# How to Use Claude for RFPs and Bid Management, Stage by Stage

How to use Claude for RFPs, stage by stage: which Claude surface and skill handles each bid job, what a person still checks, and where to start.

<KeyTakeaways
  items={[
    'Run Claude through a bid in a fixed order: qualify, shred the RFP, draft, then check for contradictions, with a person making the call at each stage.',
    'Save each job you repeat on every bid as a skill, so it runs the same way on every tender.',
    'Give Claude every file the buyer sent as its own file, including embedded spreadsheets and every Q&A, because Claude can only check what it is given.',
    'Put a gap rule in every drafting prompt: name the file behind each answer, or write GAP where the files say nothing.',
    "AutoRFP.ai's connector gives a bid team's Claude setup one library of approved answers, each returned with its source through read scopes, while full scopes let Claude write to projects and content as the signed-in user, with approvals kept in AutoRFP.ai's workflow.",
  ]}
/>

Using Claude for RFPs means running each stage of a bid as a repeatable method in Claude, from the go/no-go before anyone writes to the contradiction check before submission, with the bid team making every decision.

A bid team sets it up as one Claude project per live tender, a skill for each repeated job, a connector to the approved answer library, and Cowork or Claude Code for work across a folder of files. Claude does the first pass and flags what the files cannot support; people decide, fill the gaps and sign off.

I built most of the skills this guide uses, and wrote it for bid and proposal managers who already work in Claude. Buyers writing or scoring RFPs with Claude are doing a different job. Plan names and features are as of November 2026.

Below: a map of bid jobs to Claude surfaces, the seven stages, where Claude stops, and the team setup.

## How Do Bid Managers Use Claude for RFPs?

Bid managers who get real value from Claude use it for the first pass of every stage, in a fixed order, and keep each decision with a person. Three habits make that work:

- **Decide before you write.** In my walkthrough of a mock RFP in Claude (YouTube, October 7, 2026), the go/no-go came first: "I have not written a word. I've maybe read the RFP."
- **Save the method as a skill.** A skill runs the same method the same way on every bid. As I put it in that video, "that skill is my complicated prompt that is saved in my organization."
- **Ask where each answer came from.** Even with good skills, "occasionally you will get back, 'Yeah, you're absolutely right. I made that up.'" Hence the gap rule: name the file behind each answer, or write GAP.

Claude suits this method-led work. It reads a long tender pack, follows a saved method the same way each time, and produces working files such as a compliance matrix, a scorecard or a punch list.

<BlogCta id="Proposal Win Rate Report" />

## Match Each Bid Job to a Claude Surface and Skill

Each bid job has a natural home in Claude: a project chat for decisions and summaries, Cowork or Claude Code for work across a folder of files, and connectors wherever the job needs the CRM or the approved answer library.

| Bid job                                   | Where it runs                                                      | Skill from the library                                                                              | What comes back                                                                              | Who decides                                 |
| ----------------------------------------- | ------------------------------------------------------------------ | --------------------------------------------------------------------------------------------------- | -------------------------------------------------------------------------------------------- | ------------------------------------------- |
| 1. Decide whether to bid                  | A chat in the tender's project                                     | AI Go/No-Go Agent Skill                                                                             | A weighted recommendation, a source behind each finding, and "REQUIRES USER INPUT" questions | The bid lead and leadership                 |
| 2. Shred the RFP                          | Cowork or Claude Code, on the tender folder                        | Compliance Matrix Builder                                                                           | A compliance matrix, submission checklist, evaluation criteria and key dates                 | The bid manager, who assigns owners         |
| 3. Research the buyer and the competition | A chat with CRM, email and meeting connectors, plus web search     | Internal Research and External Research, merged by Completed Insight Brief; Competitor Intelligence | A sourced insight brief with "Missing" for each gap; a battlecard with dated sources         | The seller and the bid manager              |
| 4. Assign owners and prepare SMEs         | A chat with the answer library and file connectors                 | Assign Owners; SME Interview Prep                                                                   | Proposed owners and reviewers; only the questions the records cannot answer                  | The bid manager                             |
| 5. Draft the answers                      | The tender's project, with the library reached through a connector | None: project instructions with a gap rule                                                          | Drafts with a source on each, and GAP where the files say nothing                            | The SMEs and reviewers                      |
| 6. Write the executive summary            | A chat in the tender's project                                     | RFP Executive Summary Generator                                                                     | A six-block summary and flags for SMEs to check                                              | The bid manager                             |
| 7. Check before submission                | Cowork or a project chat, on the near-final response               | RFP Contradiction Checker                                                                           | A punch list quoting both sides of each conflict, by severity                                | The SMEs resolve; the bid manager signs off |

