Build the pack once as a library: a company profile, one-page CVs for key people, a plant list, insurance and policy documents, and a method statement for each core activity such as drainage and concrete bases. Then let AI assemble what each tender asks for. A competent person must review and sign every method statement before it is issued.
Most groundworks tenders ask the same few things in different clothes: who you are, who will do the work, what you will do it with, how you will do it safely, and what you have done before. The slow part is finding, updating and reformatting the same material every time. AI is good at that reformatting. It is poor at knowing your site, which is why the competent person's review is not optional.
What a reusable groundworks pack contains
Treat the pack as nine documents, each with a single owner and a refresh date. The table shows what each holds, what AI does with it, and how often it changes.
| Document | What it holds | What AI does | Refresh |
|---|---|---|---|
| Company profile | Short and long versions: services, size, plant, typical contract values | Rewrites to a word limit from your facts | Yearly |
| Key-person CVs | One page each, approved by the person | Reformats to one template, flags gaps | On any change |
| Plant list | Each machine, owned or hired, inspection dates | Cross-checks against method statements | Quarterly |
| Certificates summary | Insurance, accreditations, expiry dates | Builds a one-page table from the certificates | On renewal |
| Policies index | Titles, versions and owners of safety, quality and environmental policies | Nothing: directors own the policies | Yearly |
| Method statements | One per core activity, in a standard layout | Drafts, then tailors to a site | After each job review |
| Case studies | Three finished jobs with facts and a photo | Shapes them to the question asked | After each win |
| Programme template | Standard sequence and durations by activity | Adapts the template to the tender's dates | Yearly |
| Compliance matrix | Built per tender: each question, weight, limit, source document | Extracts it from the invitation | Every tender |
Notice the policies row. Policies are signed by directors, so AI does not write them. It only needs the index, so it can tell you which one answers a question.
Set up a private library the assistant can read
Start with a folder, because an assistant is only as good as the files you give it and the names on them. Use one folder per document type, put the version and date in every file name, and keep a consent log for CVs.
Tender library
01 Company profile-short-v3.docx profile-long-v3.docx
02 People cv-AB-v2.docx cv-CD-v4.docx consent-log.xlsx
03 Plant plant-list-2026-10.xlsx
04 Certificates insurance-2026.pdf accreditations-2026.pdf
05 Method stmts ms-open-trench-drainage-v4.docx ms-concrete-base-v2.docx
06 Case studies cs-estate-drainage.docx cs-hardstanding.docx
07 Tenders one folder per tender, named by client and dateThen copy the current versions into a project in ChatGPT or Claude, so the assistant keeps the library as standing context. Count the files first. Two profile versions, five CVs, the plant list, two certificate files, three method statements, three case studies, the policies index and the programme template come to 18. As of November 2026 that rules out a free ChatGPT project, which holds five files, fits the 25 allowed on Go and Plus, and leaves room for a live tender's documents within the 40 on Pro, Business and Enterprise. Even Claude's free plan gives you five projects, and each project file can be up to 30 MB. Sharing differs: a ChatGPT project owner can invite collaborators on any plan, with a cap that rises with the plan, while Claude keeps project sharing to its Team and Enterprise plans. Both vendors change these limits, so check their pages before you plan around them.
The CVs carry personal details and most tender documents are confidential, so the plan matters more here than it does for a marketing draft. On ChatGPT Business and Claude Team your content is not used to train models by default. Each charges $25 per Standard seat on monthly billing ($20 on an annual plan) and needs at least two seats, so a firm where only the estimator uses AI still pays for two. If you are on a personal plan, switch model training off in the account's privacy settings before anything is uploaded, and strip home addresses and dates of birth from every CV whichever plan you use.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowThe company profile and case studies: facts first
Every tender profile starts the same way, and so does the temptation to fill it with adjectives. Compare these two openings for the same nine-person firm (illustrative).
BEFORE
We are a leading groundworks contractor delivering exceptional
quality and outstanding service on projects of all sizes.
