Public contracts are some of the steadiest revenue a small business can win: councils, hospitals, universities and agencies pay on time, renew for years, and are often required to spread work across smaller suppliers. Yet most owners never bid. The tender pack lands as six PDFs, 80 pages of conditions and a 40-question response form, and the reaction is the same one we hear about AI itself: "I don't even know where to start."
This is exactly the kind of work AI is good at — reading long documents, extracting requirements, checking compliance and drafting structured answers. Used well, it can cut the time to produce a solid bid from 40 hours to around 15. Used badly, it produces the generic, buzzword-heavy answers that evaluators score a 2 out of 5. This guide shows you the difference, step by step, with the prompts to use.
Why tenders are different from normal proposals
If you have read our guide to using AI for proposal writing, you already know the basics. Public tenders add three things that change how you should use AI:
- Scoring is mechanical. Each answer is marked against published criteria by evaluators working through a scoresheet. An answer that is elegant but misses one stated requirement loses marks. Persuasion matters less than coverage.
- Compliance is pass/fail. Miss a mandatory document, exceed a word limit or leave an insurance certificate out of date, and your bid can be rejected before anyone reads it.
- Evidence beats claims. "We deliver high-quality service" scores nothing. "We responded to 96% of 1,140 support tickets within four working hours in 2025" scores well.
That shifts AI's role. It is less a copywriter and more a tireless bid coordinator: reading everything, building checklists, mapping your evidence to criteria, and flagging gaps.
Step 1: Decide whether to bid at all
The most expensive mistake in tendering is bidding for contracts you were never going to win. Before writing a word, use AI to run a bid/no-bid check. Paste the specification and the evaluation criteria into your assistant (see the confidentiality note below first) and ask:
"You are an experienced public-sector bid manager. Read the attached tender documents. Summarise in a table: contract value and length, submission deadline, mandatory pass/fail requirements (turnover, insurance levels, accreditations, years of experience), evaluation weighting between price and quality, and the three quality criteria with the highest weighting. Then list any requirement a 12-person [your sector] business with €1.4M turnover would struggle to meet."
Score the result against a simple rule of thumb: if you fail any mandatory requirement, stop. If you can only evidence strong answers for less than about 70% of the quality weighting, it is usually a no-bid unless the contract is strategically important. Ten minutes here can save you a week.
Step 2: Build a compliance matrix
Every serious bid team works from a compliance matrix — a spreadsheet listing every "must", "shall" and "will" in the tender, where it is answered, and who owns it. Building one by hand from 80 pages takes most of a day. AI does a first pass in minutes:
"Extract every requirement from these documents into a table with columns: ID, exact requirement text (quoted), source document and page or section, type (mandatory / scored / information), word or page limit if stated, and the response question it belongs to. Do not paraphrase the requirement text. If a requirement is ambiguous, flag it in a final column."
Two rules make this reliable. First, insist on quoted text and page references so you can check every line against the source — AI will occasionally miss or merge requirements, and our guide to preventing AI hallucinations in client work explains why verification is non-negotiable. Second, run the extraction twice, once per document, rather than on the whole pack at once; long-document extraction is more accurate in smaller chunks.
The ambiguity column is a bonus: those flags become your clarification questions. Most tenders allow questions to the buyer before a deadline, and the answers are shared with all bidders — a cheap way to remove guesswork.
Step 3: Build your evidence bank once
The real reason small businesses find tenders exhausting is that they start from zero every time. The fix is an evidence bank: a single document of reusable, verified material that AI can draw on. It should include:
- Three to five short case studies, each with a client type, problem, what you did and a measurable result.
- Your key numbers: turnover, headcount, years trading, retention rate, response times, satisfaction scores.
- Policies buyers routinely ask for: quality, health and safety, data protection, equality, environmental and modern slavery.
- CVs or short bios for the people who would deliver the contract.
- Social value commitments you can genuinely deliver — local hiring, apprenticeships, carbon reduction.
Store this in a project or custom assistant so every bid draws on the same facts. If you have not set one up yet, our guide to building an AI knowledge base walks through it. Once it exists, the second tender takes roughly half the time of the first.
Step 4: Draft answers that score
Now the writing. The secret is to make AI answer the scoresheet, not the question in isolation. For each quality question, give it the question, the scoring criteria, the word limit and the relevant evidence:
"Draft a response to question 4.2 below in no more than 500 words. The evaluators will score it 0–5 against these criteria: [paste]. Structure the answer with one short subheading per criterion, in the same order. Use only facts from the evidence bank provided; where you need a fact that is not there, write [EVIDENCE NEEDED: description] instead of inventing one. Write in plain British English, active voice, no marketing adjectives. End with a two-sentence summary of the benefit to the buyer."
Then use a second prompt to mark your own work before the evaluator does:
"Act as a strict public-sector evaluator. Score this answer 0–5 against each criterion. For anything below 4, quote the weak sentence and say exactly what evidence or detail would raise the score."
Run that loop two or three times. It is the single most valuable habit in AI-assisted bidding, because it forces specificity — the thing that separates a 3 from a 5.
What AI should not write
Keep a human firmly in charge of pricing, commitments you would be contractually held to, and anything about named staff. AI drafts; the person who will deliver the contract signs off every promise.
Confidentiality, disclosure and other rules
Three practical risks are specific to tendering.
Confidential documents. Tender packs often carry confidentiality terms, and your answers contain pricing and commercial detail. Use a business plan of your AI tool where your inputs are not used for model training, and check the tender's conditions before uploading anything. If in doubt, strip names and figures before pasting.
AI disclosure. Some public buyers now ask bidders to declare whether AI was used to prepare their response — UK government guidance issued in 2024 explicitly allows this — and some cap or penalise answers that look machine-generated. Read the instructions to bidders carefully and answer honestly. Our article on whether to tell clients you use AI covers the wider question.
Sameness. If ten bidders ask the same assistant the same question, evaluators see ten near-identical answers. Your evidence bank, your numbers and your named people are what make a response unmistakably yours. AI should structure and polish that material, not replace it.
A realistic time budget
Here is how a typical mid-sized tender (a €250,000, three-year services contract with eight quality questions) breaks down for a small team with an evidence bank in place:
- Bid/no-bid review: 1 hour (previously 3–4)
- Compliance matrix and clarification questions: 2 hours (previously 8)
- Drafting eight quality answers: 6 hours including self-scoring loops (previously 20)
- Pricing, forms and certificates: 4 hours (largely unchanged — this is human work)
- Final review against the matrix: 2 hours (previously 4)
That is around 15 hours instead of 40. The savings compound: the time you free up lets you bid selectively on more of the right contracts rather than scrambling on whatever arrives. Start with one lower-value opportunity, build the evidence bank as you go, and treat the evaluator feedback you receive — win or lose — as the best training data you will ever get.
Where does your business stand on AI?
Take the free 3-minute AI Readiness Quiz and get a personalised score with your next steps.
Take the Free Quiz →