- **Skills, projects or prompts:** see [what skills, projects and prompts are, and when to use each](/blog/claude-skills-vs-projects-for-rfps).
- **What Claude can reach:** stages 3 to 5 depend on [Claude connectors](/blog/claude-connectors-for-rfps), including the one to your approved answer library.
- **Each tender as a folder:** start with [getting started with Claude Code as a bid manager](/blog/getting-started-with-claude-code-for-rfps).

The surfaces are my recommendation for bid work, with Anthropic's features as of November 2026.

## Run the Seven Bid Stages in Claude, in Order

A Claude bid runs in seven stages, in the order a good bid team already follows: Claude does the first pass, and a named person makes the call. The stages sit inside the wider process in the full RFP response guide.

### 1. Qualify the Bid Before Anyone Writes

Decide whether the bid is worth writing before anyone drafts.

- **Skill and surface:** the AI Go/No-Go Agent Skill, in a chat inside the tender's project.
- **What Claude does:** scores seven weighted categories against the tender pack and your company profile, cites the section behind each finding, and returns Strong Go, Conditional Go, Executive Review or No-Go.
- **What to check:** answer every "REQUIRES USER INPUT" item before anyone trusts the score.
- **Watch for:** deal-breakers outside the main RFP, such as a mandatory standard named only in the buyer's Q&A.

For how the skill is built and shared, see [building the go/no-go skill yourself, file by file](/blog/how-to-create-a-claude-skill-for-rfps).

### 2. Shred the RFP Into a Compliance Matrix

Turn every file in the pack into one list of requirements, so nothing mandatory goes unanswered.

- **Skill and surface:** the Compliance Matrix Builder, in Cowork or Claude Code on the tender folder.
- **What Claude does:** quotes each requirement exactly, tags it mandatory, desirable or informational, adds the criteria, a submission checklist and key dates, and lists what it could not read.
- **What to check:** that "could not read" list first, then an owner on every row and a clarification question for anything ambiguous.
- **Watch for:** a text-only read. It can miss a security workbook embedded in the RFP's Word file, so add that workbook, and each Q&A, as its own file.

The full method is in [shredding an RFP with Claude](/blog/rfp-shredding-with-claude).

### 3. Research the Buyer and the Competition

Learn the buyer from your own records first, then from public sources and the competition.

- **Skills and surface:** Internal Research and External Research, merged by the Completed Insight Brief skill, plus the Competitor Intelligence skill, with CRM, email and meeting connectors and web search.
- **What Claude does:** writes an insight brief with a source on every line and "Missing" where the records are silent, and a battlecard that flags sources older than 12 months.
- **What to check:** the seller and the bid manager close each Missing gap.
- **Watch for:** filling the brief from the RFP or the buyer's website, which Internal Research rules out. Both Internal Research and Completed Insight Brief put government tender processes out of scope, so use them on private-sector deals.

I call competitor positioning "one of my favourite skills to run on RFPs we get" (LinkedIn, November 2, 2026).

### 4. Assign Owners and Prepare the SME Interviews

Give every requirement an owner, and spend expert time only where your records fall short.

- **Skills and surface:** the Assign Owners skill and the SME Interview Prep skill, with the answer library and file connectors.
- **What Claude does:** proposes an editor, an accuracy reviewer and a narrative reviewer for each requirement, and assigns nothing without approval. It then rates each question Covered, Refresh, Thin or Unknown and writes only the ones the records cannot answer.
- **What to check:** the bid manager approves the plan first.
- **Watch for:** interview questions the library already answers.

### 5. Draft From Approved Content and Flag the Gaps

Draft from approved content, and keep every gap visible instead of filled.

