How to Write an Upwork Job Post for an AI Automation Project

A practical field guide from Automation Ace.

Short answer: a good Upwork job post for an AI automation project names the tools and outcome in the title, describes the problem and current process, states the result you want in measurable terms, lists the apps (with plan levels and who owns the accounts), gives volume and known exceptions, defines what done looks like, proposes a contract shape, and asks two to four screening questions that a generic proposal cannot answer.

The better your post, the better the proposals. As an automation freelancer on Upwork (Troy Tessalone on Upwork), Troy Tessalone reads a lot of job posts; the ones below get thoughtful, specific proposals, and the vague ones get templates.

Short answer

• Title: tools plus outcome.
• Problem: what happens today and why it hurts.
• Outcome: what should happen automatically, in measurable terms.
• Apps, access, volume, and exceptions.
• Definition of done, including testing and documentation.
• Contract shape and screening questions.

Anatomy of a strong job post

Copy-and-edit template

TITLE
Zapier + OpenAI expert: score and route inbound leads into HubSpot

ABOUT US (2-3 sentences)
We are a 12-person B2B services firm. Leads come from our website
form (Webflow), Calendly, and LinkedIn Lead Gen Forms.

THE PROBLEM
Leads are copied into HubSpot by hand, often hours later. Reps waste
time on poor-fit leads, and some leads are entered twice.

THE OUTCOME WE WANT
- Every lead reaches HubSpot within 1 minute, deduplicated by email
- An AI step scores fit (1-100) with a short reason, saved to HubSpot
- Qualified leads get a reply draft; the owner gets a Slack alert
- Errors alert us instead of failing silently

APPS AND ACCESS
Webflow, Calendly, LinkedIn Lead Gen Forms, HubSpot (Sales Hub
Starter), Slack, Zapier (our account, Professional plan), OpenAI API.

VOLUME
About 300 leads per month, peaks of 40 per day after webinars.

KNOWN EXCEPTIONS
Existing customers filling the form, partner referrals, spam.

WHAT DONE LOOKS LIKE
Built in our accounts, tested with 25 real leads, documented
(what each Zap does, owners, how to fix common errors), and running
two weeks without manual fixes.

HOW WE WANT TO WORK
Fixed price with milestones: (1) design doc, (2) build and test,
(3) handoff and two weeks of support. Open to a short paid discovery.

TO APPLY, PLEASE ANSWER
1. How would you prevent duplicate contacts?
2. What would you do when the AI step fails or returns bad output?
3. Share one similar project and what you would do differently now.
4. What will this cost us to run each month in Zapier and AI usage?

Write a title that attracts specialists

Specialists search for their tools. Compare:

• Weak: “Automation expert needed”
• Better: “Zapier expert to connect Typeform, Airtable, and Slack”
• Best: “Make.com + OpenAI: extract invoice data from email PDFs into QuickBooks”

If you do not know which tool to use, say so: “Automation expert (Zapier, Make, or n8n) to sync Shopify orders to Airtable.” A good freelancer will recommend one and explain why; see Zapier vs Make vs n8n.

Describe the problem and the outcome, not the tool

“Build me a Zap” tells a freelancer what you think the solution is. “We reply to leads hours late and miss some entirely” tells them what to solve, and invites a better design. Describe the current process in a few numbered steps, who does each, and how long it takes. A rough map is enough; for a method, see current state vs future state process mapping.

Scope, volume, and exceptions

• Volume (records per day or month) changes the right tool, the plan you need, and the running cost; see Zapier tasks vs Make operations pricing.
• Exceptions are where most of the work is: duplicates, blanks, special customers, approvals. List the ones you know; see the exception inventory.
• Out of scope: say what is not included, such as rebuilding your CRM or migrating historical data.

What to add for AI and agent projects

• The decision or text the AI handles: “classify support emails into 6 categories,” “draft replies,” “extract 12 fields from PDFs.”
• Examples: attach 10 to 20 anonymized real inputs and the outputs you would want.
• Review level: should AI act on its own, or suggest for a person to approve? See Human in the Loop.
• Data sensitivity: personal, health, or financial data changes which AI services and settings are acceptable.
• Model preference or account: whether you already have an OpenAI, Anthropic, or other account to use.

If the project is an agent rather than a fixed workflow, describe the goal, the tools it may use, and when it must ask a person; see AI agents vs traditional automation.

Budget and contract shape

Give a realistic range or say you want a fixed-price proposal after a short discussion. Proposing milestones (design, build and test, handoff and support) attracts freelancers who work in stages and makes acceptance clear. If scope is fuzzy, offer a small paid discovery; see choosing the right contract type.

Screening questions that filter proposals

Good screening questions cannot be answered with a template:

• “How would you prevent duplicate records in this workflow?”
• “What would you do when the AI step fails or returns bad output?”
• “Describe one similar project and what you would do differently now.”
• “What would this cost to run monthly?”
• “What would you not automate here?”

After you post

1. Invite five to ten specialists with relevant work rather than waiting for proposals alone.
2. Reply to good questions quickly; the best freelancers are evaluating you too.
3. Shortlist using the scorecard in what to look for when hiring an AI automation specialist.
4. Interview two or three, then start with discovery or milestone one.

Common job post mistakes

• No apps listed, or no plan levels (some features need paid plans).
• “Simple task” for work with many exceptions.
• Asking for free samples or full designs before hiring.
• Leaving out who owns accounts and where it will be built.
• No definition of done, so the project never quite ends.
• Pasting passwords or sensitive data into the post.

Need ideas for what to post? See 10 AI automation projects to outsource. Or invite Troy directly through his Upwork profile.

Frequently asked questions

How do I write an Upwork job post for an automation project?

Name the tools and outcome in the title, describe the problem and current process, state the result you want in measurable terms, list the apps with plan levels and account ownership, give volume and known exceptions, define what done looks like, propose a contract shape, and add two to four screening questions.

What should the title of an automation job post say?

Name the tools and the outcome, for example "Make.com + OpenAI: extract invoice data from email PDFs into QuickBooks." Specialists search for their tools, so a specific title attracts better proposals than "automation expert needed."

What should I include in a job post for an AI project?

Describe the decision or text the AI handles, attach 10 to 20 anonymized example inputs and desired outputs, say whether AI acts alone or suggests for review, note any sensitive data, and say whether you already have an AI provider account to use.

What screening questions should I ask automation freelancers?

Ask how they would prevent duplicate records, what they would do when an AI step or API fails, what they would do differently on a similar past project, what the workflow will cost to run monthly, and what they would not automate.

Should I include a budget in my Upwork job post?

Yes, a realistic range or a request for a fixed-price proposal after a short discussion helps freelancers scope their answer. Proposing milestones such as design, build and test, and handoff makes acceptance clear.

Should I ask for a free sample before hiring an automation freelancer?

No. Ask for relevant past work and good answers to your screening questions instead. If scope is unclear, pay for a short discovery that produces a design and estimate you own.