Adapt content for new locales or audiences with human review

Stand up a **locale adaptation partner** — paste source copy plus your glossary and tone rules — emit **adapted draft + terminology receipts + risk flags** — then route every publish path through the same human-review habit as marketing pipelines ([Set up a content review pipeline](/tutorials/set-up-a-content-review-pipeline)) — no invisible multilingual autopilot.

Plus: three Admin-Agent passes — mine your last ten reviewer corrections into an updated gloss fragment you paste next time, diff adapted draft against source claims (**no new promises**), and draft **sampled-review** routing notes when volume would otherwise force rubber-stamping.

Audience Everyone · Admins
Time ~12 min
Prerequisites You can describe *who* the reader is and *what* must stay sacred ([Add your first context file](/tutorials/add-your-first-context-file) — voice, banned terms, and product names). Agent basics ([Create an agent from scratch](/tutorials/create-an-agent-from-scratch), [Give your agent its job description](/tutorials/give-your-agent-its-job-description)). Strongly helpful: [Set up a content review pipeline](/tutorials/set-up-a-content-review-pipeline) — this lesson assumes you already treat drafts as guilty until reviewed.
You'll end up with One **Adaptation** agent charter — sections **Source**, **Adapted draft**, **Glossary receipts**, **CLAIM_CHECK**, and **Reviewer checklist** — plus a ritual: you paste → agent proposes → **you** approve externally ([Run a workflow](/tutorials/run-a-workflow)) or in Chat — nothing ships unattended.

When a tutorial shows italic text in quotation marks, it usually mirrors a label or helper string inside Auxot. Product copy changes between releases — if something reads differently in your workspace, trust what you see on screen.

Callouts with a Worth knowing gold accent are meant as must-read context before you move on. Blockquotes that open with Tip are lighter, optional depth.

Why this matters

Same paragraph, new country or new audience is never “run translate and post.” Names drift. Regulatory hooks appear. Humor lands wrong. The expensive failure is confident, clean prose that promises something legal never approved.

Auxot does not silently rewrite your whole site: you paste chunks because you prompted. The adaptation agent’s job is to compress reviewer labor: surface terminology consistency, flag stretched claims, attach CLAIM_CHECK bullets tied to sentences; still your sign-off before anything customer-visible moves (Ship clear customer communications when the asset faces buyers).

Keep sacred vocabulary in context (Add your first context file): internal product names, trademark casing, and words legal vetoed; so drafts inherit discipline instead of vibes.

Machines propose wording — humans own accuracy and tone.


Quick start

  1. Freeze a locale brief — one context snippet: audience, reading level, currency/date norms, and jokes on/off; stop the agent improvising culture (Add your first context file).
  2. Mint Adaptation — charter forbids inventing features; UNKNOWN when source unclear; ends with Reviewer checklist checkboxes.
  3. Paste real source — landing paragraph, email hero, or policy stub; redact customer identifiers.
  4. Run once cold — scan CLAIM_CHECK; if it invents metrics, tighten charter; rerun.
  5. Wire review — same pause-for-human spine as your drafting pipeline (Set up a content review pipeline); optional workflow step (Run a workflow).

Done? Three saved adaptation packets (source + adapted + reviewer notes): searchable, dated filenames, ready for audits (Run a data privacy review before you ship when data promises appear).


The agent can do that?

1. Turn rejections into glossary patches

Chat → Admin Agent:

Rejected adaptation notes (paste): […]. Existing glossary excerpt: […]. Produce <=12 bullet **ADD / REPLACE / NEVER SAY** lines — markdown — no essay — cite which rejection drove each line.

Why it’s non-obvious: Glossaries rot when living only in reviewer heads; structured deltas reuse after you paste feedback once.

2. CLAIM_CHECK tied to sentences

Adapted draft: […]. Source: […]. Output bullet list — each bullet quotes one adapted sentence — label **MATCHES_SOURCE**, **SOFTENS**, or **NEW_CLAIM** — **NEW_CLAIM** forbidden unless I appended explicit approval note below.

Why it’s non-obvious: Transliteration accidents create liability; forcing sentence-level receipts beats trusting fluent tone (Require human approval before risky actions when publishing channels are sensitive).

3. Sampled-review routing note

We ship >40 adaptations weekly — propose sampling plan — what fraction hits human full read — what signals escalate to always-review — plain bullets — <=10 lines — assume reviewer time capped.

Why it’s non-obvious: 100% review collapses into rubber-stamping; explicit sampling preserves honest gates (Set up a content review pipeline).


Go deeper

Skills for repeat locales

Promote stable adapter instructions (Create a Skill): attach the locale brief; avoid copy-pasting charter walls weekly.

Multi-step handoffs

Translate-then-legal-then-brand (Chain steps so agents hand off cleanly): define what crosses each boundary; no orphan drafts.

Fleet hygiene

When glossary Skills change: broadcast behavior deltas (Update your agents without breaking the team); adapters drift silently otherwise.

Privacy

Locale work often touches examples with personal data: scrub before paste; align handling (Run a data privacy review before you ship).


Walkthrough

Step 1: Pick one channel pilot

Email hero beats an entire landing refresh; narrow scope trains reviewers faster.

Step 2: Paste synthetic risky source

Include a statistic and a joke; confirm CLAIM_CHECK and tone flags fire.

Step 3: Dry-run reviewer checklist

Complete boxes manually; add missing checkbox; feed back into charter.

Step 4: Pair with workflow pause

Reject loop once with narrow feedback; confirm second draft tracks glossary (Set up a content review pipeline).

Step 5: Archive packets monthly

Bundle aids Run a quarterly review of your agents evidence; show adaptation volume + escalation counts.


What’s next

Reference