- **Skill and surface:** no dedicated skill. Use the tender's project instructions, with the library reached through a connector.
- **What Claude does:** drafts each answer with the file behind it, and writes GAP where the files say nothing.
- **What to check:** approved wording stays word for word, SMEs fill each GAP, and reviewers approve every answer.
- **Watch for:** a prompt with no gap rule, which can let a capability your files never mention through as met.

To see one draft built end to end, follow [a step-by-step walkthrough of one mock RFP in a Claude project](/blog/how-to-respond-to-an-rfp-with-claude).

### 6. Write the Executive Summary From the Finished Response

Write the summary last, so it claims only what the answers support.

- **Skill and surface:** the RFP Executive Summary Generator, in a chat inside the tender's project.
- **What Claude does:** reads the completed response as its list of claims, mirrors the RFP's criteria, writes six blocks of prose and ends with "Flags for SME validation".
- **What to check:** SMEs check each flag, and the bid manager owns the page.
- **Watch for:** running it before the response is finished.

A guide to executive summary prompts in Claude is coming.

{/* LINK (wave 2): when post 9 is live, replace the sentence above with a down-link, anchor "executive summary prompts for Claude" (slug executive-summary-ai-prompts-for-rfps). */}

### 7. Check for Contradictions Before Submission

Catch the answers that disagree with each other before the buyer does.

- **Skill and surface:** the RFP Contradiction Checker, in Cowork or a project chat on the near-final response.
- **What Claude does:** compares every answer with every other and with approved content, product documents and the website, then returns a punch list that quotes both sides, sorted by severity. It does not rewrite the response.
- **What to check:** the SMEs who wrote the answers pick the right version, and the bid manager signs off.
- **Watch for:** skipping a rerun after late edits and the executive summary.

<BlogCta id="Claude skills library" />

## What Claude Caught and Missed on a Public Tender

We ran each stage once on [State of Maine RFP# 202506081](https://www.maine.gov/dafs/bbm/procurementservices/sites/maine.gov.dafs.bbm.procurementservices/files/rfps/RFP%23%20202506081%20Electronic%20Death%20Investigation%20Case%20Management%20System_Final.docx), a state agency software RFP posted on July 15, 2025, for a fictional bidder; nothing was submitted. Each step ran in a fresh Claude Code session (Opus 5.5) on November 6, 2026, and the go/no-go, matrix and summary steps re-ran on skills version 1.1 on November 9, 2026.

| Stage                             | Result in one run                                                                                                                                                                                                        |
| --------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| 1. Qualify (v1.1)                 | No-Go, 49.2%; the deal-breaker override applied (CJIS, Q&A 175)                                                                                                                                                          |
| 2. Shred (v1.1)                   | Full pack: 457/457 (100%); mandatory 419/419. Text of the Word files only: 391/457 (85.6%); 390/390 of the items in its input; the "Not read" list names the Appendix D workbook first and asks for it. No invented rows |
| 3. Research                       | Not part of the run                                                                                                                                                                                                      |
| 4. Assign owners and prepare SMEs | Not part of the run                                                                                                                                                                                                      |
| 5. Draft                          | All 3 unsupported items were left as GAP or N. All 5 approved answers were verbatim. Without a gap instruction: Claude coded 1 of 10 not-offered or silent capabilities as met. With the gap rule: 0 of 10               |
| 6. Executive summary (v1.1)       | 447 words, 0 of 21 claims unsupported. It dropped the availability caveat that both earlier skill runs flagged                                                                                                           |
| 7. Contradiction check            | 8 of 8 planted contradictions found; severity matched the key on 5 of 8                                                                                                                                                  |

The misses were a workbook Claude never received, a prompt without a gap rule, and a caveat the summary dropped. So give Claude every file, keep the gap rule, and rerun the contradiction check after the summary.

Every figure is from one run, on bidder files and answer keys our team wrote, in Claude Code sessions that could run scripts, which a chat in the Claude app may not.

## Where Claude Stops on Bid Work

Yes, a bid team can use Claude to write RFP responses. Whether a general assistant is enough on its own is answered in [how far ChatGPT or Claude go against purpose-built RFP software](/blog/chatgpt-or-claude-vs-rfp-software). Plan for three limits:

- **Additions nobody approved.** Watch for a capability your files never mention marked as met, or new commitments in paragraphs Claude adds around approved wording. A gap rule and a quote-first instruction are the controls, and [Claude's strengths and weaknesses on bid work](/blog/claude-ai-strengths-and-weaknesses-for-rfps) has the detail.
- **Work across a team.** As I said in the walkthrough video, "You'll notice in Claude, collaboration is incredibly difficult." Several writers, approvals and answers kept current across bids need more than one person's setup, so weigh whether to build on general AI or buy RFP software.
- **Other assistants.** A side-by-side run of Claude, ChatGPT and Microsoft 365 Copilot, scored blind, is in progress.

