the best file format for ai proposals isn't pdf or word.

elisabeth hitz · july 14, 2026 · 5 min read

for feeding a proposal to ai, the best format is markdown, not pdf or word, because those two carry visual layout the model has to decode before it can read a word. every listicle that ranks proposal formats is answering a different question than the one you are asking. they rank formats for handing a document to a human. you want to hand it to claude. those are opposite jobs, and the answer flips.

this trips up smart people because the advice is not wrong, it is just aimed at the wrong step. "export as pdf" is great advice for sending a proposal to a client. it is bad advice for getting ai to analyze, rewrite, or extract from that same proposal. once you see the two jobs as separate, the format choice gets obvious.

why do the listicles say pdf or word?

because they are ranking formats for delivery to people. a pdf looks identical everywhere and cannot be easily edited, which is perfect for a final document going to a client. a word file is easy to redline and comment on, which is perfect for collaboration. both are excellent at what they are for: presenting to a human eye.

but presenting to a human and feeding a model are not the same task. everything that makes a pdf good for delivery, the fixed layout, the exact positioning, the visual polish, is dead weight to a model. it stores where each glyph sits on the page instead of what the page means, so the model has to reverse-engineer the structure before it can start. word is better because it keeps some structure, but it still wraps your content in layout machinery and styling the model has to wade through. neither was built to be read by software that only cares about meaning.

why markdown wins for ai input

markdown carries structure as plain text and nothing else. a heading is marked as a heading. a list is marked as a list. a table is real rows and columns. there is no font, no page geometry, no positioning to decode. the model reads the structure directly, the same way it reads the enormous amount of markdown it was trained on. this is the full argument behind markdown vs pdf for claude, and it applies double to proposals because proposals are where the stakes live: pricing tables.

a pricing table in a pdf is exactly the kind of thing that gets mangled when claude reconstructs it. the model guesses which number goes in which cell, and a small slip turns into a wrong price in front of a client. in markdown that table has explicit cells, so there is nothing to guess. for a document whose whole point is accurate numbers and clear scope, that is not a small difference.

the rule that keeps it simple

you do not have to choose one format forever. you have to use the right one at the right step. here is the whole workflow.

  • keep the master in clean text. write and store your proposal in markdown, a google doc, or plain text, not in a pdf. this is your working copy and the one you feed to ai.
  • do all ai work against the markdown. when you want claude to tighten the language, check the scope, adapt it for a new client, or pull out the numbers, point it at the clean text version. this is where the accuracy and the token savings come from.
  • export to pdf last, for the client only. once the content is right, export the final, polished pdf and send that to the human. the pdf is the delivery, not the input. never do the reverse and try to feed the exported pdf back into ai.

that ordering is the whole trick. text in, pdf out. the moment you invert it and treat the client-facing pdf as your source of truth for ai work, you inherit every decoding error the format can produce. keep the clean version as the master and the pdf as a snapshot you generate at the end.

this is also just good input hygiene, which is most of what separates ai that works from ai that does not. if your results have been inconsistent across the board, it is usually the same story, and it is worth reading why your ai is not working for the wider pattern. format is one lever, but it is a big one for anything with a table in it.

faq

should i send my client a markdown file then?

no. clients want a polished pdf, so deliver the pdf. markdown is for the ai step, not the human step. the point is to use each format for the job it is good at, not to pick one for everything.

what if my proposal template only exists as a pdf?

convert it to markdown once with a free tool or by asking claude to convert it, then keep that markdown as your reusable master. you pay the conversion cost a single time instead of fighting the pdf on every edit.

is word really that much worse than markdown?

word is better than pdf because it keeps real structure, but it still wraps content in styling and layout the model has to work through. markdown is leaner and closer to what the model was trained on, so it is the cleaner input of the two.

join the closer method

this is the boring, high-leverage stuff we drill inside the self-paced closer method community. feeding ai clean is lesson one.

come get the rest