AFTER
We are a nine-person groundworks contractor. We excavate, lay
drainage, build concrete bases and lay hardstanding for housebuilders
and small commercial clients. We own seven items of plant, hold the
accreditations listed on page 4, and finished 31 contracts in the last
three years, the largest worth [value].The second version works because every clause can be checked, and the checking is the point. Write a fact sheet before you prompt: staff numbers, plant owned, typical contract sizes, contracts completed, accreditations exactly as printed, insurance levels from the certificates. Then ask for the profile at three lengths, 100, 250 and 500 words, using only those facts and marking anything missing as [MISSING]. Case studies work the same way: three finished jobs, each with the client type, the scope, the value band, one problem and how it was solved, and a photo. Do not name a client without their permission, and do not ask the assistant for a reference, a testimonial or a statistic, because it will invent one. The wider craft of answering scored questions is covered in construction tender writing with AI.
CVs: one page per key person, built from what they approve
CVs are where tenders are won on competence, and where AI is most tempted to flatter. Agree the one-page layout first: role on this contract, years in groundworks, relevant experience in three or four lines, tickets and cards with expiry dates, training with dates, and two short project lines. Ask each person to approve their own CV and record the date in the consent log, then remove home address, date of birth and ID numbers, which a tender does not need unless it asks. The assistant's job is to reformat and to flag gaps, never to fill them.
Reformat the CV below into my one-page template. Keep every fact
exactly as written. Do not add experience, job titles, tickets or
dates. If an expiry date or training date is missing, list it under
GAPS instead of guessing.Here is a before and after from a draft (illustrative). The raw CV line read: "Supervised drainage and base works on a 14-house estate, 2023 to 2024." The assistant returned: "Managed multi-million pound infrastructure programmes, leading large teams through complex drainage delivery." The facts were a supervisor, one estate, 14 houses. The rewrite adds a budget, a job title and a team size that nobody said. Restore the original line, and if you want it to read better, rewrite it yourself: "Supervised drainage and concrete bases on a 14-house estate, 2023 to 2024." Insist that a tender reader can check every sentence against what the person did.
Check the GAPS list each time, because a missing card expiry is a real finding. It is better to discover that an operator's ticket lapsed in March while building the pack than when a client's auditor asks.
The plant list: the document that must match everything else
A plant list is a table, and a method statement that names a machine you do not own is an easy way for a tender to undermine itself. Keep one line per item with quantity, owned or hired, the date of the last thorough inspection, and the operators with current tickets.
| Item | Qty | Owned or hired | Last inspection | Operators with tickets |
|---|---|---|---|---|
| 3-tonne mini excavator | 2 | Owned | [date] | [initials] |
| 8-tonne excavator | 1 | Owned | [date] | [initials] |
| 3-tonne tracked dumper | 1 | Owned | [date] | [initials] |
| Plate compactor | 2 | Owned | [date] | [initials] |
| Pipe laser and level | 1 | Owned | [calibration date] | [initials] |
| Trench support box set | 0 | Hired when needed | n/a | [initials] |
Do not let AI fill the date columns. Use it for the cross-check instead. Paste the plant list and your open-trench drainage statement together and ask: "List every machine or item of equipment named in the method statement and say whether it appears on the plant list." A typical result (illustrative): "The method statement uses a trench support box, a pipe laser and a plate compactor. The plant list shows the laser and the compactor but no trench box owned." That mismatch is real. Either the statement should say the boxes are hired for the job, or the firm needs to own them, and a reviewer reading only the statement would never notice. The checks that feed the plant list are covered in AI for civil construction: plant pre-start checks and service schedules.
Method statements for dig, drain and base
A method statement says how a specific piece of work will be done safely: the sequence, the people, the plant, the hazards and the controls. For groundworks, three statements cover most tenders: open-trench drainage, concrete bases and hardstanding, and reduced dig with spoil handling. Keep them in one standard layout so the review is the same every time. A general introduction to the document is in how to write a method statement for a small build with AI help.
| Section | What it says | Where the facts come from |
|---|---|---|
| 1 Scope and location | What is built, where, and by when | The tender documents |
| 2 Sequence of work | Numbered steps from setting out to reinstatement | Your supervisors |
| 3 Plant and equipment | Each item and who operates it | The plant list |
| 4 People and competence | Supervisor, operators, banksman, excavation inspector, tickets held | CVs and certificates |
| 5 Hazards and controls | Collapse, buried services, plant near people, falls into trenches, dust and noise, manual handling, contaminated ground | Your risk assessment |
| 6 Inspection and testing | Excavation inspections, pipe test, compaction checks | Your procedures |
| 7 Emergency arrangements | Service strike, collapse, injury, contacts and rescue | Site set-up |
| 8 Environment and waste | Silt, spoil, waste carrier, working hours | Environmental policy |
| 9 Review and sign-off | Reviewer, role, date, version | The competent person |
Section 6 is where drafts go vague. Published excavation-safety rules require a competent person to inspect excavations daily, before work starts and as needed during the shift, and again after rain or any other event that raises the hazard. Confirm the rule where you work and write the inspection frequency into the library statement yourself, so it cannot be softened later. Now the prompt.
