| Verdict | Applies when |
|---|---|
| UPDATE | Real demand, weak execution. Deserves a ranking it is not getting. |
| MERGE | Two or more pages splitting one intent. Consolidate into the strongest URL. |
| CUT | No demand, no links, no reason to exist. Remove or de-index. |
| KEEP | Performing, or doing a non-search job. Leave it alone. |
My first proper content audit took eleven days and covered about 1,400 URLs. I exported everything into one spreadsheet, colour coded it within an inch of its life, then stared at it for most of a week because I had no idea what to do with row 407.
Row 407 was a 2019 post pulling 40 sessions a month. Two decent referring domains. Position 14 for a query with real commercial intent. Written by someone who had left the company two years earlier. Update it? Fold it into something newer? Delete it and redirect? Every option was defensible, which is exactly the problem. Audits do not fail at data collection. They fail at the moment somebody has to commit to a verdict.
I have since run this on a 60-post blog and on a publisher archive of roughly 38,000 indexable URLs, and the shape barely changes. Below is the version I wish I had at the start: the data that matters, a scoring model that produces a decision instead of a feeling, and the rules for each verdict. Where Google has said something concrete, I link the source rather than the folklore version.
What an audit is really for
An audit is not a spring clean and it is not a report. It is a resource allocation exercise: you have a finite number of writer hours next quarter, and this tells you where to spend them. That framing kills a lot of bad habits. You stop scoring pages on abstract quality and start scoring the gap between what a page earns and what it could earn.
| Outcome | What it buys you |
|---|---|
| Recover decayed traffic | Pages that used to rank and no longer do are the cheapest wins available. The topic is proven, the URL has history, and the fix is editorial. |
| Consolidate split intent | When four posts each half-answer one query, none wins. Merging concentrates links and relevance into a URL that can compete. |
| Reduce quality drag | Google assesses sites as a whole, not only page by page. Pages that are thin, indexed and visible come off first. |
Set the scope first
Three decisions upfront save days later. Skip them and you will restart halfway through, which I have done twice.
1. Fix the time window and freeze it
Use a rolling twelve months so seasonality cancels out. Search Console holds 16 months and then deletes, with no backfill afterwards, including via BigQuery. Export both comparison windows on the same day and keep the files.
2. Pick one unit of analysis
URL-level for blogs and resource libraries. Cluster-level for large sites, where the real question is whether a whole section should exist. Mixing them mid-audit produces a sheet nobody can act on.
3. Declare what is out of scope
Legal pages, checkout flows, login screens, author archives, and tag pages get excluded upfront and handled as a separate technical task, or they swamp your zero-traffic list.
Build the inventory
One row per URL, joined in a spreadsheet or query tool rather than by eye. The classic failure is mismatched URL formats: trailing slashes, protocol differences, and uppercase paths break joins silently and leave healthy pages looking like deletion candidates.
| Source | Pull | Watch for |
|---|---|---|
| Crawler (Screaming Frog, Sitebulb) | Status code, title, H1, word count, canonical, indexability, internal inlinks | Internal inlink count is the most under-used column here; zero usually means accidentally orphaned. Screaming Frog’s free tier stops at 500 URLs. |
| Search Console | Clicks, impressions, average position and top query per page, across two 12-month windows | The interface caps exports at 1,000 rows. The API reaches 25,000 per request and 50,000 per day per site per search type; the BigQuery bulk export has no row limit. Above ~1,000 URLs the UI export misleads you. |
| Analytics | Sessions, engagement rate, conversions and assists by landing page | Stops you deleting the quiet page that drives demo requests. No conversion tracking? Fix that before auditing. |
| Backlink tool | Referring domains per URL | Decides whether a cut becomes a redirect. A page with earned links is never simply deleted. |
| Your own head | Does sales, PR, or support rely on this? | The column everyone forgets. Some pages exist for reasons the data cannot see. |
Eight columns that decide it
You can pull fifty columns. You need about eight. Every extra one is something else to argue about in the review meeting, and arguments stall audits.
| Metric | Reads as | Starting threshold |
|---|---|---|
| Clicks, 12 months | Delivered value | Under 12 a year is functionally zero for B2B. Raise it for consumer sites. |
| Click change YoY | Decay signal | Down 30%+ on steady impressions means the SERP changed. Down on both means lost relevance. |
| Impressions | Latent demand | High impressions with low CTR is the strongest update signal on the sheet. |
| Average position | Distance to the win | Positions 5 to 15 are the sweet spot. Below 30 needs a rewrite. |
| Referring domains | Redirect obligation | Two or more quality domains means this URL cannot return a 404. |
| Conversions or assists | Business value | Any recorded conversion removes the page from the cut pile. Fewer than three internal links is often the whole problem. |
| Query overlap | Cannibalisation | Two URLs sharing a top query, or swapping positions, are merge candidates. |
Turning data into a verdict
Scoring exists to remove personal taste. It needs to sort 1,400 rows into four piles fast enough that you still have energy left to do the work. Score two axes on a five-point scale, then read the verdict off the matrix below. Ten-point scales invite endless debate over whether something is a 6 or a 7.
