The Upwork proposal prompt pack

Marketplace proposals fail in two places. Most fail because the job was never winnable and someone bid anyway. The rest fail because the proposal opened with a paragraph about the bidder rather than about the job.

These prompts attack both. The first two are triage. The rest assume you decided to bid and want the proposal to survive the first eight seconds of a client's attention.

Before running any of this, read the platform's terms on client contact. The prompts are about writing a better proposal on the platform, not about routing around it.

1. Read the post properly

"Here is a job post: [paste]. Answer four questions. What is the actual deliverable, stated as a thing that exists at the end? What has this client already tried, based only on what the post says? What does the post reveal about how they will judge proposals? What is ambiguous enough that a proposal should ask about it rather than assume?"

2. Decide whether to bid

"Job post: [paste]. Client history: [total spend, hire rate, average rate, number of prior hires]. My positioning: [one sentence]. Tell me whether to bid, and give the strongest argument against bidding first. Consider: whether the budget matches the scope, whether the client has hired before, whether the post reads as written by the decision maker, and whether my positioning actually matches the ask or is close enough to be a stretch. End with bid or skip and one sentence of reasoning."

3. Draft the opener

"Write the first three sentences of a proposal for this job: [paste post]. Rules: sentence one is about their problem in their own words. Sentence two says something specific about how I would approach it that a generic bidder could not have written. Sentence three is a question about the ambiguity in the post. No greeting, no name of my company, no mention of my years of experience."

4. Price it

"Job: [paste]. My rate: [number]. Estimated effort: [hours or days]. The client's stated budget: [number or range]. Give me three ways to structure this: fixed fee, hourly, and a smaller first phase. For each, say what risk it puts on me and what risk it puts on them. Recommend one and say why."

5. Handle the specific ask

"The post ends with: [paste the client's instruction, for example start your reply with a specific word, or answer three questions]. Write that part exactly as asked, briefly. Clients use these to filter, so the answer must be obviously deliberate rather than obviously copied."

6. Critique the draft

"Here is my full proposal: [paste]. For each sentence, tell me whether it would still be true if submitted to a different job in the same category. Every sentence that survives that swap is doing no work. Then tell me where a client skimming on a phone would stop reading, and what the first thirty words actually communicate."

7. Write the follow-up question

"The client replied with: [paste]. Write one message that moves this towards a scope and a number. If the reply suggests they are shopping on price alone, say so and give me the version that disqualifies politely rather than the one that competes on price."

8. Review a month of proposals

"Here are [n] proposals I sent last month with their outcomes: [paste, each with hired, interviewed or no reply]. Find the patterns. Which category performed best? Was there a length that correlated with replies? Did the openers of the winners share a structure? Be explicit about where the sample is too small to conclude anything, and treat time as a confounder because recent proposals have had less time to get a reply."

The eighth prompt only works if the data exists. Proposals are recorded per job and per profile, and outcomes come back as interviewed, hired or no reply, which is exactly the input that prompt asks for. upwork_stats returns the summary.

The feed gate handles the first two prompts structurally. Client spend, hire rate, budget and category thresholds filter what reaches a human, so most of the skip decisions are made before anyone reads a post.

The Bid agent that would draft proposals is on the roadmap. When it ships it drafts into the approval queue, because every external send starts in review, and connects stay human-only regardless.