Recover from a Google Algorithm Update Hit: A 4-Stage Claude Diagnostic

December 11, 2025. A B2B SaaS client's head of growth sent me a screenshot at 11:47 PM. The Google Search Console line chart had dropped a cliff on the rollout start date — total clicks down 38% week-over-week, 47 of their top 50 URLs losing rank. By the time I logged in the next morning, the team had already done the panicked thing: rewritten three H1 tags, added 200 words to two pages, and submitted a reconsideration request (which Google doesn't accept for core updates, but nobody wanted to be the one to point that out at 1 AM).

The rewrite-and-pray approach doesn't recover from a Helpful Content or core update hit. It can take 6–9 months to see the kind of recovery that's actually durable, and the work isn't patching symptoms — it's figuring out which of the four failure modes your pages fell into, then triaging the worst offenders by traffic potential, not by gut feel.

After running this diagnostic on three core update hits (June 2024, December 2024, December 2025) and watching the same framework produce different action plans each time, I settled on a 4-stage method that runs almost entirely in Claude. It scales to 50–500 affected URLs, surfaces things a human reviewer would miss at row 30, and — most importantly — produces a prioritized fix list instead of an undifferentiated panic response.

Here's the workflow.

Why Manual Diagnosis Fails at 50+ URLs

The instinct after a core update is to open your worst-affected URL and stare at it. You find a paragraph that looks weak, you rewrite it, you move to the next URL. By URL 15, you're tired. By URL 30, your judgement is contaminated by what you've already seen. By URL 50, you're pattern-matching your own first 49 impressions instead of what's actually on the page.

The reason a multi-stage Claude diagnostic works: each stage asks one question only, so the model's pattern recognition stays focused. Stage 1 isn't "is this page good" — it's "does this page match the intent Google is rewarding right now for this query." Stage 2 isn't "is this page authoritative" — it's "what E-E-A-T signals are present and which ones are missing for this specific topic." Single-question scoring produces auditable output. Composite "quality scores" produce vibes.

You'll get through 50 affected URLs in an afternoon. The same work manually takes two analysts a week, and the second analyst will disagree with the first about 30% of the time.

The Data You Need Before You Start

Export these three things:

1. From Google Search Console — Performance report filtered to the 28 days before the update rollout and the 28 days after, segmented by URL and query. You want clicks, impressions, CTR, and average position for every affected URL. The before/after delta is your "what changed" signal.
2. From your CMS — For each affected URL: title, H1, meta description, word count, last modified date, author, internal links pointing in, internal links pointing out. This is your "what's actually on the page" signal.
3. From your rank tracker or a manual SERP check — The current top 5 results for each affected URL's primary query, with their word count, publish date, and author. This is your "what Google is rewarding right now" signal.

The first two you can export in 10 minutes. The third is the chore — but it's the most important. Without a current SERP baseline, you're scoring your page against yesterday's Google, not today's.

The 4-Stage Diagnostic

Each stage is a separate Claude session with a single, focused prompt. Don't try to combine them — the output of Stage 1 is the input to Stage 3, and the output of Stage 3 drives your final fix list. Running them in one mega-prompt produces a wall of text that doesn't tell you what to do first.

Stage 1: Intent Match Diagnostic

The question: For each affected URL, does the page still satisfy the intent Google is rewarding for this query — or has Google's interpretation of the intent shifted, and the page is now answering yesterday's question?

What to feed Claude: A table with one row per affected URL containing: URL, primary query, current avg position, position before update, current title, current meta description, H1, and the first 300 words of the page body. Also feed the current top 5 results for each query (URL + title + meta description + first 300 words).

The prompt:

For each row in the table, compare my page's title, H1, and opening 300 words against the top 5 currently ranking results for the same query. The query intent is what Google is rewarding RIGHT NOW, not what I assumed when I wrote the page. For each row, output:

1. Intent verdict (one of: ALIGNED, PARTIAL MISMATCH, FULL MISMATCH)
2. Intent type the SERP rewards (one of: how-to, comparison, listicle, opinion, definition, transaction, case study, data study)
3. Intent type my page delivers
4. One-sentence reason citing the specific evidence from the SERP vs. my page

Do not score quality yet. Do not suggest rewrites. Just diagnose intent alignment. Output as a markdown table.

