# Check an RFP response before you submit it: prompt chain Check a response to a request for proposals against every mandatory item, every judged criterion and every attachment, test its figures against your own records, and get a ready, ready after fixes or not ready verdict with a fix list that you decide on. **Made for:** Anyone who prepares or signs off a bid, such as a proposal manager, a grants officer, an office manager or a small business owner. ## Before you start Have these ready: - **The Call For Proposals** (required): Paste the requirements, the judging criteria and the submission rules exactly as issued. Paste only what your organization lets you put into an AI tool. - **Your Draft Response** (required): Paste the full text as it stands now, including any price table. - **Your Records** (optional): Paste the price sheet, the job list and any certificate wording that the response should match. Strongly advised. Without them, every figure is marked no record and only the response and the call are checked. **How to use this chain** 1. Open a new chat in any AI assistant. 2. Copy the Step 1 prompt, replace every [PASTE HERE: ...] slot with your material, and send it. 3. Keep each answer: later steps ask you to paste it in. Run each step in the same chat or a new one; a new chat keeps steps from blurring together. 4. After a step that has a check, run the check before you move on. ## Step 1: List Every Item The Call Asks For Split the call into separate items, each quoted word for word and marked as a must, a judged criterion, a format rule, a form or a deadline. Copy this prompt: ````text You are a careful bid reviewer who reads a call for proposals line by line. Your task is to list every item in the call below. The call: [PASTE HERE: The Call For Proposals] Rules: - One row for each item. Split a sentence that holds two items. - Quote each item from the call word for word, as short as you can while keeping its numbers and conditions. Do not correct it and do not reword it. - In the Type column, write exactly one of: must, judged, format, form, deadline. - Must is something the response has to contain. Judged is a criterion with a weight. Format is a length or file rule. Form is a document to attach. Deadline is a date or time. - Do not add an item the call does not have. If the text does not look like a call for proposals, return no table and say so in one line. Return one markdown table and nothing else: | # | Type | Item (quote) | ```` **You get back:** A numbered table of every item the call asks for, each quoted from the call and typed. **Check after Step 1: Every Item Quotes The Call** Check by eye: Each quote in the Item (quote) column appears in the call for proposals, with shortened fragments in the same order. If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. ## Step 2: Find Where The Response Answers Each Item For each item, quote the words in the response that answer it, or say that none do. Copy this prompt: ````text You are a careful bid reviewer who searches a response for the words that answer each item. Your task is to find, for each item in the table below, where the response answers it. The items: [PASTE HERE: the answer from Step 1 (List Every Item The Call Asks For)] The response: [PASTE HERE: Your Draft Response] Rules: - One row for each item. Keep its number and its quote. - In the Where It Is Answered (quote) column, quote the words in the response that answer the item, word for word and as short as you can, or write exactly None found. A passage that only names the topic does not count. Do not join two passages. - In the Your Check column, write Count the pages yourself for a format row measured in pages, Check the date yourself for a deadline row, and for a format row measured in words, show the count of words in the response against the limit. For any other row, write Nothing. - For a form row, quote where the response says the form is attached, or write None found. - Use only the response above. Do not guess an answer from what is usual. Return one markdown table and nothing else: | # | Item (quote) | Where It Is Answered (quote) | Your Check | ```` **You get back:** A table that puts the words of the response beside each item, or None found, with the checks left to you. **Check after Step 2: Every Answer Quotes The Response** Check by eye: Each quote in the Where It Is Answered (quote) column appears in the your draft response, with shortened fragments in the same order. If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. ## Step 3: Judge Each Item Say whether each must, judged and form item is covered, partly covered or not covered, and whether the response backs it with a record. Copy this prompt: ````text You are a careful bid reviewer who decides whether a passage gives everything an item asks for. Your task is to judge each must, judged and form item below against the words found for it. The items and the passages found: [PASTE HERE: the answer from Step 2 (Find Where The Response Answers Each Item)] The items by type: [PASTE HERE: the answer from Step 1 (List Every Item The Call Asks For)] Rules: - One row for each item whose Type is must, judged or form. Leave out the format and deadline rows; a person checks those. - In the Coverage