Using AI in Bid Writing Without Getting Disqualified

​Almost every bid team in Africa is now using AI in some form. Proposal managers paste in RFP clauses, consultants ask a chatbot to tighten a methodology, and business development teams use it to translate a submission into French or Portuguese. The question is no longer whether to use it. It is how to use it without handing an evaluator a reason to mark you down or throw you out.

​The risk is not that AI makes bad writers out of good ones. The risk is that it makes confident, fluent, wrong submissions very quickly, and that it tempts people to upload material they have no right to share. Both problems are avoidable once you know where the lines are.

​What the Rules Actually Say

​Start with an honest picture. Procurement rules on AI are still catching up, and they differ by buyer.

​In the UK, guidance encourages contracting authorities to ask suppliers to disclose their use of AI in creating a tender, and to put proportionate controls in place so bidders do not feed confidential authority information into AI systems. The same guidance notes that if suppliers use AI to create tender responses, buyers may carry out additional due diligence on whether the supplier can actually deliver. One legal commentary on that guidance makes the practical point plainly: using AI as a bidder is not prohibited, but it should only be applied where the buying authority can understand and manage the risks. (www.burges salmon +2)

​In the US, a Commerce Department framework published in 2026 describes buyers who want to know whether proposals were drafted, generated or otherwise prepared with AI, plus the human review and validation applied to that content. It also raises the possibility that the government may use AI detection tools at its discretion. Individual tenders are already writing this into their terms. One UK tender annex, for example, permits bidders to use AI to help draft responses, but requires them to manually agree to an AI-use policy. (www.commerce.gov +2)

​For African bidders, the key point is this: I did not find a blanket ban on AI in the standard AfDB or World Bank procurement documents. That does not mean you are free to do anything. These funders have firm rules on fraud, corruption and misrepresentation, and a submission containing false statements is treated the same way whether a person or a machine wrote the sentence. Before each bid, read the instructions to bidders for any AI clause, and if the document is silent and you plan to rely heavily on AI, consider sending a short clarification question. Silence is not permission to skip disclosure if disclosure is asked for anywhere in the forms.

​Six Ways AI Gets Bidders Disqualified or Marked Down

​1. Invented or inflated experience

​This is the most dangerous failure. AI tools are built to produce plausible text, and plausible is not the same as true. Ask a tool to "strengthen" a project description and it may quietly add a contract value, a client name, a completion date or a role that was never yours. A reference project that says you were prime contractor on a 150 MW scheme, when you were a subcontractor on one package, is a material misstatement. In a prequalification, where experience is judged narrowly, one false claim can end the bid and, on funder-financed work, can lead to sanctions that follow your firm into future tenders.

​The rule: AI may rephrase facts you supply. It may never supply the facts.

​2. Inflated CVs and key personnel

​Evaluators scrutinise CVs in detail, and AI makes it easy to over-polish them. Typical problems include a junior role rewritten to sound like a lead role, years of experience that don't add up across the document, and responsibilities that sound impressive but don't match the project's scale. Many forms also require the named expert to certify that the CV is accurate and that they are available. If a tool embellished a CV and the expert signed it without reading it closely, the false statement is the firm's problem and the expert's problem.

​The rule: every CV goes back to the person it describes for line-by-line approval before it is submitted.

​3. Confidentiality and data breaches

​Pasting a full RFP, a client's draft terms of reference, staff CVs or your pricing build-up into a public AI tool is a data-handling decision, not just a drafting shortcut. One legal analysis warns that bidders should be wary of uploading their tender responses to AI, since these often contain personal data in CVs and commercially sensitive pricing. Another notes that buyers may not be able to verify what happens to information once it is entered into an AI system, including what is retained, reused or used to train models. (trowers) (jdsupra)

​Some tender documents are marked confidential or come with non-disclosure terms. Staff CVs contain personal data, which may be covered by data protection law in the expert's country. Pricing is the crown jewel. Treat all three as off limits for any tool you have not approved for confidential use.

​4. Generic, non-responsive writing

​Evaluators score against stated criteria. AI tends to produce smooth, general text that talks about "robust stakeholder engagement" and "innovative methodologies" without ever naming the site, the river, the regulator or the risk. On a quality-and-cost-based evaluation, generic prose scores poorly because it does not show that you understood this particular assignment.

​Evaluators are also getting better at spotting the signs. One bid-writing firm lists tell-tale giveaways, including residual prompts or placeholders accidentally left inside the narrative of an AI-generated draft. A line like "Insert client name here" or "As an AI language model" in a final submission is the sort of mistake that signals a rushed, unchecked bid. (bwlgroup)

​5. Failing to disclose where disclosure is required

​If a tender asks whether AI was used, the answer has to be accurate. Some buyers are adding a direct question along the lines of whether the bidder has used AI or machine learning tools, including large language models, to assist in any part of the submission. Answering "no" when your team drafted half the methodology with a chatbot is a misrepresentation, and one that is easy to avoid by simply being truthful. (burges-salmon)

​6. Missing the mechanical requirements

​AI is poor at the unglamorous parts of compliance that decide most disqualifications: certified translations, notarised documents, signed declarations, the exact currency for contract values, mandatory forms in the prescribed order, and page or word limits. A fluent draft that ignores the required schedule format is still non-compliant. Evaluation committees are not obliged to chase missing documents, and no amount of polished prose rescues a missing audited account.