Draft a method statement for open-trench foul and surface water
drainage in my standard layout, sections 1 to 9. Use only the PLANT
LIST, the SEQUENCE NOTES and the SITE FACTS below. Where a site fact
is missing, write [SITE: what is needed]. Do not state trench depths,
support methods, gradients, test pressures or competence tickets
unless they appear in my notes. Put every hazard in a table with its
control and the person responsible.A first draft (illustrative) will look plausible and still contain errors a reviewer must catch. Here is an extract.
5 HAZARDS AND CONTROLS
Trench collapse | Trenches up to 1.5 m deep do not need support in
stable ground | Supervisor
Buried services | Services located using ground radar before digging
| Banksman
6 INSPECTION
Excavations are inspected weekly by the supervisor.Four faults. "Trenches up to 1.5 m deep do not need support" is a depth threshold lifted out of context: where a rule mentions a figure like that, it still makes a competent person's examination of the ground the deciding test, and shallow trenches collapse too. Whether support is needed depends on the ground, the depth, water, nearby loads and your own rules, so replace the line with your actual policy and make the competent person decide support before anyone enters a trench. "Ground radar" is not on the plant list, so either add the item or name the equipment you really use for locating services. The services control is assigned to a banksman, but the person who uses the locator needs the training for it. And "inspected weekly" is far weaker than the published rules above. Because one wrong line in a safety document can matter more than a hundred right ones, the draft goes to a reviewer, not straight to the client. Electricians face the same rule, and who signs off AI-drafted RAMS for electrical work shows how that trade handles it.
Read the invitation first: a compliance matrix
Before drafting anything for a live tender, turn the invitation into a table: every question, its weight, its word or page limit, whether it is pass or fail, and which library document answers it. Paste the invitation into the assistant and ask for the table, then count the questions yourself against the document, because an extraction that silently skips a mandatory pass-or-fail question is how bids are rejected before anyone reads them.
| Question | Weight | Limit | Library source |
|---|---|---|---|
| Health and safety approach | 20% | 500 words | Policies index, drainage method statement |
| Method for drainage and bases | 25% | 800 words | Two method statements, tailored |
| Resources and plant | 15% | 300 words | Plant list |
| Key personnel | 15% | One page each | CVs |
| Programme | 15% | One page | Programme template |
| Environment | 10% | 300 words | Environmental policy, waste notes |
This example adds up to 100%, which is itself a check: if your extracted weights total 95 or 110, the assistant misread something. The matrix also tells you what to spend time on. A 25% question deserves a tailored method answer, and a 10% one can be the library paragraph with the site details changed. If a pass-or-fail question has no library source, that is the day-one finding, not the day-before-deadline one. Whether to bid at all is a separate decision, which an AI bid/no-bid screen for construction handles, and the pricing side of the same job is in the groundworks quote template for dig, drain and base lines.
The competent person's review: what is checked and what is signed
The reviewer is someone whose training, experience and knowledge equip them to judge the work, for example the contracts manager or a senior supervisor, not the person who ran the prompt. They read the statement as if they would have to defend it to an inspector, because they might. Give them a checklist so the review is the same every time and the assistant's polish does not hide a gap.
- Site facts. Every [SITE] marker has been replaced with a real fact: ground conditions or the report, services found or drawn, access, working hours, neighbours, depths and levels, start date.
- Plant. Every machine named exists on the plant list for this contract, and hired items are described as hired.
- People. Every role is filled by a named person whose current tickets appear in the CVs.
- Sequence. The steps can actually be done in that order on this site, including where spoil goes and how plant and pedestrians are kept apart.
- Controls. Each hazard has a control that someone on site can carry out, with the inspection frequency written as your rule, not the assistant's.
- Emergency. Contacts, the nearest hospital route and the service-strike procedure are filled in for this site.
- Control of the document. Version, date, reviewer's name and role are filled in, and the library copy remains unchanged.
Budget about an hour per statement as a planning figure. The sign-off block should name the reviewer, their role, the date and the version, and it should be a person's signature, not a line the assistant wrote. After the job, add what happened to the library: a trench that needed different support, a service nobody expected, an access problem. A statement improves most when it records the last site's surprises.
