Claude Code for content refresh and decay detection

Updated: May 21, 2026

TL;DR

• Content decay is a gradual, measurable revenue problem: pages lose rankings, AI citations, and pipeline contribution long before a traffic drop appears on your dashboard.
• Claude Code can analyze exported Google Search Console and Ahrefs data to identify decaying pages, flag structural issues, and output a prioritized refresh list in a fraction of the time a manual audit takes.
• Prioritize refresh work by business impact, not vanity metrics: traffic loss x conversion rate x revenue per conversion tells you which pages are worth fixing first.
• Use a 90-day baseline comparison as your decay signal. A consistent 20%+ drop in clicks or impressions across that window is a reliable trigger for a refresh.
• Detection is automated. The actual rewrite still requires human expertise and a structured framework like CITABLE to restore passage retrieval eligibility.

Most B2B marketing teams still rely on quarterly content audits to find aging pages. By the time those audits surface a problem, competitors have already filled the passage retrieval slots in ChatGPT, Claude, and Perplexity that your content used to occupy. This guide covers how to use Claude Code to automate content decay detection, connect your GSC and Ahrefs data, and build a prioritized refresh list you can action next week. It covers one module of the full Claude Code for SEO playbook, which details the complete Claude Code workflow across technical, on-page, and content operations.

The revenue impact of B2B content decay

The hidden cost of aging content

Content decay is the gradual, often invisible decline of a page's organic traffic, keyword rankings, and AI citations over time. Impressions stay flat, rankings look stable, and the dashboard reads healthy, until a competitor ships fresher content on the same query and your page stops appearing in AI-generated answers.

The AI citation loss is the hardest signal to detect because it happens at the retrieval layer, not the crawl layer. Dense passage retrievers select semantically relevant sections to build a single synthesized answer, outperforming keyword-based retrieval by 9–19 points on top-20 passage accuracy. Based on our analysis of AI citation patterns, these retrievers tend to favor recency, structural extractability, and cross-source consistency when selecting which passages to surface. A page that ranked strongly 18 months ago may still hold its Google position while dropping out of LLM answer generation entirely.

Ahrefs data illustrates the divergence: pages in the top 10 on Google made up 76% of AI Overview citations in mid-2025, but that figure dropped to 38% by early 2026. Traditional rankings no longer predict AI visibility with any reliability. If you're not monitoring citation rate independently, you're missing a substantial portion of your AI share of voice.

Slow manual audits hinder content refresh

A quarterly audit sprint can take weeks to complete and produces a spreadsheet that's already partially out of date by the time it reaches your inbox. For a lean content team managing 200+ pages, that cadence means pages can decay for six months before anyone acts.

Claude Code changes the operational math. Rather than manually comparing date ranges in GSC, pulling Ahrefs position history, and cross-referencing against a content calendar, you export the data and let Claude Code run the comparison. The output is a ranked list, not a raw data dump, which means your team spends time on refresh decisions rather than data wrangling. I've written more about structured AI search audits if you want the fuller context.

The proven method for content decay detection

Extracting GSC data for decay

Start with Google Search Console. In the Performance report, set your primary date range to the most recent 90 days and add a comparison range to the same 90-day window from 12 months prior. Export both datasets as CSV. You want clicks, impressions, CTR, and position for all queries, filtered to your site's key landing pages.

The comparison format matters. You're not looking for absolute traffic volume. You're looking for pages where impressions are holding steady but CTR has dropped (typically a new SERP feature absorbing clicks above your result), and pages where both impressions and clicks are declining together (ranking decay or content relevance loss). Export directly from the GSC interface using the Export button in the top-right corner and choose CSV for Claude Code compatibility.

Ahrefs: spotting underperforming content

In Ahrefs Site Explorer, pull two reports. First, open the Organic Keywords report, filter to your highest traffic-driving pages, and export position history for each. Second, open the Backlinks report and export the Lost Backlinks view for your chosen timeframe.

The position history export gives Claude Code the trend line it needs to calculate decay velocity, not just a snapshot. A page that dropped from position 4 to position 9 over six months is a different problem than a page that held position 4 for two years and dropped in the past 60 days. The second scenario usually means a competitor published something structurally better and the retrieval system updated its passage preferences accordingly.

The lost backlinks export matters because information consistency across sources directly influences LLM citation behavior. Losing high-authority referring domains reduces the cross-source signal that LLMs use to validate claims, particularly for fact-dense B2B pages.

