A content audit is four decisions, not a spreadsheet

Why most content audits change nothing

The standard method is to export every URL, pull metrics against each, colour the rows, and hand over a spreadsheet. It is thorough and it usually fails, for a reason that has nothing to do with the analysis.

A spreadsheet with three hundred rows and an action column is not a plan. It is three hundred small decisions nobody has authority or time to make, presented as though the hard part is already done. The hard part has not started.

An SEO audit has the same failure mode, and the defect list regrows for the same reason: cataloguing is not deciding. The difference with content is that the decisions are more consequential, because removing a page is irreversible in a way that fixing a title tag is not.

The version that works inverts the order. Start from the four decisions you are willing to make, then collect only the data that separates them. Everything else is interesting and does not change what you do on Monday.

01

Keep

The page works, or is too new to judge. No action. Most pages land here and that is fine.

02

Merge

Two pages answer the same question. One absorbs the other. This is where most of the upside is, and most of the risk.

03

Rewrite

The page targets something real and answers it badly. Worth the effort only where the query has genuine value.

04

Remove

The page earns nothing, serves nobody, and costs crawl budget and attention. The decision people avoid, and the one that frees capacity.

The data that actually separates those four

Four decisions need surprisingly little data. Most audit templates collect twenty columns and use three.

• Clicks and impressions, ninety days. Clicks say whether anyone arrives. Impressions without clicks say the page is visible and unconvincing, which is a rewrite, not a removal.
• The queries it actually ranks for. The single most useful column, and the one most often missing. A page that misses its target and ranks for something else is not a failure; it is a page whose brief was wrong.
• Inbound internal links. A page with real editorial links has been endorsed by your own site. Removing it breaks something. A page with none is cheap to remove and was probably never supported in the first place.
• Last meaningful update. Not the modified date, which reflects plugin churn. When did the content last change in a way a reader would notice.

Two columns people collect and rarely use: word count, which tells you nothing on its own, and bounce rate, which for an informational page usually means the reader got their answer and left.

Clicks	Impressions	Internal links	Decision
Some	Some	Any	Keep
None	Many	Any	Rewrite, it is visible and unconvincing
None	Few	Several	Investigate, something links here for a reason
None	None	None	Remove
Some	Some	Any, and a sibling ranks for the same query	Merge

Ninety-day window, Search Console plus an internal link crawl

The page that ranks for nothing it was written for

This is the most interesting case in any audit and the one a spreadsheet handles worst.

A page was briefed against a keyword. It does not rank for that keyword. It does rank, quietly, at position nine for four queries nobody planned, and those queries convert better than the original target would have. On a colour-coded sheet it is red, because the sheet compares performance against intent rather than against value.

The action is not rewrite and it is certainly not remove. It is to re-target the page at what it already earns: change the title, lead with the question it actually answers, and let the original brief go. This is the cheapest win available in most audits and it is invisible unless you pull actual query data rather than target keywords.

The reverse case is worth naming too. A page that ranks exactly as briefed and earns nothing is a page whose brief was aimed at a query with no commercial value. That is a planning fault, not a content fault, and rewriting it will not help.

Ranks for nothing, earns nothing, nothing links to it

Remove. This is the clearest case in an audit and the one people hesitate over longest. Hesitating costs crawl budget and reader attention every month it stays.

Misses its target, ranks for four unplanned queries

Re-target. Change the title and the opening to match what it earns. No rewrite needed, and the page is often at position nine because it is answering a question nobody else answered directly.

Hits its target exactly, earns nothing

Leave it and fix the planning. The page did what it was asked. The brief asked for the wrong thing, and no amount of rewriting the page changes that.

What to do about pages that are simply old

Age gets treated as a defect in most audits, and it is not one. A page from 2021 that still answers its question correctly is a page that has had four years to accumulate links and ranking history. Refreshing it because of its date is how sites reset assets that were working. A related judgement, deciding which queries are now answered on the results page and worth conceding, is covered in the AI Overviews piece.

The distinction worth making is between pages that are old and pages that are wrong. Only the second needs anything.

Condition	Signal	Action
Old and still correct	Stable rankings, steady clicks	Leave it. Update the date only if something changed.
Old and factually stale	Names a deprecated tool, a superseded figure, a dead link	Fix the specific claim, not the whole page.
Old and structurally wrong	Answers a question people stopped asking	Re-target or remove, depending on what it earns.