​Where AI Genuinely Helps

​None of this is an argument against AI. Used well, it removes hours of low-value work so your experts can spend time on the parts that win.

• ​Reading the tender: AI is excellent at turning a long RFP into a compliance matrix: every mandatory requirement, evaluation criterion, deadline, form and submission rule in one table. It is also good at flagging contradictions between the instructions, the terms of reference and the forms, which is the raw material for sharp clarification questions.
• ​First drafts from your own material: Give it your verified project descriptions, your team's actual approach notes and the evaluation criteria, and ask for a structured first draft. You are using it to organise and phrase things you already know.
• ​Editing and translation: For bids in English, French, Portuguese or Arabic, AI can improve clarity and consistency. Have a fluent human check technical terms, because machine translation can quietly change meaning in contractual language.
• ​Consistency checks: Ask it to compare the executive summary, methodology, work plan and team structure and list any mismatches: a role named differently in two places, a task in the methodology that has no matching deliverable, or a team member who appears in the staffing table but not in the organisation chart.
• ​Red-team review: Paste in your draft answer (without confidential material) alongside the evaluation criteria and ask where an evaluator could mark you down. It will not replace a senior reviewer, but it catches the obvious gaps early.

​A Safe Workflow, Step by Step

1. ​Decide your policy before the bid starts. Write down which tools are approved, what data may be entered, and who signs off. A one-page policy beats a dozen private habits across the team.
2. ​Read the tender for AI clauses. Check the instructions to bidders, the forms and any declarations. If a disclosure question exists, plan your honest answer now.
3. ​Build a verified fact base. Before any drafting, assemble approved reference projects, CVs, financial data and company descriptions, each confirmed by a named owner. This is the single most useful step, because AI works from what you give it. If your references and CVs are scattered and inconsistent, even a good tool produces a weak, unreliable proposal.
4. ​Use AI for structure, not substance. Generate the compliance matrix and outline first. Then draft sections from your fact base, with an instruction like "use only the facts below, and mark anything you are unsure about."
5. ​Strip confidential data before prompting. Replace client names, prices and personal details with placeholders where the tool is not approved for confidential use, and reinsert the real details in your own document afterwards.
6. ​Verify every claim. Someone who was on the project reads each experience statement. Every number, date, standard and place name is checked against source documents. Check cited standards and regulations in particular, because AI can cite ones that don't exist or misstate what they require.
7. ​Rewrite for specificity. Add the site conditions, the local regulator, the seasonal constraints and the risks that only someone who has done this work would name. This is where your expertise shows, and it is exactly what generic AI text lacks.
8. ​Run a compliance pass by hand. Walk the matrix line by line against the final documents: forms, signatures, certifications, currencies, page limits, file naming and upload instructions.
9. ​Clean the file. Search for leftover placeholders, bracketed prompts, odd formatting, mixed spellings (British versus American) and inconsistent tenses. These are cheap to fix and costly to miss.
10. ​Disclose honestly and keep a record. If asked, state which tools were used, for what purpose, and what human review followed. Keep an internal log so you can answer questions later.

​A Simple Disclosure Statement

​If a tender asks you to declare AI use, plain language works best. Something along these lines is accurate and calm:

​"We used generative AI tools to assist with structuring, editing and translating parts of this submission. All factual content, including experience, personnel and financial information, was supplied from our own records and reviewed and verified by named members of our team before submission. No confidential tender information or personal data was entered into unapproved tools."

​Adjust it so that every sentence is true for your bid.

​A Pre-Submission Checklist

• ​[ ] Has every reference project been checked against the contract or completion certificate?
• ​[ ] Has each key expert read and approved their own CV, and confirmed availability?
• ​[ ] Were any confidential documents, personal data or prices entered into a tool that is not approved for them?
• ​[ ] Does the submission answer this client's specific problem, or could it be sent to any client?
• ​[ ] Are there any leftover prompts, placeholders or invented citations?
• ​[ ] Have all mandatory forms, certified documents, signatures and currency requirements been met?
• ​[ ] If the tender asks about AI, is our answer accurate and consistent with what we actually did?

​The Bigger Picture

​The firms that win with AI will not be the ones that generate the most text. They will be the ones that keep an accurate, well-organised record of what they have done, use AI to move faster on structure and polish, and keep experienced people responsible for truth and judgement. Buyers, for their part, are working out how to assess a flood of fluent proposals, and the safest response for a bidder is to be clearly accurate, clearly specific and clearly honest about process.

​AI can shorten the path to a good submission. It cannot take responsibility for it. That stays with you.

​How is your firm handling AI in bids: banned, tolerated or built into the workflow? Share what's working, and what almost went wrong, in the comments.