column, write exactly one of: covered, partly covered, not covered. Covered means the passage gives everything the item asks for. Partly covered means it gives some of it. Not covered means None found. - In the Backed By column, write exactly one of: record, claim only, not needed. Record means the response gives a figure, a named job, a named person or a named document. An item that asks for proof, such as cover or a certificate, is claim only unless the response attaches or names the proof. Claim only means it states something with nothing behind it. Not needed is for a form row. - In the What Is Missing column, name the one thing that is absent, or write Nothing. - Do not give credit for a passage that only names the topic. Return one markdown table and nothing else: | # | Item (quote) | Type | Coverage | Backed By | What Is Missing | ```` **You get back:** A table that judges each item as covered, partly covered or not covered, with what backs it and what is missing. **Check after Step 3: Every Coverage Cell Is One Of The Three Allowed Words** Check by eye: Every row of the table has one of "covered", "partly covered" or "not covered" in the Coverage column. If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. ## Step 4: Tie The Response To Your Records Match each price, total, date range, job and certificate in the response to your records and recompute every total. Copy this prompt: ````text You are a careful bid reviewer who matches a response's claims to the records behind them. Your task is to tie each price, total, date range, job and certificate in the response below to the records. The response: [PASTE HERE: Your Draft Response] The records: [PASTE HERE: Your Records] Rules: - One row for each price, total, date range, named job and certificate the response states. - In the Response States column, quote it word for word, as short as you can. - In the Record column, quote the matching passage from the records word for word, as short as you can. If no passage speaks to it, write exactly No record. - In the Verdict column, write exactly one of: matches, differs, no record. - In the Sum column, recompute every total from its parts and show the sum. For any other row, write Not a total. - If the records are empty or say none given, write no record in every Verdict cell and say so in one line below the table. - Do not choose which figure is right when they differ. Show both. Return one markdown table and nothing else, apart from the one line if the records are empty: | # | Response States | Record | Verdict | Sum | ```` **You get back:** A table that sets each price, job and certificate beside your records, with a verdict and the sum for every total. **Check after Step 4: Every Verdict Is One Of The Three Allowed Words** Check by eye: Every row of the table has one of "matches", "differs" or "no record" in the Verdict column. If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. ## Step 5: Give The Verdict And The Fix List Turn every gap and mismatch into one fix row, then give one verdict by a fixed rule and stop. Copy this prompt: ````text You are a careful bid reviewer who gives a clear verdict and a short fix list and leaves the decision to the person. Your task is to give the verdict and the fix list for the response below. The call: [PASTE HERE: The Call For Proposals] The response: [PASTE HERE: Your Draft Response] The records: [PASTE HERE: Your Records] The items: [PASTE HERE: the answer from Step 1 (List Every Item The Call Asks For)] The judgment of each item: [PASTE HERE: the answer from Step 3 (Judge Each Item)] The response tied to the records: [PASTE HERE: the answer from Step 4 (Tie The Response To Your Records)] Rules: - Make one row for each problem: a must or judged item that is not covered or is partly covered, a form that is not covered or is partly covered, a Backed By of claim only, and a verdict of differs. A problem raised in two tables gets one row. - The row has four cells. Where quotes the words in the response, or says The whole response if nothing answers the item. Problem says in plain words what is wrong. Evidence quotes the call, the record or the missing item. Suggested Wording uses only facts and numbers from the response and the records. - If no safe wording exists for a claim, write exactly Cut it or find a record first. For an item with no passage, write exactly Add it from your own records; the response gives nothing to draw on. - Do not write a number of your own in the table, and do not write a step number. Do not work out a new total. - Verdict by this rule only: Not ready if any must item is not covered, any figure differs, or a form is not covered. Ready after fixes if any other row was raised. Ready by these checks if none was. - If there are no problems, write one row with None found in every cell. - Write the verdict line first, in words only, as "Verdict:" then the verdict then one deciding reason. On the next line write exactly: This verdict comes from the checks above. An AI tool can miss things, so read the flagged parts yourself. - On the third line write "Check Yourself:" and quote the deadline and format items of the call, then say these are for the person to confirm. - Do not rewrite the response, and do not choose to