Your docs are your new sales team

A buyer is choosing between you and two competitors. They don’t open three tabs and compare. They open one chat window and type: which of these should I use for X? The model reads, the model decides, and the model names a winner, before a human ever lands on your site. The whole evaluation happened in a place you weren’t invited to, and the only thing the model had to go on was what you wrote down.

That’s the part that should keep you up at night. The model didn’t read your pricing page or your landing hero. It read your docs.

Your documentation is no longer support content for users who are stuck. It is the first sales call, and the prospect on the other end is a model that will repeat what you wrote to the human who actually buys.

This post is about what changes when the most important reader of your docs is a machine that can’t ask a follow-up question, and what you do about it.

The reader changed, and nobody updated the writing

For twenty years we wrote docs for one reader: a human who already chose you, already signed up, and is now stuck on step four. That reader is forgiving. They’ll scroll. They’ll skim three pages to find the one paragraph. They’ll fill the gaps with a support ticket. Docs written for that reader can afford to be a little scattered, a little assumed, a little “you probably already know what a widget is.”

The naive move is to keep writing for that reader and assume the model is just a faster version of them. It isn’t. The model is a different reader with different failure modes, and writing for the old one quietly fails the new one.

Here is the failure, concretely. The human-who’s-stuck can recover from a vague sentence, they go ask someone. The model can’t. When a model reads “Apollo integrates with the tools your team already uses” and a buyer asks does it connect to my calendar?, the model has nothing to stand on. It either hedges (“it may support calendar integrations”) or, worse, guesses. A vague doc doesn’t produce a vague answer the buyer can correct. It produces a confident answer the buyer can’t.

So the cost of fuzzy writing inverted. It used to be a minor annoyance the reader routed around. Now it’s a missed recommendation the reader never sees you lose.

The reader changed. The model is your top-of-funnel now, and it reads literally.

The naive funnel and the one that’s forming

Walk the old funnel. A prospect searches, lands on a marketing page tuned for the click, gets routed to a demo, talks to a person, and somewhere deep in that journey, usually after they’ve already bought, they open the docs. Docs sat at the bottom of the funnel, the place you go once you’re committed. Marketing was the front door; documentation was the furniture inside.

That order is inverting. The new prospect doesn’t start at your marketing page. They start at a chat window, ask a comparison question, and the model assembles an answer from whatever it can read about you and everyone like you. The first impression isn’t your hero image and its careful adjectives. It’s whatever the model could verify about what you actually do.

Notice what moved. The marketing page didn’t get less important to humans, it got skipped by the reader who narrows the list. And documentation, which used to be the last thing anyone read, became the first thing the decider reads. The furniture became the front door.

This isn’t a reason to stuff keywords into your docs. The model isn’t fooled by that the way an old search engine was. It’s a reason to write docs that survive being read literally by something that will quote them under pressure. Because the thing about a model is that it doesn’t reward the page that sounds best. It rewards the page it can safely repeat.

Write for the reader who can’t ask a follow-up

The whole shift fits in one rule. Your old docs were a conversation, the reader could always come back and ask. Your new docs are a deposition. The model reads once, extracts what it can defend, and answers a stranger’s question with it, with no chance to clarify.

The naive way to write for a model is to write more, more pages, more words, more hedging to cover every case. That backfires. Volume isn’t legibility. A model drowning in qualifiers extracts mush, and mush is what it repeats. The reader who can’t ask a follow-up doesn’t need more words. They need words that resolve.

So three things change in how you write, and each one is the model-reader’s version of an old human-reader habit.

State the capability as a fact, not a vibe. “Apollo can read your calendar and flag the meeting that matters before you walk in” is a sentence a model can lift verbatim and stand behind. “Apollo helps you stay on top of your day” is a sentence the model has to interpret, and interpretation under pressure is where hallucinated answers come from. The human skimmer tolerated the vibe. The model converts the vibe into a guess. Say the concrete thing, because the concrete thing is the only thing that survives the trip to the buyer.

Answer the question the buyer will actually ask, in the words they’ll use. The human-who’s-stuck searched in your vocabulary because they were already inside your product. The buyer asks in their vocabulary, from outside. If your doc explains “proactive briefings” but the buyer asks “can it tell me what’s important each morning,” the model has to bridge that gap, and it might not. Write the bridge yourself. The doc that names the buyer’s problem in the buyer’s language is the doc the model reaches for.

Make every claim checkable, because the model is checking. A capability stated with the shape of how it works, the input, the action, the result, reads as grounded. A capability stated as a slogan reads as marketing the model has learned to discount. The difference isn’t honesty; it’s legibility of honesty. Show the mechanism in one clause and you’ve given the model a reason to trust the claim enough to repeat it.

Same three instincts a great support writer always had, be specific, speak the reader’s language, show your work, pointed at a reader who quotes you to someone you’ll never meet.

The part everyone gets wrong: this is not SEO with a new name

It’s tempting to file all this under “optimize for AI search” and hand it to whoever owned keywords last year. That’s the trap, and it’s worth naming because it’s the comfortable misread.

The old game rewarded the page that ranked. You could rank a thin page with the right structure, the right backlinks, the right phrases in the right tags, and a human would click it, bounce, and you’d still have won the click. The model breaks that. It doesn’t click and bounce. It reads, evaluates, and either recommends you or doesn’t, and a thin page optimized to rank gives it nothing true to recommend. You can’t trick the reader that does the reading itself.

So the work isn’t to make your docs findable. It’s to make them quotable, true enough, specific enough, and complete enough that a careful reader would stake a recommendation on them. Those are different jobs. Findable is a structure problem. Quotable is a substance problem. Suppose, for the sake of argument, two companies write up the same feature: one in eight breezy sentences, one in three sentences that each name an input and a result. The model will quote the second one and skip the first, not because it’s longer, but because there’s something there to hold onto.

The bottleneck moved. It used to be discovery, getting the human to your page. Now it’s verifiability, giving the model something it can defend. The companies that win the new funnel won’t be the ones with the most content. They’ll be the ones whose every claim survives being repeated by a skeptic who read it once.

The turn: docs were always the most honest thing you published

Here’s the part that isn’t about models at all.

Documentation is the only thing a company writes where lying is expensive immediately. Your marketing can promise the moon and the bill comes due slowly, in churn, months later. Your docs can’t. A doc that overstates what the product does generates a support ticket the same afternoon. That’s why, in every company, the docs were quietly the most honest artifact you had, the one place the writing had to match the software or it broke on contact.

The model just made that honesty the front door. The reader who decides whether you make the shortlist is now reading the most truthful thing you publish, and rewarding you in proportion to how clearly you told the truth. That’s a strange and good thing. For once, the incentive points the right way: the company that most plainly describes what it actually does is the company the model most readily recommends. Clarity stopped being a nicety and became distribution.

Which means the work in front of you isn’t a marketing project. It’s an honesty project with a deadline. Go read your own docs the way a model will, flatly, literally, looking for the one true sentence to quote to a buyer who’s deciding right now. If you can’t find it, neither can the model. And the buyer who asked never even knew you were in the running.

---

That’s part of what we’re building toward at Apollo, a company brain where the truest description of what you do isn’t trapped in a doc someone forgot to update, but kept current and legible for every reader, human or model, that comes to decide. Your docs were always your most honest writing. Now they’re your sales team too. The least you can do is let them tell the truth clearly enough to be quoted.