Preparing the technical section of a tender with AI
You’ve been handed two PDFs running to sixty pages between them, you’ve got the section on a subject you half know, and it’s due Thursday. Nobody has time to read them end to end, least of all you.
It’s one of the places where AI helps most, and also where it’s easiest for it to make you look bad. Both are true.
The first job isn’t writing, it’s understanding what’s being asked
The usual mistake is to open the document and start drafting. Before that, two questions save you half a day:
“What exactly does the specification ask for in section X? Quote the text verbatim.” The quoting is the point. If it can’t quote, it hasn’t read it — and you now know that whatever follows is padding.
“What criteria is this section scored against, and how many marks is it worth?” In a well-drafted tender that’s written down, and it completely changes where the effort is worth spending. Writing ten magnificent pages about something worth three marks is a classic.
With that you have the frame. And it’s information that was in the document, not invented.
What AI does well here
Structure. Giving you the skeleton of the bid: which sections, in what order, what goes in each. It’s what saves most time when you don’t know where to start, and it’s the hardest thing to get wrong, because the structure of a technical bid is fairly standard.
Turning requirement into response. The specification says “the contractor shall guarantee end-to-end traceability”; you have to write what you’ll do about it. That jump — from what they ask to what you offer — it does well.
Finding what you’ve missed. “Which requirements in the specification aren’t covered in this draft?” It’s a cross-check you no longer do well at eleven at night.
What it does badly, and here it’s serious
It invents references. Regulations, standards, certifications, tool names. They sound perfect and either don’t exist or don’t say what’s attributed to them. In a technical bid that isn’t a minor slip: if the evaluator spots it, you lose credibility across the whole document.
It promises things for you. It writes with commercial enthusiasm and commits to service levels, timescales and coverage that someone then has to deliver. Anything that’s a number — an SLA, a response time, an availability figure — you check yourself, without exception.
It ignores the document and covers for itself. This is the treacherous one. You give it the specification, it answers confidently, and it turns out it hasn’t read it: it’s describing what such a specification usually contains. You can spot it in ten seconds by asking about a specific detail that only appears in there. We cover it in what happens when you give an AI a PDF, including the page limits almost nobody knows about.
Why more than one opinion helps
A technical bid has no correct answer. It has approaches, and the approach is exactly what scores.
When you ask a single AI for the outline, it gives you one — and it sounds convincing, because it always sounds convincing. You have no way of knowing whether it’s the right one or the first one it thought of.
With several in front of you something useful happens: they normally agree on the underlying structure (that’s the sector standard, trust it) and differ on emphasis: one leans on operations, another on governance, another on tooling. That disagreement is yours to choose from, and choosing with two options in front of you is different from accepting the only one you were given.
That’s what The Judge does with a brief like this: several AIs prepare the response separately and the judge delivers a single assembled piece with the best of each, telling you what it took from where and what it discarded.
An order that works
- Upload the documents and ask for a summary of the section with verbatim quotes.
- Ask about the scoring criteria for that section.
- Ask for the structure, not the text.
- Fill in what’s yours: real capability, experience, names, figures.
- Ask for a cross-check against the specification: which requirement isn’t answered.
- Go through every number and every standard referenced by hand. All of them.
Step 6 isn’t optional. It’s what separates a bid that wins from one that leaves you exposed.
What not to delegate
The part that says what you will do. An AI doesn’t know who’s on your payroll, what you’ve done that’s comparable, or what you can commit to without it costing you money.
That’s exactly what gets scored, and it’s the only thing it can’t write for you.
If tenders are your line of work rather than a one-off, we’ve put the whole method on its own page: AI for public tenders — from summarising the specification to cross-checking between AIs, including what not to upload anywhere.