Time saved, and where the saving stops
These hours are planning assumptions for a nine-person firm, to be replaced with your own after the first tender. The first build is the investment. Each later tender is a tailoring job.
| Task | Building the library (once) | Each later tender |
|---|---|---|
| Company profile and case studies | 3 hours | 0.5 hours |
| Five key-person CVs | 6 hours | 0.5 hours |
| Plant list and certificates summary | 5 hours | 0.5 hours |
| Method statements, three core activities, with review | 12 hours | 3 hours to tailor two and review |
| Library set-up and folder rules | 4 hours | 0 |
| Compliance matrix and answers | 0 | 3 hours |
| Final read-through | 0 | 0.5 hours |
| Total | 30 hours | 8 hours |
Suppose the firm bids 12 tenders a year and each used to take 24 hours, which is itself an assumption. That is 288 hours before. With the library it is 30 plus 12 times 8, or 126 hours, a saving of 162 hours. At $45 an hour that is $7,290, and a single extra win would be worth more. If you bid only four tenders a year, the comparison is 96 hours before against 62 after, a gain of 34 hours, so the library still pays but less obviously. Work out your own bid count before you decide how polished the library needs to be.
The saving stops at five places. AI cannot price the job, so the numbers still come from your estimator. It cannot see the ground, so the site facts stay with your supervisors. It cannot give references or testimonials, and anything it produces in that line is invented. It cannot take responsibility for a safety document, which is why the review and signature are human. And it cannot decide whether you should bid, which is a business decision about capacity and risk. If a tool promises to write the whole tender unattended, ask what it does when the invitation changes after clarifications, and who is accountable when it states something untrue.
Questions about AI and groundworks tender packs
Can AI write a method statement I can issue without checking?
No. A method statement describes how real people will do real work on a real site, and someone competent must check the sequence, hazards and controls and take responsibility for them. AI drafts structure and wording quickly, but a competent person reviews it against the site and signs it. Never issue an AI-drafted statement unread, and never let it invent depths, plant or ticket details.
Is it safe to upload CVs and tender documents to an AI assistant?
CVs hold personal data and tender documents are often confidential. Use a business plan where content is not used for training by default, strip what a tender does not need, such as home addresses and dates of birth, and get each person's agreement to their CV being reused. If the tender's terms limit sharing documents, check before you upload anything.
How long does a reusable pack take to build?
It depends on how many documents you already have. For the example firm in this tutorial the first pack takes about 30 hours of office and supervisor time and each later tender about 8, but those are planning assumptions to adapt, not benchmarks. Build it in stages, starting with the profile, the plant list and the two method statements you use most.
Do I need a different method statement for every tender?
You need a site-specific one for every contract, but not from a blank page. Keep activity-level statements, such as drainage in open trench or concrete bases, and tailor each to the site's ground, access, services and programme before issue. That tailoring is where the competent person's review matters most, so allow time for it in every bid.
Further reads
- Roofing RAMS Template: AI Drafts, a Competent Person Signs It Off — The same draft-then-sign-off pattern in another trade.
- AI Tools for Construction Estimating: Bid Checks After the Takeoff — Checks to run on the price before you submit.
- AI for Construction Contractors: Bid Invitations and Follow-Up — Handle bid invitations and follow-up.
- Commercial Plumbing Tenders: Can a Small Firm Use AI to Answer? — How a small firm answers commercial tenders.
- ChatGPT for Civil Engineers: Reports, Method Statements and Emails — Engineers' view of AI-drafted method statements.
- Can AI Do Construction Bid Levelling? What It Catches and Misses — Level subcontractor prices before you submit.
- Can AI Help a Small Business Respond to Tenders and RFPs? — What AI can and cannot do on tenders and RFPs.
- 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.
- 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 Price Excavation Work: Hourly, Per-Load or Fixed, Checked With AI — One driveway dig priced three ways from your own machine costs, stress-tested against slower ground, with a contingency taken from your past overruns.
- Earthworks Estimating: Can AI Calculate Cut and Fill Volumes? — Build a cut and fill sheet with end-area volumes and swell factors, see where AI goes wrong, and check the answer by hand.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI's Projects help article, Anthropic's Projects help article and Claude pricing page (checked November 2026); published excavation-safety rules on inspection frequency. The hours in the time comparison are illustrative planning assumptions.