What you're looking for: Pages that scored PARTIAL or FULL MISMATCH are the easiest wins. If the SERP has shifted from "how-to" to "comparison" for a query, and your page is still a how-to, you can usually recover 60–80% of lost traffic with a structural rewrite (not a 200-word patch). These pages go to the top of the fix list.

For one December 2025 hit I worked on, 18 of 50 affected URLs were intent-mismatched. The client's blog had leaned heavily on how-to formats; the SERP had shifted to comparison + data study for those queries. The fix was changing the H1, the H2 structure, and adding a comparison table — not new writing.

Stage 2: E-E-A-T Gaps

The question: For URLs that passed Stage 1 (intent-aligned), what E-E-A-T signals (Experience, Expertise, Authoritativeness, Trustworthiness, the four quality signals Google's quality raters score) are missing or weak for this specific topic?

What to feed Claude: Same URL list, but now include: author byline, author bio (or note "no bio"), publication date, last updated date, presence of original data/screenshots/expert quotes, outbound citations, presence of an about page, presence of a contact page, and the topic of the article. The topic matters because E-E-A-T weights are topic-dependent — YMYL (Your Money or Your Life, 涉及金钱或人身安全) topics like finance and health require demonstrable credentials; a recipe page doesn't.

The prompt:

You are a Google Quality Rater. For each row in the table, evaluate E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) on these four dimensions, scoring 1–5 each:

1. Experience — Is there evidence the author has actually done the thing the page is about? (First-person accounts, original photos, named case studies, specific data.)
2. Expertise — Does the author bio establish topical expertise? For YMYL topics, are credentials relevant and verifiable?
3. Authoritativeness — Is the page cited by other authoritative sources? Are outbound links to authoritative sources present?
4. Trustworthiness — Are there contact/about/editorial policy pages? Is the source transparent? Is the content factually consistent with cited sources?

Output a table with the four scores, a one-sentence reason per dimension, and a "primary gap" — the single E-E-A-T dimension most responsible for this page's quality weakness. Flag YMYL pages scoring below 12/20 as priority E-E-A-T fixes.

What you're looking for: Pages where the primary gap is Experience are the second-easiest win — adding one original screenshot, one first-person paragraph, or one named case study often moves the needle. Pages where the primary gap is Trustworthiness (no about page, no contact page, no editorial standards) are a site-wide fix, not a page fix. Pages where the primary gap is Expertise but the topic is YMYL are the hardest — they sometimes require getting a credentialed co-author, not just a bio tweak.

For a finance client hit in December 2024, the most common primary gap was Authoritativeness on competitive terms. The SERP winners for "best high-yield savings accounts" were all linked from major news outlets; the client's pages had zero referring domains from Tier 1 sources. No content rewrite was going to fix that — the action item was a 6-month link earning campaign targeting tier-1 finance press.

Stage 3: Content Quality vs. the SERP

The question: For pages that passed Stages 1 and 2, does this page actually add something the SERP winners don't already provide?

What to feed Claude: The full body of your page, plus the full body of the top 3 currently ranking results for the same query. (Top 3, not top 5 — the bar for "information gain" over the top 3 is what matters.)

The prompt:

Compare my page against the top 3 currently ranking results for the same query. Score on these dimensions, 1–5 each:

1. Information gain — Does my page contain specific facts, data, or perspectives that the top 3 do not? If a reader read the top 3, would they learn anything new from mine?
2. Topical completeness — Does my page cover every subtopic the top 3 cover, plus at least one they miss? Are there important subtopics in the top 3 that I'm missing?
3. Depth-to-word-count ratio — Is the page dense with actual information, or padded with restatements and transitions? A 600-word page with 30 unique facts beats a 2,400-word page with 8.
4. Structural clarity — Does the H2/H3 hierarchy match how a reader scanning the page would naturally try to find answers? Are key facts at the top of each section or buried at the end?

For each dimension, cite the specific line or section from my page or the top 3 that drove the score. Flag for "needs rewrite" if the average is below 3. Flag for "needs structural edit" if Information gain is 1–2 and the others are 4–5.

What you're looking for: "Needs rewrite" pages are a 4–8 hour investment. "Needs structural edit" pages are a 1–2 hour investment (move the comparison table to the top, kill the 800-word introduction, add the missing subtopic). For pages that pass all four dimensions, the diagnostic says "the content isn't the problem" — which is itself a useful answer. It means the issue is something other than the page itself (link profile, internal linking, technical SEO), and you should stop editing the content.

