one order?
Agent buyers are usually small businesses that want something to handle the inbox, qualify leads or look up order details. What sets this apart is tools and memory. An agent decides which tool to call and what to remember, and that's where most of the testing time goes.
Keep one order to one agent with one job: up to three tools (search a knowledge base, look up a row in a sheet, draft a reply) and one memory approach (a short conversation window, or stored notes per customer).
- Custom AI agent workflow: the system prompt, tool descriptions, memory setup and the build itself (for example, an n8n workflow using the AI Agent node), plus a test transcript of five cases. Include at least one case where the agent should hand off to a human.
- Agent prompt and tool setup: for clients who already built an agent that picks the wrong tool, loops, or forgets context. They send the prompt, the tool definitions and a transcript of a failed run.
No guaranteed results. Agents make mistakes, so build a human approval step before anything sends an email to a customer or touches money.
Selling n8n workflows: what changes when the client self-hosts?
Many n8n buyers come from your YouTube builds and are a bit more technical. Some run n8n Cloud and some self-host on a VPS, and that difference shapes the order. Ask for their n8n version and setup in the brief. A workflow built on a newer version, or one using a community node, can fail to import on an older install.
- n8n workflow built to a brief: your video template adapted to their apps and fields, delivered as a JSON file they import, with notes on which credentials to create.
- Self-hosting setup notes: a written step-by-step for their server. They tell you the provider, the operating system and whether they use Docker. You cover things like the webhook URL behind a reverse proxy, timezone settings and keeping data when updating. You write the steps; they run them. You never need server access.
Never send passwords or API keys in a request applies twice here, because server credentials are as sensitive as app keys.
Zapier for solopreneurs: what do one-person businesses ask for?
Zapier buyers are coaches, shop owners and freelancers who want a few hours a week back without learning a new tool. Their worries are plan limits and task usage: multi-step Zaps, Paths and premium apps depend on the plan they pay for.
- Zap built to a written brief: built inside their plan's limits, with Filters so it doesn't burn tasks on rows it should skip, delivered as a share link they copy into their own account, plus a note on reconnecting each app.
- Broken Zap diagnosed: they send the error text from the Zap history and a screenshot of the failing step. Most fixes are a field mapping, a Formatter step or a filter condition, so this works well at a lower price with a 1β2 day window.
No guaranteed results.
Make for agencies: what is different about scenario work?
Make buyers are often agencies running client reporting, lead routing or content pipelines. Their scenarios use routers, iterators and aggregators, and operations count matters because it drives their bill.
- Make scenario built to a brief: delivered as a blueprint file with a short note per module explaining what it does, so their team can maintain it.
- Scenario debugging: they send the blueprint and the error from the execution history (a bundle validation error, an iterator creating duplicates, a filter that never passes) and get the fix in writing or as a corrected blueprint.
Agencies sometimes pass your work to their own clients. Followers see the public name you choose on your page and at checkout, so you can sell under a studio name rather than your legal name. No guaranteed results.
Custom GPTs for coaches and educators: what do you actually hand over?
Coaches, course creators and teachers want a GPT that sounds like them, answers their students' recurring questions, or quizzes learners on their material. The catch: a GPT built in your account stays yours, so the real deliverable is a build kit they paste into their own GPT editor.
- Custom GPT configured: written instructions that fit the instructions field, conversation starters, a recommended knowledge file list, and test prompts with expected answers. You can share a link to your own test version so they can try it first.
- Knowledge file and instructions setup: turning their messy course notes, FAQ and transcripts into clean, well-labeled files the GPT can actually use, plus instructions that tell it when to say "I don't know."
For educators, design the GPT to guide and question rather than hand out finished answers. No guaranteed results: GPTs can still be confidently wrong, so tell clients to test with real questions before sharing it with students.
Prompt engineers: how do you sell prompts, not just tips?
Prompt buyers are marketers, recruiters and operations leads who use ChatGPT or Claude every day and get uneven results.
- Custom prompt pack for a workflow: for example, eight prompts for a recruiter's week (job description, outreach message, screening summary, rejection note), each with clearly marked variables in square brackets and an example output. Delivered as a document.
- Prompt that's misbehaving fixed: they paste the prompt, name the model, and show two bad outputs and one example of what good looks like. A short prompt can be a Paid Question because it fits in text. A long one with sample files works better as Custom Work.
No guaranteed results: a fixed prompt can drift after a model update, so include a one-line test they can rerun to check it still works.
What about tips, goals and brand inquiries?
Tips suit the follower who copied your free n8n template and got a day back. A one-time thank-you, nothing for you to deliver.
Goals suit a public project: "Fund the free automation template library" or "New mic and capture setup for build videos." Followers make one-time contributions, every contribution is yours whether or not the goal is reached, and there's no deadline unless you set one.
Brand Collaboration Inquiries matter if AI software companies keep pitching you sponsored reviews. A Work with Me form on your page collects the brand's budget, timeline and deliverables into a separate brand inbox, so pitches don't get buried among client requests. If reviewing AI tools is most of what you post, the tech reviewers page covers that side in more depth.
How do I start taking paid AI questions this week?
- Create your page. There's no follower minimum and nothing to apply for. Creators in 41 countries can get paid.
- Turn on Paid Questions with a reply window you can always meet, and a card that says what to include: the task in steps, the apps they use, and no keys or passwords.
- Publish one Custom Work listing for the build your audience asks for most, usually the workflow from your most-watched video. Fill in what's included, what you need, what's excluded and the delivery time.
- Write your correction and refund policy so it's settled before anyone pays.
- Put the link where the requests arrive: your bio, pinned under your best build video, and as your reply to every "can you build this for me?" comment.
Followers pay as guests, with no account to create. The full flow is on How it works. When you're ready, start your Brillay page.
Frequently asked questions
Do I need a discovery call before building an automation?
Not if the listing does that job. Publish the scope, what you need and what's excluded, and the brief collected with the order answers what a discovery call would. If a request is too vague to build, a Paid Question works well as the first step: the client describes the workflow and you reply with what to build and which listing fits.
Can I deliver an n8n workflow or a Make blueprint as a file?
Yes. Custom Work delivers a written note, up to 5 files and links in one package, so you can attach the JSON or blueprint with a setup guide. For Zapier, include the share link. A screen recording up to 20 minutes long can go in the same delivery if a walkthrough helps.
What if the client says the workflow doesn't work after delivery?
They can open a support conversation on the Custom Work order once you've delivered, and you reply there with files if needed. Whether a fix is included is decided by the correction policy you wrote, which they saw before paying. The usual causes are app connections that were never authorized or a different n8n version, so cover both in your setup note.
How much does Brillay cost an AI consultant?
There's no subscription. You only pay the platform fee described in the pricing section above, on orders that actually get paid, and card processing is separate.
Can I sell a full AI agent or a complete automation system?
Not as one order. Delivery is capped at five days, with no milestones, so sell a clearly scoped piece: one agent with one job, one workflow, or a written automation map. Each next piece is a new order at your listed price.
Do my clients need an account to pay?
No. They pay by card as guests on your page, under the public name you choose. Nothing to join and no membership.
Explore more