The audit & editor

What happens when you audit a page

You can start an audit two ways: click Audit → on any row in the Optimization Queue, which opens that page’s editor and starts the audit automatically, or click Run audit directly inside the editor. Either way, the four engines run concurrently and stream their results back live — you watch each dimension’s score and issue count populate one by one instead of staring at a spinner until everything finishes.

Only one audit runs per page at a time; triggering a second one while the first is still going tells you to wait. If a run stalls or crashes, it’s automatically cleaned up so it never blocks the next attempt or leaves phantom issues sitting open in the queue.

The four engines, five dimensions

One audit scores a page across five dimensions, produced by four engines running in parallel (accuracy and citability come from a single engine). Each dimension contributes to the composite Optimization Score, with accuracy and citability carrying the most weight:

The composite is a blend over whichever dimensions actually applied to this page — if one can’t run (say, Authority has no matched prompts or competitor evidence to compare against yet), it drops out and the rest take over, so a page with less data still gets a fair score across what did run.

Accuracy & Citability

Reviews the page text for factual-risk and citation-readiness signals. It is not an external fact-check or a calibrated prediction of citation likelihood. Accuracy looks for unsupported claims, misleading statistics, and shaky logic. Citability looks at structure, clear definitions, authoritative tone, source attribution, and structured data. Without an AI model connected, this engine falls back to a deterministic heuristic pass — flagging phrases like “studies show” with no named source, vague quantifiers like “a lot of,” uncited numbers, and run-on sentences — and the result is labeled simulated in the UI so you know it’s the lighter-weight pass.

Authority

A competitive read: how complete is your topical coverage next to the pages actually winning citations on your matched prompts? When it can, this engine fetches the real competitor pages your tracking runs have seen cited on this page’s matched prompts, compares your outline against theirs, and proposes ready-to-insert passages for the topics they cover that you don’t. Without fetchable competitor pages, it falls back to a lighter token-overlap comparison against your matched prompts. With neither matched prompts nor any competitor citation evidence yet, Authority has nothing to compare against and sits out of the composite entirely.

AI Coverage

Does this page actually answer the questions an assistant would ask about it? When the page has confidently matched tracked prompts, those real prompts are the scored query set. Their covered, partial, or not-covered verdicts generate findings, and the orchestrator derives the persisted score from the shared finding-penalty ledger rather than trusting an engine-supplied number. Without any confident matches, the engine fans out 5 to 8 exploratory queries from the page’s title, topics, and category, but reports AI Coverage as not applicable and produces no findings or score from that exploration.

AI Visibility — the measured dimension

This dimension combines what Monitor observed in trusted tracking answers with a deterministic read of the current page. For every answered prompt-platform observation on a matched prompt, the exact page citation contributes 50 points, current-page query coverage contributes 30, 12, or 0, and a brand mention contributes 20. Observations are averaged with page-to-prompt match confidence as the weight. It is a transparent product index, not a calibrated probability of future visibility.

The sidebar shows the working behind that number as a per-prompt evidence panel: each matched prompt gets a row with quick chips — cited or not cited, brand mentioned or absent, and whether the current page covers query, gives partial coverage, or has no coverage — plus a “cited instead: …” line naming the competitor domains winning that prompt whenever any do.

Inline fixes, insertable passages & recommendations

Every issue the engines retain lands in the sidebar as one of three kinds. Only an inline fix has an exact in-document anchor and wavy underline; add-content passages and page-level recommendations live in the sidebar:

Any issue can be dismissed instead — you’ll pick a short reason (Fixed manually, Wrong, Off-voice, Too risky, Other) so the pattern is visible later. A dismissal awards no recovery points; it removes that finding from the recovery denominator. Accepted findings among those that remain can still recover the dimension toward 100.

Editing the document by hand

The audit’s click-to-accept fixes aren’t the only way to change a page — you can also edit any block directly. Hover over a paragraph, heading, list item, quote, code block or table and a pencil appears at its right edge; click it and the block becomes a text box you rewrite freely, with Save and Cancel. Saving replaces just that block, and clearing the text before you save deletes the block.

Hand edits and accepted audit fixes build up in the same draft and behave identically afterward: nothing reaches your live site until you publish, and every save is checked against the version you started from, so two edits can’t silently overwrite each other.

How your score recovers as you fix things

You don’t need to re-run the audit to see the workflow projection move. Every accept or dismiss recalculates the projected Optimization Score instantly, per dimension, using the fixed finding-penalty table:

Each fixable dimension starts at 100, subtracts fixed penalties for its grounded findings, counts at most three findings per category, and applies a dimension-specific floor. Accepted findings earn back their share of the remaining penalty pool. Dismissed findings leave that pool without earning points; open and stale findings stay unresolved. More severe findings carry more penalty weight, so accepting the few things that matter most moves the projection more than clearing minor notes.

The composite Optimization Score shown at the top of the sidebar is simply the same weighted average described above, recomputed from these live per-dimension scores — and it’s always recalculated on the server at save time, never trusted from whatever the browser last displayed. It is a deterministic audit-workflow and prioritization summary, not a calibrated measure of overall content quality, citation probability, or business impact.

Audit credits

Running or re-running a full editor audit uses one audit credit on every plan — your first audit on a page and every subsequent re-audit each draw from the current period’s audit-credit allowance. The free scores that got the page into the queue in the first place — the QuickScore and the Priority Score — never cost a credit. The full editor audit consumes a credit whether it uses model-backed engines or the keyless heuristic fallback. For exact allowances per plan and what happens when you run out, see Plans, quotas & credits.