Pinpointing content decay

Export your blog's CMS metadata as CSV (title, URL, publish date, last modified date, word count). Upload this alongside your performance data and instruct Claude Code to flag structural patterns that predict decay. Look for pages where the last modified date is well behind the publish date, indicating zero maintenance cycles. Flag pages with notably low word counts competing for commercial queries, since longer, more comprehensive content typically performs better for fact-heavy searches. Identify topic drift across pillar sections, which reduces entity clarity for LLMs. The output is a content health score per page that combines performance decay with structural risk.

High-impact content refresh list

Once Claude Code processes your exports, it generates a ranked list of pages sorted by decay severity. The prioritization formula that determines which pages to fix first is:

Refresh priority score = (traffic loss x conversion rate x revenue per conversion) / estimated refresh effort

A page losing 500 monthly clicks at a 2% conversion rate and $15,000 average contract value represents significant monthly pipeline exposure. A page losing 50 clicks in the same conditions represents notably lower exposure. Fix the first one this week.

Estimate refresh effort in hours for typical B2B SaaS pages. A statistics update with no structural changes is relatively quick. A full CITABLE restructure with new sections, updated schema, and third-party validation research requires significantly more time. A complete rewrite from outdated positioning is the most time-intensive effort. Use these considerations as starting points to calculate your priority score, adjusting for your team's expertise and content complexity, and avoid spending extensive hours on a page that drives minimal pipeline when a lighter refresh on a different page is worth far more.

Build your AI content monitoring system

┌─────────────────────────────────────────────────────┐
│ GSC Export: current vs. 90-day prior                │
└───────────────────┬─────────────────────────────────┘
                    │
                    ▼
┌─────────────────────────────────────────────────────┐
│ Ahrefs Export: Organic Keywords + Lost Backlinks   │
└───────────────────┬─────────────────────────────────┘
                    │
                    ▼
┌─────────────────────────────────────────────────────┐
│ Claude Code: current-vs-baseline diff               │
└───────────────────┬─────────────────────────────────┘
                    │
                    ▼
┌─────────────────────────────────────────────────────┐
│ Decay score per page (clicks, position, citations) │
└───────────────────┬─────────────────────────────────┘
                    │
                    ▼
┌─────────────────────────────────────────────────────┐
│ Prioritized refresh list (by business impact)      │
└───────────────────┬─────────────────────────────────┘
                    │
                    ▼
┌─────────────────────────────────────────────────────┐
│ CITABLE refresh → Publish → Track citation lift    │
└─────────────────────────────────────────────────────┘

Pipeline summary: GSC export → Ahrefs export → Claude Code diff → decay score per page → prioritized refresh list → CITABLE refresh → citation lift tracking.

How to link GSC and Ahrefs data

Upload both CSV sets into a Claude Code session with a single instruction file that maps column names to variables. Tell Claude Code you want a joined dataset keyed on URL, comparing current period metrics against the baseline period for each page. Include columns for clicks delta, impressions delta, position delta, and a flag for pages where impressions are flat but CTR dropped (SERP feature displacement). From here, every decay finding has a traceable source column, which you'll need when presenting refresh ROI to your CEO or head of revenue.

Selecting your decay baseline period

Use 90 days as your standard comparison window, with 12 months prior as the baseline. Shorter windows may surface more normal fluctuation and algorithm testing noise. Longer windows can mask recent recoveries or improvements.

The exception is seasonal content. If you publish annual reports, product comparison pages tied to vendor releases, or event-driven content, compare the current period to the same period in the prior year to avoid misidentifying seasonal troughs as decay. This works reliably for established content with at least two full annual cycles of data. For content where the annual pattern is shifting, such as a comparison page whose competitive set has changed significantly year-over-year, layer in a rolling 90-day trend alongside the year-over-year comparison to distinguish true decay from a shifting seasonal baseline. For first-year seasonal content without historical data, establish a baseline over the initial cycle before applying decay detection.

How to set decay thresholds

A consistent drop in clicks across a 90-day window, confirmed by a corresponding drop in impressions or position, is a strong signal to trigger a refresh flag. Pages where clicks dropped but impressions held steady at the same position are SERP feature displacement, not content decay. The fix is structural optimization for AI Overview eligibility rather than a full content rewrite.

For AI citation monitoring, track citation rate by query cluster using your AI visibility tracker. Define a set of relevant buyer queries for your content (typically 10-20 queries per cluster), run them across your target LLM platforms, and count how many responses cite your page. A drop from appearing in 3 out of 10 relevant AI responses to 1 out of 10 signals the page has lost passage retrieval eligibility and needs structural attention.

