When every bid reads the same: designing tenders that still tell suppliers apart

Evaluators are noticing it across the sector. Responses are fluent, well-structured, fully compliant and increasingly interchangeable. Same requirement, same tools, same output. The result is bunching at the top of the score range, evaluators struggling to articulate why one supplier scored above another, and moderation notes that don't hold up when someone asks for reasons. That last point is the real risk: if you can't explain the differential, you can't defend the award.

Banning AI isn't the answer, it's not practical or enforceable. The better response is to design a process that assesses delivery capability rather than writing ability which is precisely what the competitive flexible procedure at section 20(2)(b) exists to let you do.

Ask for evidence, not description. A method statement can be generated in seconds; a named contract, a real referee, an actual retention figure, a genuine escalation and what it cost cannot. Require specifics that can be checked.

Score the specific, not the fluent. Section 23(2) requires award criteria to be clear, measurable and specific. Use that: write scoring guidance that caps a well-written but unevidenced answer, and say so up front so the market responds accordingly.

Build in stages AI can't sit in. Interview the named delivery team rather than the bid team. Use scenario exercises, site visits, observed practice. The competitive flexible procedure lets you design these in but they must be set out in the tender notice and associated tender documents before you start.

Cut your word counts. Volume favours generated text. Compression favours people who actually know the contract, then clarify and verify!

AI has helped plenty of small, capable providers clear a writing barrier that never had much to do with how well they deliver. The aim isn't to catch people out. It's to make sure the thing you are scoring is the thing you are buying.