Skip to main content

AI-ready content structuring

AI-ready documentation

This is for SaaS teams whose AI assistant or support bot gives confident, wrong answers, and who suspect the content is part of the problem. Usually it is. Retrieval-augmented generation is only as good as the passages it retrieves, and most documentation was written to be read top to bottom by a person, not cut into pieces and matched against a question. We restructure the content so the right passage gets found and says the right thing on its own.

Price

$2,500 audit; implementation from $6,000

Response

Reply within 1 business day

After delivery

30-day fix window

Who it's for#who-its-for

  • Heads of support who launched a bot on top of the help center and now spend part of every week correcting what it tells customers.
  • Engineering leads building RAG features on product docs, who have tuned the model and the retriever and still get answers from the wrong version or the wrong product.
  • Docs leads asked to make the documentation "AI-ready" without anyone defining what that means in practice.
  • Product and customer experience leaders evaluating AI search or chat vendors, who want the content fixed before paying for a tool that will put its gaps in front of customers.

It is less useful when the assistant fails for reasons outside the content, such as a retriever that ignores metadata or a prompt that invites guessing. The audit says so plainly when that is the case, so you do not pay to rewrite pages that were never the problem.

What you get#what-you-get

The work comes in two stages, priced separately, so you can stop after the first.

The audit tells you why answers go wrong and what to fix first:

  • A retrieval review: we run real customer questions against your current content and trace each wrong answer to its cause, whether that is a missing page, a contradiction, stale content or a passage that makes no sense out of context.
  • A content map of the duplicates, contradictions and version conflicts that confuse retrieval.
  • A structure and metadata proposal covering page types, product and version fields, audience, and a chunking approach that fits your stack.
  • A prioritized fix list you can hand to your own team.

Implementation is the fixes themselves:

  • Pages restructured so each section stands alone: one topic per section, headings that name the task, and no answers that depend on the paragraph above.
  • Duplicates merged, contradictions resolved, and outdated pages retired or clearly marked.
  • Frontmatter metadata added and validated in CI, so every chunk carries its product, version and audience.
  • Terminology cleanup, so the same feature is not called by several different names across the docs.
  • If useful, an llms.txt file and plain Markdown versions of pages for AI tools.

Not included: building or hosting the assistant, choosing a vector database, or prompt engineering. Those stay with your engineers or your vendor, and the audit hands them the content-side findings they need.

How it runs#how-it-runs

  1. Question set. You share real questions from tickets, chat logs or bot transcripts, and we build a test set from them.
  2. Audit. We trace the failures to their causes in the content and deliver the report with the fix list.
  3. Plan. If you go ahead with implementation, we agree the scope from the fix list and we send a fixed quote.
  4. Implementation. We restructure and rewrite in your repository or help center, and re-run the question set after each round to see which answers changed.
  5. 30-day fix window. For 30 days after delivery, if a question from the set, or a new one in the same area, exposes a content problem we missed, we fix it at no charge.

We work on the content, not on the model or the vector database. If your engineers own the retrieval pipeline, we coordinate with them on chunk boundaries and metadata fields so the content and the pipeline agree.

Pricing#pricing

Audit
$2,500
about €2,300
Implementation
From $6,000
about €5,500

The audit has a fixed price and stands on its own: you keep the report and the fix list whether or not you hire us for the rest. It also makes the implementation quote precise, because the fix list already names the pages involved.

Implementation starts at $6,000 and depends on what the audit finds: how many pages need restructuring rather than light edits, how many products and versions share the same content, whether metadata can be added in bulk or needs page-by-page decisions, and whether the content lives in Git or in a help center that has to be edited through its own interface.

Prices are in US dollars; the euro figures are approximate conversions for reference. If the questions customers ask are not covered anywhere yet, support-ticket mining finds and writes the missing answers before the restructuring starts.

Frequently asked questions#faq

Why does our support bot make things up?

Often because retrieval hands the model a passage that is outdated, contradicts another page, belongs to a different product version, or only makes sense next to the paragraph before it. The model then fills the gap fluently. Fixing the content removes many of those failure modes; it does not replace good retrieval and prompting.

Do you change our AI vendor or model settings?

No. We work on the content and its metadata. We document what changed and why, so your engineers or your vendor can adjust chunking or filters to match.

Is AI-ready documentation different from good documentation?

Mostly it is good documentation with less tolerance for shortcuts. A person can skim past a stale paragraph or work out which version a page covers; a retriever cannot. The work makes those implicit things explicit.

Does this work for a hosted help center, or only docs-as-code?

Both. Hosted help centers and docs-as-code sites have the same content problems. If the content is also moving to a new platform, the restructuring can happen during the migration instead of twice.

What do you need to start the audit?

Access to the content, a sample of real customer questions, and, if you have them, transcripts where the assistant answered badly.

Send us what you have

Send a link to your docs, an export, or a few lines on what needs to change. We reply within 1 business day with questions or a quote.

Start a projectEmail us