An RFP lands with a date on it. Sixty questions, a scoring matrix, a security annex, and eleven working days. You are not shopping for proposal software this week — you are trying to answer the document without dropping the three deals already in your pipeline.
Search for help and nearly every result is an RFP platform offering prompts as a download. The prompts are usually fine. They are also a lead magnet, which means none of them will tell you where a chat window is the wrong tool. This page does both. For the wider workflow rather than the prompt set, start with how to answer RFPs and security questionnaires with AI.
Read the RFP before you answer it
Most lost bids are lost in the first hour, by a team that drafted before reading the scoring criteria. Two prompts, run on the document itself, before anyone writes a sentence.
Here is an RFP. Extract every question as a numbered list, worded exactly as it appears, with its section heading and any word limit. Mark each question that asks for evidence, a reference, or a certification rather than a description.
From this RFP, list the evaluation criteria and their weightings, every mandatory requirement, and anything stated as grounds for disqualification. Quote the wording for each. If a weighting is not stated, write NOT STATED rather than estimating one.
That last instruction is the whole prompt. A model asked to summarise scoring will invent a plausible weighting that appears nowhere in the document, and a bid plan built on it sends your best writer to the wrong section.
The evidence rule
An RFP answer is a contractual claim. Everything you submit should trace back to something your company has already approved: a previous response, a signed security page, a published policy, a product document. So one rule goes in front of every drafting prompt.
Answer this question using only the source material I have pasted below. Where the material does not contain the answer, write UNANSWERED and say what document or owner would be needed. Do not generalise from similar products and do not soften a gap into a vague claim.
Run a bid this way and the first pass splits in two: answers already defensible, and genuine unknowns with a named owner. That list is the useful output. A draft with no blanks is usually fiction.
Prompts that draft in the buyer's own language
Scoring panels reward answers that use the vocabulary of the question. Points are lost by answering a question about "incident notification timelines" with a paragraph about your platform's monitoring.
Rewrite this approved answer to respond directly to the question as worded. Open with the direct answer in one sentence, then the evidence, then the limitation if there is one. Reuse the buyer's terms for their systems and roles. Stay under [n] words. Do not add capability that is not in my source text.
More for this part of a deal: Solution Framing, Demo & Proposal.
The security and compliance section
This is where an assistant will get you in real trouble, because the answers look like boilerplate and are not. A model will happily assert a certification you do not hold. Never let a drafting prompt produce, unsourced: certification status, audit dates, data residency, subprocessor lists, retention periods, uptime commitments, breach history, or encryption specifics. Those come from one approved source or they do not go in the document.
Here is our approved security documentation and a questionnaire. For each question, either quote the approved answer or write REFER TO SECURITY OWNER. Never compose a new security claim. Where the approved answer is longer than the word limit, shorten it without removing any qualifier or condition.
Before any of this, check what your company allows you to upload. A customer's RFP is their confidential document, and a security questionnaire often arrives under an NDA with opinions about third-party processing. Ask whoever owns that policy once, then work inside the answer.
Prompts that find the questions that will lose the bid
Every bid has two or three questions where you are weaker than the incumbent. Finding them early is worth more than polishing the answers you already have.
Read these questions and my capability notes. Identify the five questions where my answer is weakest and explain what the buyer is likely trying to protect themselves against in each. Do not reassure me.
Play a sceptical procurement evaluator reading my answers. For each, name the follow-up question you would ask and what would make you score it lower than a competitor.
The second is the prompt the vendor downloads never include, and the one that changes a bid. Pressure-testing your own paper is the same discipline as pressure-testing a deal, which is why it also shows up in ChatGPT prompts for MEDDPICC.
Reusing last year's answers without last year's mistakes
Reuse is where the speed comes from and where stale claims creep in. Two prompts:
Compare this question with these three previous answers. Rank them by fit and say exactly what would need to change in the best one to answer the question as worded.
Review this reused answer against our current product documentation. List every statement that is now out of date, no longer true, or describes a feature that has been renamed.
Executive summary and win themes
The summary is often the only part the decision-maker reads, and should be built from the buyer's stated priorities, not your feature list.
Here are the buyer's stated objectives, in their words, and my differentiators. Draft a one-page executive summary that opens on their objective, names three win themes, and ties each to evidence I have supplied. No superlatives. Nothing I have not given you.
If the summary has to survive an internal committee after you submit it, pair it with a proposal your champion can sell internally. Pricing and terms questions belong with Negotiation, Procurement & Closing.
Take one prompt free, with no account, at /free-prompt. Pro is $12 per month or $89 per year, and lifetime access is $149 — See pricing.
Where RFP software genuinely beats a chat window
A chat window has no answer library with ownership and review dates. It has no version control when four contributors edit the same response. It leaves no audit trail of who approved a security claim, which matters the day a customer asks. It does not fill a submission portal or flag that an answer you reused was marked out of date in March.
One bid a quarter, handled alone: prompts and a folder are enough. Several a month with a bid team: buy the software and use prompts inside it. The decision is about volume and contributors, not prompt quality.
Either side of the document, AI prompts for account executives covers the stages, Deal Strategy & Stakeholder Management holds the stakeholder prompts, and AI prompts for sales maps the catalogue. Browse the library → or read the full taxonomy.
Frequently asked questions
Can ChatGPT write an RFP response?
It can draft one from source material you supply, and it is good at matching an approved answer to the wording of a question. It should not generate claims about your product, security posture or certifications from nothing. The workable split is that the model restructures and shortens what your company has already approved, and a person owns anything new.
What is a good prompt for answering RFP questions?
One that constrains the source. Paste the question and your approved material, then instruct the model to answer only from that material, to open with the direct answer in one sentence, and to write UNANSWERED where the material does not cover it. The instruction to leave a gap visible is what makes the output usable.
How do I use AI for a security questionnaire?
As a matching exercise, not a writing one. Give it your approved security documentation and have it return the approved answer for each question, or flag the question for your security owner. Never let it compose a new security claim, and never let it shorten one in a way that drops a qualifier or a condition.
Is it safe to put RFP documents into ChatGPT?
That depends on your company's policy and the tier you are using, not on the model. The document belongs to the buyer and often arrives under an NDA, so ask whoever owns data policy before the first bid rather than during it. Many teams land on a rule of redacting the buyer's name and using only an approved business account.
Should I use RFP software or ChatGPT?
Volume and contributors decide it. One bid a quarter handled by one person works fine with prompts and a well-organised folder. Several bids a month with multiple contributors needs an answer library, version control and an audit trail, which is exactly what the software sells and what a chat window cannot give you.