Construction Tender Writing With AI: Draft, Then Check Against Scoring

Pull the scoring rules out of the tender documents first, draft each answer from your own evidence notes, and finish by asking AI to mark the draft against the published criteria. Marks come from specific proof tied to each question, not from polished prose. Keep tender documents and client data out of consumer AI plans.

AI cannot know your projects, your people or your safety record, so a draft written from its general knowledge reads like every other bid and scores like one. The useful sequence is the reverse: gather your evidence, let AI shape it to the question, then use AI as a harsh first marker before a colleague reads it.

Follow me on Instagram@sagnikteaches

Read the scoring before you read the questions

The tender documents tell you how you will be marked, and most small contractors read the questions first and the scoring last. Reverse that. From the instructions and the evaluation section, extract the weightings between quality, price and any other category such as social value; the sub-criteria under each; the scoring scale and what each score means; word or page limits; any pass-or-fail questions or minimum scores; the price scoring method; and the dates for clarification questions and submission.

Connect on LinkedInSagnik Bhattacharya

One common price method gives the lowest compliant price the full price marks and scores everyone else in proportion: price marks equal the price weighting multiplied by the lowest compliant price divided by your price. Other methods exist, including linear scales with a cut-off and scales built around the median bid, so the method printed in your tender documents is the one to use. The figures below use the proportional method with illustrative weightings: quality 60 marks and price 40. Quality splits into method statement 20, programme 10, team and resources 10, health and safety 10 and social value 10, each scored 0 to 5 and weighted by the share of the quality marks (a score of 3 on the method statement earns 3 out of 5 of its 20 marks, or 12).

Subscribe on YouTube@sagnikteaches

Say the lowest compliant price is $812,000 and yours is $856,000. Your price score is 40 times 812,000 divided by 856,000, which is 37.94. Here is how three bids compare:

On equal quality the cheaper rival wins by 2.06 marks, which is exactly the gap between 37.94 and 40. Matching the lowest price would add at most 2.06 marks to your total, and a 5% price cut, to $813,200, adds just 2.00. One scoring band on the method statement adds 4. That is why the third row wins. These are illustrative weightings and a common formula, not a rule, so redo the sum with your tender's own numbers. The point carries over: find the question where one band is worth most, and put your hours there. If a question carries a minimum score, failing it can remove you regardless of price.

Whether the job is worth pricing at all is a separate decision, and the bid or no-bid screen deals with it before you spend a day on any of this.

Build a compliance matrix from the invitation to tender

Paste the invitation, the instructions and the evaluation section into an assistant and ask for a matrix. The matrix is your project plan for the bid: every row is a question, and every row has an owner.

From the tender documents below, build a compliance matrix with columns:
Q ref | Question (short) | Weight | Word or page limit | Scoring descriptors for scores 5 and 3 (quoted) | Evidence we need | Owner | Status

Then list separately:
- Every mandatory requirement (words such as must, shall, required) with its location.
- Pass or fail questions and any minimum score.
- Forms and appendices to complete, and how they are submitted.
- All dates: clarification deadline, submission deadline, site visit, award.
- The price scoring method, quoted word for word.
Rules: quote, do not paraphrase. If something is unclear, list it under "Questions for the buyer". Do not guess weights.

DOCUMENTS:
[paste]

Illustrative output for two rows:

Three faults turn up in output like this. The assistant merges sub-questions 2a and 2b into one row, so a half-answer slips through. It overlooks an appendix with a form that must be completed. And it mistakes the maximum score for the weight. Read the matrix against the documents once, then add the questions for the buyer to your clarification list before that deadline passes, because a clarification question costs nothing and a wrong assumption costs marks.

Clarification questions that protect your marks

The matrix's "Questions for the buyer" list is where ambiguity turns into a decision. Good clarification questions point at a sentence in the documents, ask one thing each, and could be answered without helping any rival. Assume the answer, and sometimes the question, will be shared with every bidder, so write nothing you would not want a competitor to read. Ask the assistant to draft questions from the unclear items it found, then cut the list to the few that change how you write or price. Illustrative examples:

Q1. Section 4.2 sets a 600-word limit for the method statement. Do headings, tables and diagrams count towards the limit?
Q2. The price evaluation refers to the "lowest compliant price". Please confirm whether abnormally low tenders are excluded before that price is identified.
Q3. Question 5 asks for social value commitments. Is a completed social value form required in addition to the 300-word answer?

