Question extraction identifies the response questions inside your tender documents and populates them into the question workspace, ready for planning and drafting.
Extraction is a conversation with BidBot rather than a button you press. BidBot already has full contextual understanding of every document uploaded to the project, so it can tell you what is in the tender pack before it extracts anything from it — and it works with you as it goes.
The whole process is human-in-the-loop. BidBot will not extract indiscriminately or act without your direction, so you can explore the tender pack with it safely before committing to anything.
Where to find BidBot
Open your project from the
Projectsarea.In the navigation bar across the top of the project, click
BidBot- the first item on the far left.
If your organisation has renamed BidBot in its settings, look for that name in the same position. See Team Content Controls for how bot branding is configured.
Before you start
Wait until document indexing has reached 100% and the indexing indicator in the top-right of the screen has disappeared. Extracting from a part-indexed document risks missing questions.
The simplest way to extract
You can open BidBot and simply ask it to extract questions:
"Extract questions."
BidBot will work through the uploaded documents with you, raising any unusual structures it finds and asking how you want them handled.
This works. But you will get better, more controlled results from a targeted prompt — and on most real tenders, targeting is not optional.
Why targeting matters
Very few tenders contain a single, uniform set of questions. Even a straightforward open tender with no lot structure will typically issue several distinct sets at once:
Qualification and compliance questions — known in the public sector as PQQs, SQs or PSQs.
ITT or quality questions — the scored written responses.
Commercial questions — pricing and commercial terms.
BidScript handles each of these differently, and you will rarely want to answer all of them at the same moment, in the same sitting, by the same people. Targeting your extraction lets you bring in exactly the set you are working on.
The clearest case is a multi-lot tender. If a procurement has five lots and you are only bidding for lots 1 and 4, you can have BidBot extract only the questions relating to those two lots and ignore lots 2, 3 and 5 entirely - rather than extracting all five and manually unpicking the ones you do not need.
The recommended two-step method
BidScript recommends a two-step extraction for every tender: understand first, then target.
This is particularly valuable when you do not yet know the tender pack in depth, or when several different question sets exist - which is the norm in most tender and RFP environments.
Step 1 - ask BidBot to summarise the procurement
Before extracting anything, ask BidBot what you are dealing with.
"The tender documents have been uploaded to this project. Please summarise the procurement for me: what questions exist, what they exist in relation to, and whether there are any lot structures."
BidBot returns a comprehensive breakdown covering:
Every document in the pack, and what each one is for
The overall context of the procurement
Any distinctive structures, such as lot structures or question gateways
A list of all the questions it has found, and which document each belongs to
Read this before going further. It is the map you will use to target step 2, and it often reveals structures - gateways, lot-specific sections, duplicated question sets - that are not obvious from opening the files yourself.
Step 2 - target the extraction
Now use what BidBot told you to ask for exactly the questions you want.
You can extract everything, or work one set at a time.
Targeting by question type:
Please extract the quality questions only — the ITT questions identified in {DOCUMENT NAME}. Populate them into the question workspace.Targeting by lot:
We are bidding for lots 1 and 4 only. Please extract only the questions relating to those two lots, and ignore the questions for lots 2, 3 and 5.
Repeat for each set you want to bring in. Working in deliberate blocks reduces errors, avoids timeouts, and keeps you in control of what has and has not been captured.
Writing an effective prompt
Whatever you are targeting, BidBot extracts most accurately when your prompt gives it two things:
Orientation - which file, and roughly how many questions.
Format context - how the questions are laid out.
Step 1 of the two-step method gives you both, which is why it is worth doing even when you think you know the pack.
Targeting Excel documents
Excel files are one of the most common formats for bulk questions. They can contain thousands of rows, multiple tabs and varying structures, so precise instructions matter most here.
When all questions are in a single tab
Go to the Files tab and open the Excel document. Use the visible row count to estimate how many questions it contains.
Open BidBot and complete step 1 above.
Ask for the extraction, naming the file and the question volume.
Example prompt:
In the uploaded tender files there is an Excel spreadsheet titled {FILE NAME} that contains {NUMBER OF QUESTIONS} questions. Please extract them all and populate the question workspace.When a single tab has its own internal structure
A single worksheet often contains section headings, merged-cell titles or separator rows that act as natural breaks between blocks of questions. These structures are very useful for controlled extraction.