submit, hold or fix. The person decides. Return the three lines, the fix table and, under the heading How Each Item Was Judged, the judging table from the judge step, and nothing else: | Where | Problem | Evidence | Suggested Wording | ```` **You get back:** A verdict, a Check Yourself line for the deadline and format, and a fix table with a row for every problem. **Check after Step 5: No Number Appears That The Call, The Response Or The Records Do Not Have** Check by eye: Every number in the answer also appears somewhere in at least one of "the call for proposals", "your draft response" and "your records"; this does not check what each number means. If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. ## Final checks Run these on the finished result. **Final check 1: Every Type Is One Of The Five Allowed Words** Check by eye: Every row of the table has one of "must", "judged", "format", "form" or "deadline" in the Type column. If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. **Final check 2: No Requirement In The Call Was Skipped** Open a new chat (so the check is not swayed by the work) and paste this check prompt: ````text You are a careful reviewer. You do not fix or rewrite the work; you only decide whether it passes one check. The check: No Requirement In The Call Was Skipped Read the call below. Find every sentence that holds must, shall, required, attached, a weight or a due date. Does each one have a row in the table? Answer FAIL and quote each sentence with no row. The call: [PASTE HERE: The Call For Proposals] The work to check (the answer from Step 1 (List Every Item The Call Asks For)): [PASTE HERE: the answer from Step 1 (List Every Item The Call Asks For)] Reply with exactly PASS or FAIL on the first line. On the second line, say in one sentence why. If it fails, name the exact item or line that fails. Do not add anything else. ```` If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. **Final check 3: Each None Found Was Searched For** Open a new chat (so the check is not swayed by the work) and paste this check prompt: ````text You are a careful reviewer. You do not fix or rewrite the work; you only decide whether it passes one check. The check: Each None Found Was Searched For For each row marked None found, name the key word or number the item needs, then look in the response below for a line that holds it. Answer FAIL and quote each line that answers an item marked None found. The response: [PASTE HERE: Your Draft Response] The work to check (the answer from Step 2 (Find Where The Response Answers Each Item)): [PASTE HERE: the answer from Step 2 (Find Where The Response Answers Each Item)] Reply with exactly PASS or FAIL on the first line. On the second line, say in one sentence why. If it fails, name the exact item or line that fails. Do not add anything else. ```` If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. **Final check 4: Every Backed By Cell Is One Of The Three Allowed Phrases** Check by eye: Every row of the table has one of "record", "claim only" or "not needed" in the Backed By column. If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. **Final check 5: Every Covered Row Is Backed By Words In The Response** Open a new chat (so the check is not swayed by the work) and paste this check prompt: ````text You are a careful reviewer. You do not fix or rewrite the work; you only decide whether it passes one check. The check: Every Covered Row Is Backed By Words In The Response For each row marked covered, quote from the response below the words that give each thing the item asks for. Answer FAIL and name each covered row where a thing the item asks for has no words. The response: [PASTE HERE: Your Draft Response] The work to check (the answer from Step 3 (Judge Each Item)): [PASTE HERE: the answer from Step 3 (Judge Each Item)] Reply with exactly PASS or FAIL on the first line. On the second line, say in one sentence why. If it fails, name the exact item or line that fails. Do not add anything else. ```` If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. **Final check 6: Every Total Was Recomputed** Open a new chat (so the check is not swayed by the work) and paste this check prompt: ````text You are a careful reviewer. You do not fix or rewrite the work; you only decide whether it passes one check. The check: Every Total Was Recomputed Look at the response below. Count the totals it states, such as a fixed price made of parts. Does the table show a sum for each one, and is each sum right? Answer FAIL and quote each total with no sum or a wrong sum. The response: [PASTE HERE: Your Draft Response] The work to check (the answer from Step 4 (Tie The Response To Your Records)): [PASTE HERE: the answer from Step 4 (Tie The Response To Your Records)] Reply with exactly PASS or FAIL on the first line. On the second line, say in one sentence why. If it fails, name the exact item or line that fails. Do not add anything else. ```` If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. **Final check 7: Every Record Quote Is In Your Records** Open a new chat (so the check is not swayed by the work) and paste this check prompt: ````text You are a careful reviewer. You do not fix or rewrite the work; you only decide whether it passes one check. The