One of the first things we do in a Moltbot setup is look at the client's existing documentation. Almost every time, the documentation that a human was able to use for years is not usable by the bot. The reason is that humans are very good at filling in gaps. AI is less good at it.
Good documentation for humans often assumes context. "See the onboarding guide for details." "The standard process applies here." "Contact your account manager if unsure." Humans follow these pointers. The bot either follows them in unexpected ways or fails silently.
The fix is not to rewrite everything. The fix is to close the gaps. Every implicit reference becomes explicit. Every "standard" gets defined. Every cross-reference gets inlined or linked to a specific document. Every ambiguity gets resolved.
The second thing to fix is duplication. In most orgs, the same information exists in three places with slightly different wording. If all three are indexed, the bot will sometimes pull the old version. The fix is to designate a canonical source for each topic and deprecate the rest. This is work that should have been done anyway. The bot is just the forcing function.
The third thing is structure. Documents that are one long blob are hard for the bot to use. Documents that are broken into labeled sections with clear headings are much easier. You want each section to be a chunk that makes sense on its own, because that is how retrieval actually works.
The fourth thing is freshness. The bot should know what is current and what is archived. Timestamp everything. Mark deprecated documents as deprecated. When a policy changes, update in place or explicitly mark the old one as "replaced by."
Doing all of this is typically a couple of weeks of content work for a mid-size org. Most clients push back at first and then, after the first round of bot tuning, come back and ask "so what would it take to do the content cleanup." The cleanup is the single biggest lever on bot quality. The model is fixed. The prompts help. The content is what makes it useful.
Our honest advice is to start the content cleanup in parallel with the bot setup. It pays for itself in a couple of months, and the documentation improvements outlast the specific bot deployment.