AI Builder Review — let Wally itemize what to tighten

AI Builder Review is one feature that works the same way across every Builder surface — Docs, Site pages, Deck slides, Design canvases, and Social posts. Click "Review with Wally" and Wally returns 3–8 itemized suggestions, each one pinned to a specific passage with a literal before/after rewrite. Accept the ones you like with one click; the rest stay invisible.

Where to find it:

• Builder Docs — open a doc and click "Review with Wally" in the toolbar.
• Builder Sites — open a page and click "Review with Wally" in the page actions.
• Slide Decks — open a deck and click "Review with Wally" in the slide rail header.
• Design Studio — open a canvas and click "Review with Wally" in the right panel.
• Socials — compose a post and click "Review with Wally" before scheduling.

How it works under the hood:

Each surface flattens its artifact into a list of "passages" — a stable id, a role (heading / paragraph / list-item / block-title / caption / CTA), and the literal text. Wally returns suggestions pinned to those passage ids, so when you click Apply on a suggestion the editor knows exactly which heading or paragraph to swap. New surfaces only need to teach Wally a new prompt — no new function, no new contract.

What the review pass focuses on:

By default, Wally optimizes for tightness, clarity, and tone-fit to the artifact kind. You can steer it with the "Focus" field: paste a sentence like "make it tighter," "more donor-friendly," "land the federal procurement audience," "convert harder," or "less corporate." That focus rides the prompt as an explicit goal and Wally rewrites accordingly.

What you see in the side panel:

• A list of 3–8 suggestion cards. Each card shows: which passage (heading text or first line), why Wally flagged it (one sentence), the literal before text, and the literal after text.
• A green Apply button on every card. Apply swaps the passage in place — non-destructive in the sense that Builder's normal undo (⌘Z / Ctrl+Z) reverts it like any other edit.
• A Discard button if a suggestion misses the mark.
• A "Re-run" button at the top if you want a fresh set of suggestions (e.g. after applying a few and wanting the model to re-evaluate).

What you can rely on:

• Apply is literal — Wally hands the editor the exact replacement text. No partial diffs, no editorial guesswork at the boundary.
• Suggestions never touch CTAs without flagging the change explicitly (a CTA rewrite always cites why it's better than the original).
• Reviews respect your Wally model sovereignty choice (see "Which AI features respect my Wally sovereignty choice"). If your org is on a BYO endpoint, the review pass runs against your endpoint.

Cost:

Each review pass is metered against Wally tokens like any other Wally call. The token count and the +15% chargeback show on the same usage card as Wally chat (Settings → Usage → Pay-as-you-go). A full review of a typical doc or page usually costs a few cents.

When to use it:

• Final polish pass before you publish a page, send a deck, or schedule a campaign.
• Quickly tighten copy somebody else wrote without rewriting from scratch.
• When the doc is long and you want a fast read on which paragraphs are the weakest.

Review is meant to be a final-pass primitive, not a draft generator. For drafting from scratch, ask Wally directly in chat ("draft a landing page for X") and use Review afterwards to itemize the edits.