check: Every Record Quote Is In Your Records For each row of the table whose Record is not No record, is that passage in the records below, word for word? Answer FAIL and quote each passage that is not. The records: [PASTE HERE: Your Records] The work to check (the answer from Step 4 (Tie The Response To Your Records)): [PASTE HERE: the answer from Step 4 (Tie The Response To Your Records)] Reply with exactly PASS or FAIL on the first line. On the second line, say in one sentence why. If it fails, name the exact item or line that fails. Do not add anything else. ```` If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. **Final check 8: The Line About Reading The Flagged Parts Is There** Check by eye: The answer contains the text fragment "This verdict comes from the checks above. An AI tool can miss things, so read the flagged parts yourself." (including inside a longer word). If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. **Final check 9: Every Problem Has A Row** Open a new chat (so the check is not swayed by the work) and paste this check prompt: ````text You are a careful reviewer. You do not fix or rewrite the work; you only decide whether it passes one check. The check: Every Problem Has A Row Count the problems in the two tables below: each must, judged or form row that is not covered or is partly covered, each Backed By of claim only, and each verdict of differs. Count a problem raised in both tables once. Does the fix table have one row for each? Show both numbers. Answer FAIL and name each problem with no row. The judgment of each item: [PASTE HERE: the answer from Step 3 (Judge Each Item)] The response tied to the records: [PASTE HERE: the answer from Step 4 (Tie The Response To Your Records)] The work to check (the answer from Step 5 (Give The Verdict And The Fix List)): [PASTE HERE: the answer from Step 5 (Give The Verdict And The Fix List)] Reply with exactly PASS or FAIL on the first line. On the second line, say in one sentence why. If it fails, name the exact item or line that fails. Do not add anything else. ```` If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. **Final check 10: The Verdict Follows The Written Rule** Open a new chat (so the check is not swayed by the work) and paste this check prompt: ````text You are a careful reviewer. You do not fix or rewrite the work; you only decide whether it passes one check. The check: The Verdict Follows The Written Rule The rule is: Not ready if any must item is not covered, any figure differs, or a form is not covered. Ready after fixes if any other row was raised. Ready by these checks if none was. Use the two tables below. Does the verdict follow the rule? Answer FAIL and say which line of the rule it breaks. The judgment of each item: [PASTE HERE: the answer from Step 3 (Judge Each Item)] The response tied to the records: [PASTE HERE: the answer from Step 4 (Tie The Response To Your Records)] The work to check (the answer from Step 5 (Give The Verdict And The Fix List)): [PASTE HERE: the answer from Step 5 (Give The Verdict And The Fix List)] Reply with exactly PASS or FAIL on the first line. On the second line, say in one sentence why. If it fails, name the exact item or line that fails. Do not add anything else. ```` If it does not pass: Run the step once more, and add what was wrong to the end of the prompt. If it still does not pass, fix it by hand and make a note of it. **Final check 11: A Person Decides Whether To Submit, Hold Or Fix** Ask yourself: Check the deadline and the length rule against the response yourself, count the pages, and confirm the forms are attached and signed. For each figure that differs, decide which number is right. A person who can sign for your organization must read the call and the response together before anything is submitted. If it does not pass: Carry on, but write down the problem so whoever uses the result knows about it. ## What you end up with - A verdict of ready, ready after fixes or not ready, with the one reason that decides it and a line of items you must check yourself. - A fix table with a row for every gap, unsupported claim and mismatch, each with the words in the response, the evidence and a safe suggestion. - Tables that list every item in the call, show where the response answers it, say whether it is covered and backed, and tie the figures to your records. ## Changing this chain - This template checks a response. It does not write one, compare suppliers or decide whether to bid, and it cannot count pages or confirm a signature. Those are left to you. - The checks are only as good as your records. Paste the price sheet, the job list and the certificate wording. With no records, every figure is marked no record and only the response against the call is checked. - Paste the call exactly as issued, including the judging criteria and the submission rules, because a missed condition is the usual way a response is rejected. - Paste only what your organization lets you put into an AI tool. Leave out personal details of the people named as contacts if you can. - Each fix row suggests wording only from the response and your records. Where no safe wording exists it says so, and you supply it. --- From Logic Lab (logiclabhq.com). Free to use and adapt, with credit.