Send them before the clarification deadline, quote the section number every time, and keep the buyer's replies in the bid folder. A reply can change a word limit or a scoring rule, and the bid must follow the final answer, not the first reading.

Build an evidence bank the assistant is allowed to use

A tender answer scores when it proves something, and proof lives in your records, not in a model's training. Keep one document of dated, factual entries: projects with scope, value band, duration and a result with a figure; people with roles and qualifications; your safety record and procedures; quality checks; equipment; supply chain; and real social value activity. Tag each entry with the question types it serves. The rule for every prompt is that the assistant uses entries from the bank and writes [EVIDENCE NEEDED] wherever it has nothing. Illustrative entry:

PROJECT P-07
Scope: [internal refurbishment of an occupied school hall: floor, lighting, acoustic ceiling]
Value band: $[ ]   Duration: [ ] weeks against a [ ]-week programme
Our role: main contractor, [n] operatives, [n] subcontract trades
Constraint: work only in school holidays and evenings
Result: [measurable: handed over [x] days early; no reportable accidents; [n] snags at inspection]
Evidence: client letter [ref]; photographs [ref]; programme [ref]
Relevant to: method, programme, health and safety, social value

The square brackets are for you. Fill them with real numbers from your own records, and never let the assistant supply them. If you do not have a number, the honest entry is "no figure recorded", and that is a signal to start recording it on the next job. Photographs and documents you can attach beat adjectives every time, which is also the logic behind the case study tutorial for construction firms.

Draft one answer: the method statement section

The method statement usually carries the most marks, so draft it first. An answer that scores well covers, in roughly this order: what you understand about this site and its constraints; the sequence of work with durations; the resources, meaning who and how many; the risks specific to this site and how you control them; the quality checks and hold points; how you deal with the client, neighbours and other trades; how it all links to the programme; and one measurable commitment. If you have no method statement format yet, the tutorial on writing a method statement for a small build covers the content in more depth.

Draft the answer to Question 2 using ONLY the evidence and site facts below.
Word limit: 600. Address each element of the score 4 and score 5 descriptors, in the order the question asks.
- Use specific site facts from the brief. Do not use generic claims.
- No evaluative words such as excellent, high-quality or extensive unless a figure or document follows in the same sentence.
- Where the evidence does not support a point, write [EVIDENCE NEEDED] and move on.
- End with a short list of everything you assumed.

QUESTION AND DESCRIPTORS: [paste]
SITE FACTS: [paste constraints, access, working hours, neighbours]
EVIDENCE: [paste P-07 and relevant entries]

The difference between a draft that scores and one that does not shows up at sentence level:

Expect to fix something in the assistant's draft every time. A typical slip is a commitment you cannot keep, such as "we will provide a 3D model" when nothing in the evidence supports it. Delete it. A tender promise can become a contract obligation, so every sentence must be one you are willing to deliver.

Case studies that earn marks: relevance, proof and result

Evaluators read dozens of case studies that say the same thing. The ones that score connect a past project to this tender's specific constraint. A four-part structure works: the situation (a constraint similar to this site), what you did, the result with a figure and its source, and one sentence on why it matters for this question. Give the assistant your evidence-bank entry and the question, and ask it to cut to the word limit while keeping every figure.

Rewrite the project entry below as a case study of 150 words for Question [n].
Structure: situation, action, result, relevance to this tender.
Keep every figure and document reference exactly as given. Add no adjectives and no facts. If a result has no figure, say so in a note after the text.

ENTRY: [paste]
QUESTION: [paste]

Check the result against a claim-and-proof table you build yourself, with each claim in the answer on the left and the document that proves it on the right. Illustrative rows:

A claim with an empty right-hand cell is either cut or rewritten as an intention you control, such as "we will record and report", which is safer than a boast and still specific.