Deploy Claude Code scripts for content audits

Detecting ranking decay with Claude

Paste this prompt into Claude Code after uploading your merged GSC and Ahrefs CSV:

"Compare clicks, impressions, and average position for each URL between the current_period and baseline_period columns. Flag any URL where clicks dropped more than 20% and position dropped more than 3 places. Rank flagged URLs by total click volume lost. Add a column estimating whether the drop correlates with position change (ranking decay) or position-stable CTR loss (SERP feature displacement). Output as a CSV sorted by click loss descending."

The output gives you a triage list. The prompt is repeatable on a monthly cadence, which means your team stops running manual audits and starts reviewing a pre-sorted list instead.

Detecting AI-driven traffic drops

For pages where AI citation loss is driving invisible pipeline drops rather than tracked traffic loss, use this prompt:

"Identify pages in this dataset where impressions declined less than 10% but clicks declined more than 25% in the same period. These are candidates for AI Overview displacement. List them with their current average position, CTR delta, and the primary keyword cluster driving their historical traffic."

CTR displacement is a direct marker of AI Overview absorption. Once you've identified these pages, follow the technical optimization path we cover in Mastering Google AI Overviews.

Flag structural issues in decaying content

Once you have your ranked decay list, run a second Claude Code pass against your blog CMS export to flag structural issues. Instruct Claude Code to identify:

• Sections answering more than one question per block (topic drift, reduces extractability)
• Pages where the primary claim is buried deep into the body (answer delay)
• Statistics lacking a current-year source (staleness, reduces LLM trust signal)
• Pages missing schema markup for FAQ or HowTo content

This structural pass translates the traffic data into actionable rewrite briefs, which is the handoff point between automated detection and human-led refresh work.

Which content needs refreshing for AI?

Score each flagged page using three factors: decay severity, revenue impact, and refresh effort. Decay severity combines the percentage drop in clicks, position change, and citation rate loss. A page that dropped from position 3 to position 12, lost 60% of its clicks, and went from appearing in 4 out of 10 AI responses to 0 out of 10 has high decay severity across all three dimensions.

Revenue impact uses the formula from the previous section: traffic loss x conversion rate x revenue per conversion. Divide that figure by your effort estimate in hours to get a final priority score, then work the list from highest to lowest. This prevents your team from spending full rewrite effort on low-pipeline pages while high-exposure pages wait.

Evaluating content update performance

Record baseline metrics before publishing any refresh: current position, clicks, impressions, CTR, and citation rate across your tracked query set. After publication, run your first comparison once initial indexing is complete. Initial citation improvements can appear relatively quickly for well-structured refreshes. Traffic and ranking recovery typically take longer, depending on crawl frequency and competitive intensity.

Measure citation rate lift monthly using your AI visibility tracker. Run your defined query set across target platforms and count citations. A page that moved from 0 citations to 3, measured against the same fixed query set you defined before the refresh, represents meaningful, reportable progress. The denominator stays constant: you run the same list of relevant buyer queries each month and count how many responses cite the page. Track this across all three surfaces: web search performance via GSC, citation performance via your AI tracker, and training data influence by monitoring whether your refresh content appears in longer-form AI responses that indicate retention beyond real-time retrieval.

Warning signs of underperforming content

Distinguish temporary traffic drops from permanent decay. A significant drop over a short period during a Google core update may be temporary. The same drop sustained over 90 days with no recovery is permanent decay requiring action.

Check whether a competitor published better-structured content targeting the same query. Run the query in ChatGPT and Claude. If a competitor consistently appears while you rarely do, they likely won a passage retrieval displacement and your decay is competitive, not technical. The fix is a CITABLE-structured refresh, not a backlink sprint.

For AI Overview decay specifically, watch for stable rankings in top positions combined with significant CTR collapse. That pattern means Google added an AI Overview above your result and absorbed your clicks. Mastering Google AI Overviews covers the structural optimization path for reclaiming that visibility.

Real example: B2B SaaS content audit

Identifying early content decay

One decay pattern we see repeatedly in B2B SaaS audits is statistical staleness. Pages that cite CRM data, sales benchmark figures, or market size statistics go stale quickly. A B2B SaaS page citing two-year-old industry statistics is a decay candidate even if its Google position hasn't shifted yet, because LLMs deprioritize outdated factual claims in passage retrieval.