Look for separators such as:
"Section A — Technical Questions"
"Lot 1 Requirements"
"Commercial Questions"
Empty rows used as structural breaks
Colour-coded subsections
Where these exist, extract each section in turn.
Example prompt:
We have an Excel document titled {FILE NAME} uploaded to this project containing {X} questions. These questions follow the structure {BRIEFLY DESCRIBE STRUCTURE INCLUDING SEPARATORS}. I want to begin with the first set of {X} questions listed under the section titled {SECTION NAME}. Please extract this specific block of questions and populate them into the workspace editor. We will handle the remaining sections afterward.Repeat for each section until every question you need has been extracted.
When questions are split across multiple tabs
Workbooks with multiple worksheets are common in large procurements, where each tab often represents a different scoring area, submission schedule or section of the ITT.
Go to the Files tab and inspect the workbook. Note each tab and estimate the number of questions in it.
Open BidBot and complete step 1 above.
Instruct BidBot one tab at a time.
Example prompt:
Extract all questions from the uploaded Excel file. Let's first focus on Tab {X}, titled {TAB NAME}. There are {NUMBER OF QUESTIONS} in this tab. {BRIEFLY DESCRIBE STRUCTURE INCLUDING SEPARATORS}. Please extract all questions from this tab and populate them into the workspace editor.Repeat for each remaining tab. Working this way preserves tab order, keeps section grouping intact, ensures no questions are skipped, and leaves you in control of extraction timing and load.
Targeting Word and PDF documents
Word and PDF files often carry large volumes of background information interspersed with several types of question - general information, technical, clinical, social value, and pass/fail compliance questions among them.
Read the document first
Where a document mixes extensive background with multiple question sets, look at it yourself before prompting.
Go to the Files tab.
Open the Word or PDF document.
Review its overall structure, the types of question it contains, where those questions appear, and what supporting information surrounds them.
Then open BidBot.
Documents with little background information
Where most of the content is questions to be answered, one prompt is enough.
In the uploaded tender files, I have a {DOC/DOCX/PDF} file titled {FILE NAME} that contains {X} questions I need to answer. Please extract all questions and populate them into the question workspace.Documents with substantial background information
Where the document carries a lot of context alongside the questions, use the two-step method in its narrowest form.
Step 1 - familiarise BidBot with the document:
In the uploaded tender files, I have a {DOC/DOCX/PDF} file titled {FILE NAME} that contains a large amount of background information. Please familiarise yourself with the document. {Briefly explain the structure and relevance of the surrounding information}.Read BidBot's reply and confirm its understanding is accurate before continuing.
Step 2 - extract:
Great. Now please extract all {X} questions from that document, focusing specifically on the questions relevant to me as a potential supplier.
Documents with multiple question sets
Where a document contains several categories of question - general information, compliance, technical and social value, for example - work one category at a time.
Step 1 - identify the question types:
In the uploaded tender files, I have a {DOC/DOCX/PDF} file titled {FILE NAME} that contains a large amount of information. Please familiarise yourself with the document. {Briefly explain how the questions are structured, e.g. general information, technical, compliance, social value}.Read BidBot's reply and confirm its understanding before continuing.
Step 2 — extract one type:
Great. Now please extract all {X} {TYPE OF QUESTIONS} questions from the document, focusing specifically on the {TYPE OF QUESTIONS as indicated by the relevant section headers}. Populate these into the question workspace for me.Repeat for each remaining question type you need.
Working with very large question sets
Where a single document contains 250 questions or more, those questions are almost always presented with some inherent structure - tabs, headings, section titles, column separators or formatting cues.
BidScript works with these structures effectively, but only when your prompt names them. Use step 1 to establish the structure, then extract block by block.
Monitor an extraction
While BidBot works, panels in the chat show what it is doing:
Documents being read appear in a panel labelled Reading document content.
Extracted questions populate a table in the question editor, on the right-hand side of the split screen.
The questions found, and the sources they came from, appear in a purple-bordered panel titled Inserted Questions.
Do not close the tab or navigate away while extraction is running. The process must complete uninterrupted.
Related articles
Managing the question table - columns, word limits, weightings, assignment and statuses once your questions exist.
Section classification in response documents - making sure the right sections are marked as Bid Response before you extract.
Document Workspace: Overview - the full path from upload to extracted questions.
Team Content Controls - renaming BidBot for your organisation.