Stage 4: Internal Linking

The question: Are the affected pages receiving enough internal link equity (the SEO value that flows from one page to another via internal links, similar to how PageRank flows) to compete with the SERP winners, or are they orphaned or underlinked?

What to feed Claude: A list of all affected URLs, the internal links pointing TO each one (with anchor text and the source page's topic), and the internal links pointing OUT from each one. For comparison, also feed the internal link profile of the current #1 result for each affected query (you'll need a tool like Ahrefs or Semrush for this — or a manual crawl).

The prompt:

For each row, evaluate internal linking on these dimensions:

1. Internal links in — How many unique pages on my site link to this page? Is the count roughly proportional to the page's commercial/strategic value, or is this an orphan?
2. Anchor text quality — Do the inbound anchors describe what the page is actually about, or are they generic ("click here", "read more", "this article")?
3. Topical clustering — Do the pages linking TO this page live in the same topical cluster, or are they scattered across unrelated topics? Topical clustering sends a stronger relevance signal to Google.
4. Outbound link architecture — Does this page link DOWN to relevant supporting pages on my site, or does it link out only to external sources? Pages with rich internal outbound links pass equity to their cluster.

Output a "link gap" verdict (CRITICAL, MODERATE, NONE) and a specific action: e.g. "add 3 contextual links from [specific source pages] with anchors [X, Y, Z]."

What you're looking for: This is often the cheapest fix in the diagnostic. For 20% of affected pages in a typical audit, the only thing wrong is internal linking — 2 contextual links from related pages with topic-relevant anchors can move a page from position 14 to position 8 within 4–6 weeks, without any content changes. The CRITICAL verdict pages are the highest-ROI fixes you have, because they require writing zero new words.

The Prioritization Matrix

After all four stages, you'll have 200–500 cells of diagnostic output. The output is useless unless it tells you what to do first. Sort every affected URL on two axes:

• X axis — Traffic recovery potential (low / medium / high). High means: the query still has meaningful search volume, the SERP has winners but isn't locked up, and the page was earning real traffic before the hit.
• Y axis — Fix cost (low / medium / high). Low means: intent alignment edit, internal link addition, or H1/H2 restructure. High means: full rewrite, new E-E-A-T author, or a 4-hour depth overhaul.

The quadrant that matters is high potential + low cost — these are your Week 1 fixes. For the December 2025 client, 9 URLs fell in this quadrant; we shipped the fixes in 4 days. Six of them recovered to within 15% of their pre-hit position within 5 weeks. The other 3 held at 50% recovery, which is itself useful information — those move to the "deep rewrite" backlog.

Everything else gets a date on the calendar. High potential + high cost pages go to the "rewrite over the next 4 weeks" sprint. Low potential pages (queries with declining volume regardless of who ranks) get a "monitor only" tag. The temptation is to fix the high-cost, high-glamour pages first because they feel important. Resist it. The cheap fixes in the high-recovery quadrant are what rebuild your traffic floor fastest.

The Hard Truth About Recovery Timeline

No diagnostic will recover your traffic in 2 weeks. Google's core update recovery happens at the next core update, because that's when the classifier re-evaluates the site. The December 2025 update hit on rollout; partial recovery started showing in early February 2026 once Google recrawled and reprocessed the fixed pages, and full recovery on the highest-priority URLs landed in the March 2026 core update cycle.

The 4-stage diagnostic doesn't make Google update faster. What it does is make the 6–9 month recovery window count — every fix is targeted at the specific reason that page was hit, instead of generic rewrites that move quality scores up by 0.5 without addressing the actual failure mode.

For the December 2025 client: of 50 affected URLs, 9 high-priority low-cost fixes shipped in Week 1, 14 medium-priority structural edits shipped in Weeks 2–3, 12 deep rewrites shipped in Weeks 4–8, and 15 were deemed unrecoverable and consolidated. By the time Google re-evaluated in the March 2026 cycle, the site had recovered 71% of pre-hit organic sessions across the surviving pages — which is on the high end of what's typical for a hit that severe.

The diagnostic won't tell you Google is fair. It will tell you exactly where the gap is, and which gaps are worth closing.