{/* LINK (wave 3): when post 8 is live, add a down-link here, anchor "how Claude, ChatGPT and Copilot compare on RFP work" (slug claude-vs-chatgpt-vs-copilot-for-rfps). */}

## How Should a Proposal Team Set Up Claude?

A proposal team gets consistent results from Claude across people and bids when it shares four things: the plan, the skills, one project per tender and one approved answer library.

Choose the plan first. Use Team or Enterprise before a buyer's confidential documents go in: as of November 2026, both fall under [Anthropic's Commercial Terms](https://www.anthropic.com/legal/commercial-terms) and let owners control feedback, connectors and skills, while individual plans suit trying skills on public documents. For data rules by plan and what to send IT, see [is Claude safe for confidential bid documents](/blog/is-claude-safe-for-rfps).

On those plans, Anthropic says "organization Owners can provision skills for all users" ([What are skills?](https://support.claude.com/en/articles/12512176-what-are-skills), read November 9, 2026), so everyone runs one agreed version of each method. Give each live tender one project, shared with the people working on that bid, and let every skill draft from one approved answer library instead of an export that goes stale.

<ConceptFigure
  kind="layers"
  title="A bid team's Claude setup rests on one approved answer library"
  highlight={4}
  layers={[
    { label: 'Skills, one agreed version per stage', note: 'Provisioned for everyone on Team or Enterprise.' },
    { label: 'One project per live tender', note: 'The RFP, each attachment and every Q&A as it lands.' },
    { label: "Connectors with each person's permissions", note: 'The CRM, shared files and the answer library.' },
    { label: 'One approved answer library', note: 'Each answer approved, owned and sourced.' },
  ]}
/>

AutoRFP.ai's MCP connector gives a team's Claude setup the answers the team has already approved, each returned with its source: read scopes cover lookups, and full scopes let Claude write to projects and content as the signed-in user, with approvals staying in AutoRFP.ai's workflow.

Inside its own workspace, AutoRFP.ai drafts only from approved content, cites the source behind each answer and routes anything the content cannot support to a person, which is what [AI proposal software built on approved, sourced answers](/ai-proposal-software) means for a bid team that also works in Claude.

## Start With Two Skills on Your Next Bid

Start with your next live tender: give it one project, add every file the buyer sent, and run the go/no-go and compliance matrix skills before anyone writes.

<ConceptFigure
  kind="checklist"
  title="A live tender is ready for Claude when these checks are in place"
  checks={[
    {
      lead: 'Plan.',
      text: "The team is on Team or Enterprise, with feedback and connector settings checked, before a buyer's documents go in.",
    },
    {
      lead: 'Files.',
      text: 'Every attachment goes in as its own file, including any spreadsheet embedded in the RFP, and each Q&A is added the day it is published.',
    },
    {
      lead: 'Go/no-go.',
      text: 'The qualification skill has run and a person has answered its open questions before anyone writes.',
    },
    { lead: 'Matrix.', text: 'The compliance matrix covers the whole pack and every requirement has an owner.' },
    {
      lead: 'Gap rule.',
      text: 'Every drafting prompt asks for the file behind each answer and the word GAP when the files say nothing.',
    },
    {
      lead: 'Library.',
      text: 'Claude reads approved answers from the live library through a connector, so nobody maintains a second copy.',
    },
    {
      lead: 'Final check.',
      text: 'The contradiction checker runs on the near-final response and a person signs off every answer.',
    },
  ]}
/>

1. **Install the first two skills** from the skills library, with a Claude download for each stage. Each is free, and its download unlocks with an email address.
2. **Upload each .skill file** on claude.ai under **Customize** > **Skills**.
3. **Add the rest one stage at a time,** opening each stage's guide as you reach it.

<BlogCta id="Darkest Blue Demo CTA" />