| Signal | Score | What you are judging |
|---|---|---|
| Demand axis Impressions, position, query volume | 0–5 | Is anyone looking for this? Score the data, not how much you like the topic. |
| Asset axis Links, conversions, internal support, strategic flag | 0–5 | What would you lose by removing it? Links and conversions weigh heaviest. |
| Duplication override Shares primary intent with another URL | Flag | Overrides both scores. Flagged pages route straight to a merge decision. |
| Accuracy override Outdated facts, pricing, or advice | Flag | Fix or remove this sprint. Wrong information is a trust liability. |
Do this before you score Group every URL by its top Search Console query, then look at any query with more than one URL attached. That one pivot table surfaces most cannibalisation in minutes and reshapes a surprising number of verdicts before you score anything. |
The decision matrix
This is the table I actually work from. Find the row that matches the page, apply the verdict.
| Verdict | Page profile | Action |
|---|---|---|
| UPDATE | High impressions, low CTR, position 5 to 15, or clicks down 30%+ on steady impressions | Rewrite for intent, refresh the data, fix the title and opening, add internal links from strong related pages. Keep the URL. |
| UPDATE | Low traffic, strong referring domains, topic still accurate | Expand into the definitive version. You already own the hardest asset to acquire. |
| MERGE | Two URLs ranking for the same primary query, positions trading over time | Strongest links and longest history wins. Move unique value across, then 301 the rest into it. Canonicalise instead if both must stay live. |
| MERGE | Several thin posts covering slices of one topic | Build one substantive guide at the strongest URL. Redirect the rest, repoint internal links at it. |
| CUT | Zero clicks, near-zero impressions for 12 months, no links, no conversions | Remove. 301 if a genuinely equivalent page exists, 410 if nothing does. |
| CUT | Expired events, discontinued products, dead campaign pages | 410 Gone is the honest answer. Redirect only where a real successor exists. |
| CUT | Thin, mass-produced or syndicated pages that are indexed and visible | Highest removal priority on the sheet. Visible weak pages outrank invisible ones. |
| KEEP | Steady traffic, conversions, or top-three positions | Leave it alone; fact-check in six months. Noindex if sales or support use it but search does not. Anything under six months old is too new to judge. |
Updating without faking it
Update is the most common verdict and the most commonly botched. Changing the year in the title, bumping the date, and adding two sentences is not an update. Google’s documentation warns against artificially freshening a page by changing its date when nothing meaningful changed. A real update changes what the page can rank for.
• Read the SERP, not the page. Note the top five results’ format, what they answer in the first screen, and what they include that you do not. Intent drift causes most decay and is invisible if you only look at your own page.
• Fix the first 150 words. If a page ranks but gets no clicks, the problem is the title, description, and opening paragraph.
• Replace facts, not adjectives. Dead statistics, stale screenshots, deprecated features, and broken outbound links do more damage than clumsy prose.
• Rebuild internal links. Three to five contextual links from your strongest related pages, with varied anchors. Cheapest lift in the audit, and routinely skipped.
• Add something only you have. Original screenshots, your own test results, a number you measured. Google’s 2024 systems were tuned against unoriginal content, and reworded consensus is exactly that.
• Keep the URL unless the slug is actively misleading. A new URL resets history and adds a redirect hop for no gain.
Merging: the mechanics
Merging holds most of the compounding value and most of the technical risk. Order of operations matters.
| # | Step | Detail |
|---|---|---|
| 01 | Choose the survivor on evidence | Strongest referring domains, longest indexed history, best existing position. Not the one that is easiest to edit or whose author still works there. |
| 02 | Publish the expanded page first | Let the survivor sit live for a few days first. Redirecting into a page that has not absorbed the content yet wastes the transition. |
| 03 | Redirect one to one | 301 each retired URL to the survivor, never to a category page or the homepage. An irrelevant target gets treated as a soft 404 and passes nothing. |
| 04 | Repoint the internal links | Crawl for links to retired URLs and point them at the survivor directly. Resolving through the redirect works but slows crawling at scale. |
| 05 | Update the sitemap, then wait | Drop retired URLs, keep the survivor, request indexing once, then leave it for three weeks. |
| 06 | Keep the redirects permanently | Removing 301s six months later to tidy the config file undoes the consolidation. Treat them as infrastructure. |
Cutting: the right exit
Four ways to retire a page, and they are not interchangeable. The wrong one either loses value you had or leaves the problem in place.