Thin evidence: social value, team and programme answers

Some answers will have thin evidence, and the temptation is to let the assistant fill the gap with plausible content. Resist it. For social value, list only commitments you can deliver and measure: for example, the share of spend with suppliers close to the site on your last three jobs if you record it, or a training placement if you actually run one. If you have nothing, say what you will start recording and report on, which is a commitment you control. For team answers, name roles and the people holding them, with their qualifications, and note who covers absence. For programme, attach a simple chart and write the narrative around the dates on it, not around adjectives.

A bid that promises more than you can deliver scores well once and then costs you money for the length of the contract. The scoring rewards specific, evidenced commitments, so lean towards smaller promises that you can prove.

Mark your own draft like an evaluator

Once every answer exists, ask an assistant to play the evaluator against the buyer's own descriptors. The prompt matters because a request to "review this tender" produces praise.

You are a strict tender evaluator. Score each answer below from 0 to 5 using ONLY the descriptors provided.
For each question return: score | the sentences that earned marks | what stopped it scoring one higher | the single change that would add most.
Rules: if evidence is claimed but not shown, give it no credit. Do not rewrite the answer. Do not give an overall verdict.

DESCRIPTORS: [paste]
ANSWERS: [paste]

Illustrative output for two questions:

Treat the scores as a ranking of weaknesses rather than a prediction. Assistants can be generous, so calibrate: run your last tender through the same prompt and compare with the score the buyer gave you in its debrief, then note the gap in a small log. Run important answers through two different assistants and look at where they disagree. Then give the draft to a colleague who has not seen it and ask them to mark it with the same descriptors. If you have only one day left, the colleague beats the assistant, but with two days you can have both.

A five-day plan for a small team

The price is never an AI job. Your estimator builds it, checks the form's arithmetic and signs it off, and the overview of AI for tenders and RFP responses explains where the assistant fits in the wider process.

Tender confidentiality clauses and the AI plan you choose

Tender documents often carry confidentiality terms, and your evidence bank carries client names, site details and your own rates. Read the instructions for any rule on sharing documents with third parties or on using AI tools, and follow it. Replace client names with project references while you work, and keep your rates and margins in your estimating system rather than in any chat.

The plan matters because a tender passes through the assistant several times: once for the matrix, again for every draft and again for the marking. A personal ChatGPT or Claude plan may use those chats for model training until you turn the switch off in privacy settings, so do that before the first upload or, better, keep tender work on a business plan, where OpenAI and Anthropic leave business content out of training by default. At USD list prices, ChatGPT Business Standard and Claude Team Standard each cost $25 a seat a month, or $20 billed annually, with a minimum of two seats. If a tender's instructions forbid third-party tools outright, no plan setting fixes that, so raise it with the buyer through the clarification route. The questions to ask before buying AI that touches client data are worth running through before you pay.

Reusing last year's answers without carrying over the last client

Old answers are a useful starting point and a trap. They carry the previous client's name, project details that no longer apply, figures that have aged, an accreditation that has since lapsed and a scoring format that was different. Copy from the evidence bank, which you keep current, rather than from the last bid. When a full draft exists, ask the assistant to list every proper noun, number and date in it that does not appear in the current tender documents or the evidence bank, then check each line.

A realistic mistake, as an illustration: a final PDF is reviewed, signed off and uploaded, and an appendix header still names last year's client. Nobody saw it because everyone read the answers and nobody read the headers. The proper-noun listing would have caught it in a minute, which is why that step comes before sign-off and not after. The anonymising tutorial describes a repeatable way to strip identifying details before you paste anything.

The last checks: numbers, names and placeholders

Before submission, run a check that is dull and effective. Ask the assistant to list every number, date and name in the bid and flag any that disagree. An illustrative catch: Question 3 says the team has 14 operatives, while Question 4 says 12 and the price form lists a crew of 10. Each figure may be defensible, but an evaluator who spots three different numbers stops trusting the rest.

• Every question is answered, and every mandatory form is completed and signed.
• Word and page limits are met, with the layout rules followed.
• No [EVIDENCE NEEDED] or square-bracket placeholder remains.
• Project values, years, headcounts and dates agree throughout.
• File names and formats match the portal's instructions.
• The bid is submitted at least a day before the deadline, with proof of submission saved.

