We use AI. Here is precisely how, and what to say about it.

1. What our AI operator does

1. Monitors published sources: tender portals, funding calls, regulatory announcements.
2. Scores an opportunity against your profile and the published award criteria.
3. Assembles evidence from what you gave us and from public sources, keeping the source and date of each item.
4. Drafts text, tables, criteria maps and document structures.
5. Chases verification responses, keeps the register and the calendar, and writes routine correspondence.

2. What a named human does

1. Reads the whole deliverable before it leaves us. Not a sample of it.
2. Checks every factual claim against its source, and every number against its origin.
3. Decides what to keep, what to cut and what to rewrite, and records that decision.
4. Approves the deliverable and is accountable for it. That person is named to you.
5. Ratifies before any submission, two working days before your deadline.

Nothing produced by the AI operator reaches you, a buyer, a funder or a candidate without a named human having read and approved it.

3. What we never do

1. We never state that AI was not used when it was.
2. We never present AI-generated text as the personal work of a named individual who did not write or adopt it.
3. We never invent a fact, a figure, a reference, a case study or a quotation. If we cannot source it, it does not go in.
4. We never submit anything you have not approved.
5. We never call a credential verified unless the issuing source confirmed it to us.

4. What to disclose, when a buyer or funder asks

Use this. It is accurate for the work we do, and you should adapt it only to make it more accurate, never less.

If the buyer or funder asks for the name of the reviewer at our end, we will give it. If they prohibit third-party assistance altogether, tell us and we will stop. We will not help you conceal it.

5. Where the rules differ

1. Ireland. No general published rule on tenderers using AI was found when we last checked. Individual contracting authorities set their own terms, so the only safe answer is to read the RFT for your competition. We do that and record the date.
2. European Commission and Horizon Europe. You may use AI, and you are fully responsible for anything it produced. You must be able to say which tools were used and to produce the sources behind generated or rewritten text, with citations. Nothing intended to be patented, and no team, financial or IP detail, goes into a standard tool.
3. UKRI research funding. Permitted, with disclosure expected for substantive use — idea generation, data interpretation, literature comparison, code or abstract generation. Language, translation and formatting need no disclosure. An application, or a section of one, must never be generated without human involvement. Breach is treated as research misconduct, which can mean rejection, termination of funding and multi-year bans.
4. UK public procurement. Supplier use of AI in the commercial process is not prohibited and there is no automatic duty to disclose it — the guidance treats it much like engaging a bid writer. Buyers may add optional disclosure questions, and a declaration may be required where AI will be used to deliver the contracted service. Some authorities are advised to ask anyway. We check the current position for your competition rather than relying on a general rule.
5. Australia and the United States. We check the buyer’s own conditions each time. We do not assume.

We record the check and the date on your file, so if you are asked two years later what the position was, there is an answer.

The people evaluating your bid are under stricter rules than you are

It is worth knowing, because it shapes what a good bid looks like. Evaluators at the European Research Council have been told plainly, as of 24 March 2026, that they may not use AI to summarise proposals, assess merit, or draft evaluations, and that uploading a proposal or any part of it to an external AI system is prohibited. They may use it to tidy the language of their own report. Nothing else. UKRI assessors are under the same shape of rule.

So your bid is read by a person, start to finish. That is an argument for writing it for a person: specific, checkable, and carrying the evidence only you have.

What we will not put into a tool, ever

This is the part that costs people the most and gets discussed the least. What goes into a tool matters more than what comes out.

• Anything you intend to patent. Entering it into a public tool can count as public disclosure and destroy novelty. That loss is permanent.
• Trade secrets. The protection depends on the information having been kept secret.
• Privileged legal material. Sharing it with a third-party system can put the privilege at risk.
• Other people’s personal data, without a lawful basis and, where needed, consent.
• Your pricing. In a competitive tender this is both a confidentiality breach and commercial self-harm.

Where we need help with structure or language on something sensitive, we describe the shape of the problem rather than the content of it.

One risk nobody warns bidders about

When several bidders use similar tools with similar prompts, the answers come back similar or identical. To a buyer that reads as a market with nothing to choose between — or worse, as collusion. It is the strongest practical argument for using these tools on structure and language only, and putting the specific, un-generatable evidence in by hand. Which is what we do.

6. Your data and our AI use

1. We do not use your confidential information or personal data you give us to train third-party models.
2. We keep a record of which deliverables had AI involvement and who reviewed them.
3. You can ask for that record at any time.

7. If you would rather we did not

Tell us. We will quote for a fully manual engagement instead. It costs more and takes longer, and we will say so honestly rather than quietly charging you the same.