| Method | Use when | Effect | Avoid when |
|---|---|---|---|
| 301 | An equivalent page exists and the old URL has links or history | Consolidates signals into the target, preserves journeys from old links | No equivalent exists. Bulk-redirecting to the homepage is the classic mistake. |
| 410 | Permanently retired, no successor: expired events, dead products, old campaigns | Signals deliberate removal rather than an error | The page still holds links you could consolidate instead. |
| 404 | Accidental removals, or bulk cleanups where 410s are impractical | Drops out of the index over subsequent crawls | You want to signal permanence, or the URL had real link equity. |
| noindex | The page must stay live for users but should not compete in search | Removes it from the index, keeps it accessible and linkable | You also block it in robots.txt. A blocked page is never crawled, so the tag is never read. |
The mistake that costs most Deleting a page with earned backlinks and returning a 404. Those links took months to acquire and stop contributing the moment the URL dies. Filter the cut list by referring domains before anything ships. |
What Google actually says
Content pruning has collected a lot of folklore. Four claims worth checking against the record.
| Common claim | On the record |
|---|---|
| “Deleting old content boosts rankings” | When CNET removed thousands of archive pages in 2023, Google’s Danny Sullivan said the guidance does not encourage this and that older content can still be helpful. Removing one weak page does not make a site better, though on very large sites it can help Google crawl the rest. |
| “Improve or delete, take your pick” | Mueller’s position is narrower: improve where practical, remove where it is not. Neither endorses keeping thin pages indexed in case they rank someday. |
| “All thin pages are equally bad” | Glenn Gabe documented how Mueller’s answers shifted toward content that is both low quality and highly visible. A thin page nobody sees is lower priority than one collecting impressions on commercial queries. |
| “Pruning saves crawl budget” | Google’s guidance targets sites above a million pages changing weekly, above 10,000 changing daily, or with many URLs stuck in “Discovered, currently not indexed”, and calls these rough estimates. Below that, crawl budget is no reason to delete anything. |
Context that changed the bar The March 2024 core update reworked multiple ranking systems and folded the helpful content work into them. Google expected the combined effort to cut low-quality, unoriginal content by 40%; after the rollout finished on 19 April 2024, it reported 45%. The same release added spam policies covering scaled content abuse, site reputation abuse, and expired domain abuse. A large volume of mass-produced pages is not a decay problem, it is a policy problem. |
Measuring the outcome
Log the date and verdict for every URL you touch, or you cannot separate audit effects from seasonality, core updates, and everything else that moved that quarter.
| Timing | Check | Healthy result |
|---|---|---|
| Week 1 | Recrawl. Redirects resolve in one hop, no chains or loops, no accidental noindex. | Zero chains, zero loops, retired URLs returning the intended status code. |
| Weeks 2–6 | Position and impressions, updated pages only. | Movement once recrawled. Two to six weeks is typical; nothing at three weeks is normal. |
| Weeks 6–12 | Merged clusters: survivor clicks versus the old combined total. | The survivor beats the old combined total, not just matches it. |
| Months 3–6 | Site-level clicks, indexed page count, average position. | Indexed count down, clicks flat or up. Sitewide effects take one to two quarters. |
| Quarterly | Decay check on pages losing 30%+ of clicks year over year. | A short list each quarter instead of a backlog every two years. |
Final Verdict
What I would tell my past self before day one
Having run this on sites from tiny to unwieldy, the change that mattered most was giving up on the idea that the spreadsheet would tell me what to do. It does not. The model gets you to four piles, then a person makes a hundred judgement calls about context the data cannot see. Accept that early and the audit takes days instead of weeks.
Merge is the verdict that pays best, and cut is the one people over-apply. Consolidating four thin posts into one strong page beats deleting all four, because you keep the links, keep the history, and end up with something that can compete. On the 38,000-URL archive the merges drove the traffic recovery; the cuts mainly made the site easier to maintain, which is worth doing but is not the same thing.
If you only have one week, pull the pages sitting at positions 5 to 15 and rewrite the top twenty properly. That band has the best effort-to-return ratio by a wide margin, and it is real editorial work, not a date change and a fresh year in the title.
Row 407, for the record, got merged into a newer post and the combined page reached page one for its query in about two months. It took me a week to decide and twenty minutes to execute. That ratio
