How Do I Audit Content for Search and AI?

One audit, two questions

A content audit for AI and a content audit for search are not separate exercises, and running them separately is how teams end up with two spreadsheets and no decisions. The inventory is the same; only the questions differ.

• Search: is this page indexed, does it earn impressions, and does it convert them into clicks?
• AI: is this page retrievable, is it quotable, and does it appear when you test the prompts it should answer?

The prerequisites overlap almost entirely, which is the useful part: a page that cannot be crawled or indexed fails both, so the technical layer is one pass rather than two.

Build the inventory first

Every audit that stalls stalls here, because the inventory was assembled from memory rather than from a crawl. Four columns, joined on the URL:

1. Every URL, from a crawl, plus its indexability status.
2. Search Console data per URL: impressions, clicks, CTR, position, over 12 months.
3. GA4 data per URL: sessions, engagement, conversions.
4. Last meaningful update, which your CMS knows and nobody remembers.

The crawl is not optional. Without it you audit the pages you know about, and the pages you have forgotten are exactly where the problems live, including orphaned pages nothing links to.

The search pass

Four checks per page, in order, because each one makes the next meaningful:

• Is it indexed? If not, nothing else matters. The Page indexing report gives Google's own reason.
• Does it earn impressions? Indexed with no impressions means Google does not consider it relevant to anything, which is a targeting problem.
• Does it convert impressions to clicks? Good impressions with poor CTR is a snippet problem, and the cheapest fix in the audit.
• Do the clicks do anything? Traffic that bounces is a match problem between the query and the page.

The AI pass

Shorter, because less of it is measurable. Three checks:

• Retrievable. Can AI search crawlers reach it, and does the content render without JavaScript? Our robots.txt tester answers the first in seconds.
• Quotable. Does the page contain atomic facts, claims that survive being lifted out of context, or only qualified generalities?
• Actually appearing. Test the prompts this page should answer across the major assistants, and record whether you are mentioned, cited or recommended.

That third check is manual and there is currently no good substitute, since nothing in Search Console reports AI citation. Keeping the prompt set fixed matters more than the method.

What to automate, and what to keep

Content audits stall because they are treated as one large manual task. Split by what can run itself:

• Automate the collection. The crawl, the Search Console pull, the GA4 pull, and the join on URL. There is no judgement in any of it.
• Automate the flagging. Classifying pages as not-indexed, no-impressions, poor-CTR, declining or duplicate is entirely rule-based once the data is joined.
• Keep the decisions. Whether a page with no traffic should be improved, merged or removed depends on things no dataset contains: whether it serves a customer, whether legal needs it, whether it supports a campaign.

That split is also why fully automated content audits disappoint. They are excellent at producing the classified list and structurally unable to make the calls that follow.

How often, and what triggers an unscheduled one

Twice a year is enough for a full audit on most sites, and the calendar is the wrong trigger for the rest. Four events warrant an immediate pass regardless of schedule:

1. A migration or replatform, which redistributes internal links and orphans pages in bulk.
2. A redesign, particularly one that changed templates, since a removed link module can orphan an entire section silently.
3. A core update that moved your traffic noticeably, because the audit tells you which content types were affected.
4. Publishing at scale, especially anything generated from a template, before the volume becomes a problem to unpick.

Between full audits, the cheap monthly check is the CTR pattern: pages with impressions holding and clicks falling. It takes minutes and catches the fastest-moving problem.

Finishing the audit

The most common failure is not the analysis, it is that the spreadsheet never becomes work. Convert the classified list into dated tickets before you close the file, group them by fix type rather than by page, and start with the CTR rewrites so the audit produces visible movement in its first fortnight.

Grouping by fix type matters more than it sounds. Twenty title rewrites done together take an afternoon; the same twenty spread across twenty page-by-page tickets take a month and usually stall halfway. The audit produced a classification, so use it as one rather than reverting to a page-by-page march through the spreadsheet.

Set a date to run the next one before you close this one. Audits that end without a scheduled successor tend not to have one, and the inventory goes stale within a single quarter of ordinary publishing activity.

Put the date in the calendar rather than in the document. Documents get archived.

Deciding what to do with each page

For the removal side of this specifically, see content pruning with Search Console data.