Using AI to Format Your CV for World Bank Tenders: The Dos, the Don'ts, and the Debarment Risk Nobody Warns You About

​Every consulting firm bidding on World Bank, AfDB, or AFD-financed work now has someone on the proposal team quietly feeding expert CVs into ChatGPT or Claude before submission. It tightens language, fixes formatting, translates a French CV into English, or reshapes a CV into the rigid Key Expert format that most Terms of Reference (TOR) demand.

​Used this way, AI is simply a faster typist. Used carelessly, it becomes the fastest route to a fraudulent practice finding and a multi-year debarment that follows a consultant or a firm across every multilateral development bank (MDB) in the world.

​This article sets out exactly where that line sits, grounded in how the World Bank's own sanctions system actually treats CV misrepresentation, and gives a practical workflow for using AI on tender CVs without crossing it.

​The Rule That Predates AI (And Still Governs Everything)

​The World Bank's Integrity Vice Presidency is very clear on what constitutes misconduct:

​Fraudulent Practice: Any act or omission, including a misrepresentation, that knowingly or recklessly misleads a party to obtain a financial or other benefit.

​That definition does not care whether the misleading sentence was typed by a human or generated by a language model. The tool used to produce a false claim is irrelevant to whether the claim is false. What AI changes is not the rule, but the speed and plausibility with which a false claim can now be produced.

​The Bank's own published guidance gives a concrete illustration: a consulting company claimed the project experience of its individual consultants as the firm's own and separately exaggerated the value of past projects. Both were treated as fraudulent practices, not aggressive marketing. The exact same logic applies to an individual expert's CV inside a technical proposal.

​The Precedent: Real Sanctions for CV Misrepresentation

​These cases predate generative AI, which highlights how seriously the Bank treats this misconduct even without an "AI multiplier."

• ​False Accounting Certificates (15-Month Debarment): Submitted by an individual consultant in Uganda.
• ​Falsified Work Experience (3-Year Debarment): Exaggerated experience on a contract application.
• ​Forged Professional Certifications (3-Year Debarment): A firm falsely claimed two staff held TOR-required credentials.
• ​Fabricated Past Projects (5.5-Year Debarment): A firm submitted forged documents claiming non-existent past experience.
• ​False Availability Claim by a Joint Venture (15-Month Debarment): Misrepresented the availability of a named key specialist.
• ​False Availability Claim by a Consultancy (14-Year Debarment): Misrepresented the availability of key staff for a technical assistance contract.

​Two critical takeaways from these cases:

1. ​Misrepresentation almost always centers on a Key Expert (qualifications, certifications, experience, or availability).
2. ​Debarment by the World Bank triggers a mutual enforcement agreement. A sanction on one project simultaneously closes the door to AfDB, AFD, IDB, and ADB-financed work.

​Where AI Specifically Raises the Risk

​None of the cases above involved AI. However, AI introduces failure modes that map perfectly onto the conduct the Bank has already sanctioned repeatedly.

• ​Fabricated specifics presented with false confidence: Language models are fluent, not factual. If asked to "strengthen" a CV, a model will often hallucinate plausible-sounding project names, clients, contract values, or date ranges.
• ​Inherited claims from templates: If you ask AI to "reformat this CV to match our standard template," it may carry over leftover text from the template's previous expert (degrees, certifications, or roles) without anyone noticing the contamination.
• ​Inflated seniority through tone: Asking AI to make a CV "sound more senior" without supplying new facts almost always results in upgraded job titles and implied leadership roles that do not reflect reality.
• ​Unsupported certificate claims: If you prompt AI to write that an expert "holds a PMP certification" simply because the TOR demands it, you have manufactured a false fact pattern. AI cannot verify a certificate.
• ​Availability statements detached from reality: AI easily generates boilerplate text like "the expert confirms availability." If the expert is actually committed elsewhere, generating this sentence without verification is precisely what caused multi-year debarments in the past.

​The Dos and Don'ts of AI CV Formatting

​To stay compliant and competitive, adhere to these clear boundaries when generating proposal materials.

​The Dos ✅

• ​Use AI only on material you supply. The safest use of AI is restructuring verified content (CVs, certificates, references) into the required TOR format.
• ​Trace every claim to a document. Run a one-line check before submission: Where is the certificate, transcript, or contract that proves this sentence? If the answer is "AI wrote that," delete it.
• ​Ask AI to flag gaps, not fill them. Prompt the model to "list every TOR requirement this CV does not currently address, without adding new content."
• ​Require human sign-off. Have a designated team member—separate from the person who ran the AI—verify facts against source documents.
• ​Use AI for translation and language polishing. Translating a French CV to English or tightening grammar is low-risk, provided you check the output against the original for meaning drift (especially regarding dates and job titles).

​The Don'ts ❌

• ​Do not prompt AI to "qualify" an expert. Avoid instructions like, "this expert needs ten years in dam safety, adjust the CV." This forces the model to invent facts.
• ​Do not reuse contaminated templates. Ensure previous draft sections, comments, or hidden text from other experts are completely deleted before the AI rewrite.
• ​Do not let AI invent contract values or durations. If the real project details are forgotten, find the original contract. Do not let the model guess a confident-sounding number.
• ​Do not submit unverified boilerplate availability statements. Never let AI draft a commitment clause without getting written confirmation from the named expert first.
• ​Do not assume a well-formatted CV is a low-risk CV. Fluency is not a defense. The Sanctions Board evaluates whether the underlying claims are true, not how professional they look.

​A Practical, Safe AI Workflow

​If you want to leverage AI for efficiency without risking debarment, implement this step-by-step process:

1. ​Start with the Source: Gather the person's actual documents—degree certificates, professional registrations, signed reference letters, and prior contracts.
2. ​Prompt for Structure: Feed those verified documents into the AI tool. Ask it to restructure only that material into the TOR's required format and word limit.
3. ​Run a Gap Analysis: Ask the AI separately to identify any TOR requirements not evidenced in the source documents. Treat these as flags for the expert to address with a real document, not a rewritten sentence.
4. ​Enforce Independent Verification: Have a second team member check every certification, project reference, and availability statement against the underlying paperwork before locking the CV.
5. ​Secure the Audit Trail: Keep your verification documents on file. If a Bank integrity review ever asks how a claim was substantiated, "we have the certificate" is a complete answer. "An AI tool phrased it that way" is not.

​The Bottom Line: AI is a genuinely useful tool for the unglamorous, time-consuming parts of proposal writing: formatting, translation, consistency, and TOR compliance checks. The danger is not the tool. It is treating a fluency engine as a fact engine, and submitting its output into a system that has spent two decades debarring firms for doing exactly that.