AI-driven refresh prioritization

During an AEO-led content audit for one anonymized B2B SaaS client, Claude Code was used in the detection phase to surface a set of pages where statistical staleness combined with structural issues (long sections, buried answers) had reduced citation eligibility across ChatGPT, Claude, and Perplexity. None of these pages had visible traffic drops at the time of the audit. All were already losing AI share of voice. The CITABLE framework guided the refresh: Clear entity and structure in the opening, Answer grounding with verifiable sources, Block-structured for RAG sections, and Entity graph and schema markup throughout.

Reversing decay: measurable wins

After applying that refresh protocol, the same client went from 550 AI-referred trials to 3,500+ in 7 weeks. The result came from an AEO-led process: identifying the right pages to fix, applying a CITABLE-structured rewrite, and sequencing the work by business impact. Claude Code accelerated the detection phase, but the outcome was driven by the quality and prioritization of the refresh itself. Tom Wentworth, CMO at incident.io, described a similar starting point:

"Before Discovered Labs, we were using homegrown LLM prompts, without a clear strategy for what to optimize for or exactly how best to structure content." - Tom Wentworth, CMO at incident.io

After the structured refresh process, incident.io's AI visibility moved from 38% to 64% across tracked queries.

Comparison: manual audits vs. Claude Code automation

Content refresh prioritization matrix

Conclusion

Content decay is a revenue problem before it becomes a traffic problem. The workflow in this guide gives you a repeatable system: export your GSC and Ahrefs data, run Claude Code to identify decaying pages and structural issues, prioritize by business impact using the traffic-loss formula, and apply a CITABLE-structured refresh to restore passage retrieval eligibility. The teams that do this monthly catch decay at the citation layer, before rankings and pipeline numbers confirm it. If you want to apply this workflow without building the infrastructure yourself, the Discovered Labs Starter package includes end-to-end content operations and AI visibility tracking on a month-to-month retainer with no annual lock-in. Book a call and we'll tell you honestly whether we're a fit, or start with the free AEO content evaluator to score your existing pages right now.

FAQs

How often should I run a Claude Code content decay scan?

Monthly is a practical cadence for most B2B SaaS teams. Monthly scans catch decay before it compounds, while weekly runs may surface too much normal fluctuation to act on reliably.

What percentage traffic drop should trigger a content refresh?

A consistent drop in clicks across a 90-day window, confirmed by a corresponding drop in impressions or position, is a strong signal for refresh. Isolated single-month drops should be monitored but not actioned immediately.

Can Claude Code fully automate the content refresh?

No. Claude Code automates detection, structural analysis, and prioritization. The actual rewrite requires human judgment and a structured framework like CITABLE to restore passage retrieval eligibility. Detection without quality rewriting may recover rankings but not AI citations.

How long before refreshed content regains AI visibility?

Initial citations can return relatively quickly after a well-structured refresh. Measurable citation rate lift across a query cluster takes consistent content operations and off-page consistency work over several months.

What data do I need to start a Claude Code decay scan?

You need two GSC performance exports (current 90-day period and the same window from 12 months prior) and an Ahrefs organic keywords export with position history for your top pages. Both are standard CSV exports from those platforms. Claude Code (via Claude Pro or API) processes the uploaded CSVs directly.

Key terms glossary

Content decay velocity: The rate at which a page loses organic traffic, keyword rankings, and AI citation eligibility over time, measured by comparing performance metrics across defined baseline periods.

Citation rate: The percentage of relevant buyer queries in which your brand or a specific page appears in an AI-generated response. Measured by running a defined query set on a specific platform (ChatGPT, Claude, Perplexity, Gemini) and tracking citation frequency over time.

Passage retrieval: The process by which LLMs select specific sections of text from indexed sources to include in a generated answer. Dense passage retrievers outperform keyword-based retrieval by 9–19 points on top-20 passage accuracy by selecting passages based on semantic similarity. Based on our research into AI citation patterns, sections that independently answer a single question and are structurally compact are more likely to be selected.

Content decay baseline: The reference period used to measure current performance against historical norms. A 90-day window compared to the same 90-day window 12 months prior is a commonly used baseline for decay detection.

SERP feature displacement: A traffic drop caused by a new SERP feature (AI Overview, featured snippet) absorbing clicks above a page's organic result, rather than a ranking or relevance decline. Distinguished from content decay by stable impressions alongside falling CTR.