Age alone appears in none of these rows, which is the point

The one genuine exception is anything with a year in the title or the URL. Those pages carry an implicit expiry date, and a page called “best tools 2024” is actively telling readers it is out of date. Either commit to updating them annually or stop putting years in titles.

There is a quieter version of staleness worth checking: outbound links. A page full of links to sites that have moved or closed reads as abandoned even when the argument is still sound, and it is a ten-minute fix rather than a rewrite.

Removal is the decision that frees capacity

Every other decision costs something. Keep costs nothing and changes nothing. Merge and rewrite both cost writing time. Removal is the only one that gives capacity back, and it is the one people put off indefinitely.

The hesitation is understandable and usually misplaced. The fear is losing traffic from a page that might be earning something invisible. The check for that is two minutes: does it have clicks, does anything link to it, does it rank for anything. If all three are no, the page is costing you and returning nothing.

What removal actually gives back, in rough order of value.

Crawl allowance freed 100% of their share

Reader attention not misdirected substantial

Maintenance load removed moderate

Direct ranking gain elsewhere small and indirect

Directionally reliable. The last row is the one people expect to be largest and it is not.

That last bar matters. Removing dead pages does not directly lift the pages that remain, and anyone promising that is overselling. What it does is stop you spending crawl budget, review time and internal links on pages that were never going to work, which is a real return measured over quarters rather than weeks.

One caution: removing and redirecting are different decisions. A page with no value and no inbound links can simply go. A page with inbound links, even bad ones, should redirect somewhere sensible, because the alternative is manufacturing your own broken links.

The merge decisions are where the risk sits

Merging is where an audit either earns its cost or destroys something. Two pages competing for one query is the obvious candidate, and the mechanics of consolidating them well are worth reading separately before you start.

The audit-specific point is about selection: a spreadsheet identifies merge candidates by title similarity, which is the wrong signal. Two pages with near-identical titles can serve completely different intents, and two pages with unrelated titles can compete directly.

The signal that actually works is whether both pages rank for an overlapping set of queries. That is data you already have and it takes one pivot to see.

~60%

keep, no action

~15%

removals, and the fastest to execute

~15%

re-targets, the cheapest wins

~10%

merges and rewrites, where the effort goes

Proportions from audits I have run. The middle two are where teams under-invest.

Auditing a site that publishes programmatically

Everything above assumes pages a person wrote. A site with generated page families needs a different unit of analysis, because auditing two hundred location pages one at a time is neither possible nor useful.

The unit becomes the family, not the page. The questions change with it.

• Does the family as a whole earn anything? If the top ten pages in a two-hundred-page family carry ninety-five percent of the clicks, you do not have a family of two hundred useful pages. You have ten, and a hundred and ninety pages of overhead.
• What differs between two adjacent members? If the answer is a city name and a number, a reader cannot tell them apart and neither can a search engine. That is a template problem, not two hundred content problems.
• Where does the tail flatten? Most generated families have a point past which members earn nothing at all. That point is your removal boundary, and it is usually much earlier than anyone expects.

The action on a failing family is almost never to improve every member. It is to cut the family to the members that earn, and to fix the source data so the next generation does not reproduce the same problem at the same scale.

This is also where the crawl argument stops being theoretical. On a forty-page site, dead pages cost you nothing measurable. On a site with three generated families, they are most of what a crawler sees.

Running one without the spreadsheet trap

The practical version, in the order that keeps it from stalling.

• Removals first. They are unambiguous, they free capacity, and finishing something builds the momentum the rest of the audit needs. An audit that opens with the hardest decisions never finishes.
• Re-targets second. Cheapest wins available, and they need no new writing. Title and opening paragraph, informed by the queries the page already earns.
• Merges third, one at a time, with a full crawl cycle between each so you can see what each one did.
• Rewrites last, and only where the query is worth it. Most rewrites in most audits are not worth doing and get done anyway because rewriting feels productive.

The mechanical parts are worth automating rather than eyeballing. A link map shows which pages have real editorial support and which are orphaned, which is the input the removal decision needs and the one most audits guess at. Two pages competing for the same query is the other automatable check, and it is the one a manual scan misses most often.

One practical constraint that keeps audits honest: cap the number of decisions you will act on in a quarter before you start looking. Thirty is a realistic number for one person. Knowing the cap changes how you triage, because you stop trying to grade every page and start looking for the thirty that matter most.

The alternative, which is what most audits do, is to grade all three hundred equally and then discover there is no capacity to act on any of it. The grading was never the constraint.

The measure of a good audit is not how many rows it has. It is how many URLs are different a month later.