Your AI-written proposals are telling buyers you didn't bother to show up

Before ChatGPT, employers on Freelancer.com paid more for workers whose applications were tailored to the job. After ChatGPT, they stopped. That is the finding of a paper by Galdin at Dartmouth and Silbert at Princeton, and I think every head of sales should read it as if it were about them. A job application and a sales proposal do the same work. Both are text whose real message is "I paid attention to you specifically." Once a machine can fake that attention for free, the buyer stops paying for it.

The same researchers modelled what happens when writing no longer tells you anything about ability. Compared with the pre-LLM world, the top fifth of workers get hired 19% less often and the bottom fifth 14% more often. Swap "workers" for "vendors" and you have my worry about B2B selling right now. The careful sellers lose ground, the lazy ones gain it, and buyers go looking for some other way to tell them apart. Usually that other way is price or incumbency, and neither helps a challenger.

I went to Loopio's numbers expecting good news for AI, because Loopio sells RFP software and has every reason to publish it. Nearly 80% of RFP teams used generative AI in 2025, up 10 percentage points on the year before. Over the same stretch, the average win rate sat at 39%, below the 45% average for 2019 to 2026, and teams reached the shortlist 46% of the time, down from 54% the year before. Loopio does not say AI caused the drop and I cannot prove it did. My read is simpler. When every response in the pile is fluent, complete and correctly formatted, the evaluation committee cannot rank them on the writing, so the writing stops winning anything.

The people reading your proposals are desk workers, and desk workers are sick of this stuff. A BetterUp and Stanford survey of 1,004 of them found 40% believe they received "workslop" in the past month, and they reckon 15.4% of what lands on their desk is polished but useless. Managers get more of it, 54% against 38.5% of individual contributors. The procurement lead and the VP signing off your deal are the managers in that statistic. Colin Breck, in the essay that set me off on this, cites a developer survey in which 78% stop reading when they suspect AI authorship and 71% avoid that author afterwards. In sales, an author the buyer avoids is a dead account.

Breck's sharpest point is about context. Whoever prompted the model has the CRM notes, the call transcripts and the product docs in their head, so the output reads fine to them. The buyer has none of that and gets an exhaustive, context-free document they must read line by line to find the one paragraph about their problem. Breck used AI heavily on his own academic paper, to check facts against source code, fix citations and catch errors, and let it write only the abstract.

The defence of AI proposals is real. Zhang and Gosline found people rated AI-generated ads higher in quality than human ones, and another study found readers preferred AI-written answers when the source was hidden. Loopio says its top performers submit 180 bids a year by putting efficiency ahead of effort, and that 71% of RFP software users are satisfied with their process against 51% of non-users. I accept all of it. But the satisfaction figure measures the vendor's team, and buyers are the ones who sign. The hidden-source condition no longer exists in procurement, because buyers now assume AI. Schilke and Reimann ran 13 experiments showing that people who disclose AI use are trusted less, and those exposed by someone else lose even more. Another study found creators labelled as AI users were judged less qualified and less effortful even when readers rated the text's quality the same. Effort is precisely what a proposal is supposed to prove. If you are a volume shop firing off 180 bids, fine. If you are a mid-market software company chasing a few dozen deals a year, the shortlist is the whole game.

And the documents matter more than ever: Loopio says RFPs influenced more company revenue in 2025 than at any point in the past five years. More money is riding on paperwork that persuades less.

So split the work. Let the model fill the compliance matrix, draft security questionnaire answers from your approved library, chase citations and check every claim against the product docs. Make the account lead write the executive summary and the "why us" section personally, with the client's own words from discovery calls in it. Bryan Cantrill now requires public writing at Oxide to pass the Pangram detector as human-authored, and any buyer can run the same tool on your bid.

Try this before your next pitch. Pull your last five proposals, delete the client names and hand them to a colleague who was not on the deals. If they cannot match each document to its client, your buyers could not tell either, and you should rewrite the next executive summary by hand before it goes out.

Prompted by I Don’t Want to Read What You Didn’t Write, Colin Breck.