I was at my art group last night, and the conversation somehow drifted from brushwork to bid writing, which, if you know me, won’t surprise you. One of the ladies, a former GP, brought up something that happened to us a while back, and the more I thought about it afterwards, the more I realised it’s exactly the warning I’d give anyone tempted to lean too heavily on AI for their tender responses.
Here’s what happened. A few of us decided to recreate a photograph we’d found on Pinterest. It was a beautiful lilypond, the kind of image that makes you want to grab your watercolours immediately. From a distance, it looked like a genuine photo. At first glance, there was nothing obviously wrong with it at all.
We got two sessions in – about four hours of work, which is roughly what a painting like this should take from start to finish – and it had turned into an absolute nightmare. Nothing made sense. The light and shade were in the wrong places. The lines weren’t right. The perspective and vanishing points didn’t add up. The colours weren’t harmonious in the way they would be in nature. We were all struggling, and none of us could work out why.
It was only after that second session that we discovered the truth: the “photograph” was an AI-generated image. It had never existed. And once you knew that, you couldn’t stop seeing it. The wrongness was everywhere, once you went looking for it.
That’s the thing about painting and drawing: you really have to look at what you’re copying. Artists always copy, in one form or another, but copying means studying, not glancing. Under that kind of forensic examination, the lilypond image fell apart completely.
Had we been working from a real, unadulterated photo, the end result would have felt more realistic, and we’d have finished it in roughly half the time. Instead, we were on it for four sessions, twice as long as it should have taken, because the underlying information was wrong from the start. But you couldn’t tell that just by looking at it. Not at first.
We did eventually produce something that looked okay. But it took far longer than it should have done, and a fair amount of frustrated head-scratching along the way.
Why this matters for bid writing
As a writer, I find exactly the same thing is true of AI and bid writing.
AI can string words together that look like English, structured into something that resembles a proper response. At first glance, it looks like you’ve got an answer. The paragraphs are there, the tone sounds professional, and it’s tempting to think the job’s more or less done.
But under scrutiny – the same kind of scrutiny we ended up giving that lilypond – it falls apart, because it doesn’t actually make sense. Not in the way that matters for a bid, anyway.
The logic doesn’t hold up under questioning. An AI-generated answer might describe an approach in confident, fluent language, but if you ask “why this approach, for this client, in this context?”, the reasoning often isn’t really there. It’s the equivalent of a vanishing point that’s been placed in roughly the right area but doesn’t actually align with anything else in the picture.
The claims are generic rather than specific. AI tends to produce content that could apply to almost any organisation bidding for almost any contract. It talks about “robust processes” and “a track record of excellence” without anchoring any of it to your organisation’s actual experience, people, or delivery model.
The evidence is missing, weak or invented. A polished paragraph about your safety record or social value contribution means nothing to an evaluator if it isn’t backed by real examples, figures, or outcomes. And worse, AI can sometimes produce evidence that sounds plausible but simply isn’t true, which is a risk no bid can afford to take.
Perhaps the biggest issue is that it doesn’t fit the scoring criteria. Evaluators aren’t marking your bid on how nicely it reads in isolation; they’re marking it against a specific question and a specific set of criteria. AI-generated text is very good at producing something that sounds like an answer to a question, without actually answering the question that was asked – or addressing the things the evaluator has been told to look for.
Where it leaves you
With the lilypond, the problem wasn’t that the image looked ridiculous. If it had, we’d never have tried to paint it in the first place. The problem was that it looked convincing enough to trust. AI bid answers can catch people out in a similar way.
A rushed human draft often looks rough. You can see where the gaps are. You know which sections need more evidence, where the subject matter expert hasn’t answered properly, and which parts still need shaping.
An AI-generated draft can look much cleaner than that. It can give you headings, tidy paragraphs and the sort of language we’re used to seeing in bids: collaboration, governance, value for money, continuous improvement, quality assurance. That’s exactly what makes it risky.
The flaws only become visible when someone actually sits down and interrogates the content – checking the logic, checking the evidence, checking it against the question and the scoring criteria, line by line. And if that scrutiny doesn’t happen until the bid review stage, or worse, after submission, you’ve lost time you didn’t have to lose in the first place.
None of this means AI doesn’t have a place in bid writing. It absolutely does – for first drafts, for surfacing content you’d forgotten you had, for taking the blank page away. At Bidding, that’s exactly how we use AI as part of our process: as a starting point that’s then shaped, checked, and strengthened by people who understand the sector, the buyer, and the question being asked.
But a first draft is exactly what it should be treated as – a starting point, not a finished answer. The real value, in painting and in bid writing, comes from the people who know how to look properly: who can spot where the perspective is off, where the evidence is thin, and where an answer that sounds confident is quietly failing to say very much at all.
So the next time an AI-generated bid response lands on your desk looking polished and complete, take a leaf out of our art group’s book. Don’t just glance at it from across the room. Get up close. Look at the light, the shade, the lines, the logic. Because if it was generated rather than genuinely written for this bid, the flaws are there – you just have to look for them.