Whatever the result, ask for the debrief and compare the buyer's scores with your own marking. Update the evidence bank with what you learned, because the second tender should start from the first one's best answers.

Tender writing questions from small contractors

Can AI write the whole tender response for me?

It can draft each answer, but only from evidence you supply, and a bid built from its general knowledge tends to read like every other response. Use AI to extract the scoring, shape your evidence to each question and mark the result. Facts, prices and commitments must come from you, and a person must stand behind every claim.

Will the buyer know I used AI?

Buyers cannot be sure, but generic phrasing and unsupported claims are what evaluators mark down, whoever wrote them. Read the tender instructions for any rule on AI use or third-party tools and follow it exactly. If a declaration is asked for, make it truthfully and keep a note of how you used the tool.

How long should a method statement answer be?

As long as the limit allows and no longer. Treat the word limit as a budget: spend it on site-specific sequence, resources, risks and checks, and cut introductions and mission statements first. If the limit is a page count, check the font size and layout rules in the instructions before you format anything.

Should I use AI to set my price?

No. Price comes from your estimate, your costs and your risk allowance, and the price form's arithmetic should be checked by an estimator. AI can explain how the price score works and test what a change in price does to your total mark, but a number it generates is not a basis for a commitment.

Further reads

• Groundworks Tenders: Build a Tender Pack and Method Statements With AI — A tender pack and method statements for a groundworks firm
• Can AI Do Construction Bid Levelling? What It Catches and Misses — Compare subcontractor quotes before you price the tender
• AI Tools for Construction Estimating: Bid Checks After the Takeoff — Bid checks after the takeoff, before the price goes in
• AI for Construction Contractors: Bid Invitations and Follow-Up — Handling bid invitations and following up afterwards
• Architecture RFP Response: Reuse Past Proposals With AI, Without Copying — Reusing past proposals without copying them blindly
• Can AI Draft a Construction Phase Plan? What You Must Still Check — Another document AI can draft but a person must check
• Can AI Write a Catering Tender Response? What to Keep Confidential — Bid-writing steps for caterers: requirement matrix, pricing schedule and compliance answers drafted with AI, plus what to keep out of consumer tools.
• AI Marketing Tools for Small Businesses: A Shopfitter's Case Study Pages — A step-by-step method for shopfitters: project paperwork and photos in, a checked one-page case study out, with client sign-off before it goes live.
• Bill of Quantities Template: A Worked Example Checked With AI — A worked bill of quantities with Excel check formulas and AI prompts for codes, descriptions, units and totals, with measurement left to the QS.
• ChatGPT for General Contractors: 10 Prompts for Bids, Subs and Updates — Ten prompts for bid screening, scope sheets, sub chasing and client updates, each with a sample output and a check before sending.
• Home Automation Tenders: Draft Responses With AI and Check Every Spec — Build a requirement matrix and compliance table with AI, write spec wording from datasheets, and have an engineer verify every claim before you submit.
• How to Bid Drywall Jobs: Plan Review and Scope Checks With AI — A commercial drywall bid from spec to exclusions: wall-type takeoff, fire-rated assemblies, finish levels, alternates, and where AI helps or a person must read.
• How to Bid Flooring Jobs: Commercial Takeoffs and Tenders With AI — A commercial flooring bid from finishes schedule to tender questionnaire: room-by-room takeoff, waste and prep, an AI answer bank and hand-checked counts.
• Masonry Bidding: Review Plans and Specs for Scope Gaps With AI — A bid-review routine for masons: AI tables each requirement, flags who supplies and who installs, and turns gaps into RFIs, exclusions and clarifications.
• Commercial Plumbing Tenders: Can a Small Firm Use AI to Answer? — How a small plumbing firm can draft a commercial tender with AI: method statement, programme, exclusions, insurance wording and pre-submission checks.
• ChatGPT for Construction: Eight Office Jobs With Worked Examples — Eight office jobs a small construction firm can give ChatGPT, each with a worked example, a check step and the jobs to keep human.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: published tender evaluation notices and procurement guidance on price scoring formulas (several common methods compared); OpenAI business data page; Anthropic privacy centre; ChatGPT Business and Claude Team list prices as verified on the vendors' pricing pages in late October 2026.