Clients on Upwork, Contra, and similar platforms read dozens of proposals per job. The ones written start-to-finish by ChatGPT are easy to spot: broad enthusiasm, no specifics, the same three "I'm excited to bring my expertise" sentences everyone else sent. The fix isn't a better prompt — it's using AI on the right part of the job. This is a seven-step workflow that keeps the thinking human and uses AI only where it genuinely helps.
The seven-step workflow
| Step | What you do | AI's role |
|---|---|---|
| 1. Extract requirements | List every stated and implied requirement in the job post: deliverable, tools, timeline, budget cues, "must-haves". | Paste the job post, ask it to list explicit vs implied requirements. You verify against the post. |
| 2. Identify the real problem | Name the outcome the client actually wants (more signups, less manual work, a launch date met) — not the task. | Ask it what business problem a client posting this might be trying to solve. Treat as a prompt for your own thinking. |
| 3. Select proof | Pick one past project that maps to their problem. Note the specific result you can describe honestly. | None. This is yours. AI cannot know your work. |
| 4. Draft | Feed AI your notes from steps 1–3 and the prompt below. | Produces a structured first draft from your inputs. |
| 5. Strip generic AI language | Delete every sentence that could appear in anyone's proposal (see the kill-list below). | Ask it to flag its own filler — then cut harder than it suggests. |
| 6. Personalise | Add the one detail only someone who read the post would write. Reference their product, their stack, their phrasing. | None. |
| 7. Human review | Read it aloud. If it doesn't sound like you talking, rewrite that line. | None. |
What AI should and should not do here
- Good uses: restructuring a rambling draft, tightening grammar, condensing three sentences into one, adjusting tone for a formal vs casual client, and giving non-native English speakers polish without flattening their voice. ChatGPT and Claude both have usable free tiers for this.
- Never let AI: invent a result, a client name, a metric, or experience you don't have; decide which past project to cite; write the opening line; or claim familiarity with a tool you haven't used. Those are the parts that win or lose the job, and they are the parts AI gets wrong.
The generic-language kill-list
Delete these on sight in step 5 — they signal "AI wrote this and I didn't read it":
- "I'm excited about the opportunity to…"
- "I'm confident I'm the perfect fit / ideal candidate."
- "With my extensive experience in…"
- "I have a keen eye for detail and a passion for…"
- "I'd love to bring my expertise to your project."
- Any sentence that would be equally true for a different job in a different field.
A worked transformation
Illustrative only — an invented job post for a "Shopify email flow" gig.
Generic AI opening: "Hello! I'm excited about the opportunity to help with your email marketing project. I have extensive experience in email marketing and a proven track record of delivering results. I'm confident I'd be a great fit for your needs."
After steps 1–7: "You mentioned your abandoned-cart flow is a single email and your welcome series doesn't segment by first purchase. I rebuilt exactly that setup for a homeware store last year — three-email cart flow plus a branch for first-time vs returning buyers. I can send you that flow's structure so you can judge the approach before we talk. One question: are you on Shopify Email or Klaviyo? It changes how the branching is built."
The second version names their situation, offers proof they can inspect, and asks a question that shows domain knowledge. No AI wrote those specifics because no AI had them.
A reusable draft prompt
"Here is a job post: [paste]. Here are my notes: the client's real problem is [X]; the relevant project I'll cite is [Y] with the honest result [Z]; the tool question I want to ask is [Q]. Draft a proposal under 150 words with this structure: (1) one sentence naming their specific situation, (2) two sentences on the comparable project and result, (3) an offer to share a sample or outline, (4) one specific question. Plain sentences, no bullet points, no phrases like 'excited about the opportunity' or 'perfect fit'. Do not invent any detail I did not give you."
Before you submit — checklist
- First sentence references something specific to this post, not a greeting.
- Every claim is one I can back up if asked on a call.
- No sentence from the kill-list survived.
- One concrete piece of proof the client can look at or ask about.
- Exactly one clear question or next step at the end.
- Length matches the job: short for a quick gig, fuller for a complex build.
- Read aloud — it sounds like me, not a press release.
- Any skill gap is acknowledged honestly, not papered over.
When to skip AI entirely
For a client you already have a relationship with, a tiny gig where a two-line message is the right call, or a niche where your natural writing voice is the selling point (copywriting, brand strategy), drafting through AI adds a round trip and a flattening pass you then have to undo. Write those yourself.
None of this guarantees more hires — proposal response depends on fit, timing, budget, and how many others applied. What it does is make sure the proposal that gets read is specific enough to be judged on your merits. Pair it with using AI for client research without cutting corners and pricing your services as a beginner.