A forensic SEO audit is not a regular audit run in a panic. A regular audit asks “what could be better.” A forensic audit asks one question — what changed, and when — and refuses to look at anything that cannot be tied to a date. That constraint is the whole method, and it is why forensic audits find causes while checklist audits find symptoms.
What follows is the sequence I use when a site loses meaningful traffic and nobody knows why.
Build the timeline before you form a theory
Open Search Console, set the range to 16 months, and plot clicks and impressions together. You are looking for the shape of the decline, because shape narrows the cause faster than anything else.
- Cliff (one to three days): technical or manual. A deploy, a robots change, a migration, a penalty.
- Step down over 5 to 14 days: almost always an algorithm update rolling out.
- Slow bleed over months: competitive displacement or content decay, not an event.
- Impressions flat, clicks down: SERP feature change or a title/meta rewrite — you still rank, fewer people click.
That last pattern gets misdiagnosed constantly. If impressions held and clicks fell, you did not lose rankings. Something above you in the results page started absorbing the click, or your snippets changed. Different problem, different fix.
Isolate the segment that actually dropped
Sitewide declines are rare. Almost always a subset fell hard and dragged the total. Break the traffic down by directory, by page type, by device, and by country, then compare each segment across the same before-and-after windows.
When you find that the blog held steady and the product category pages lost 60%, you have eliminated most theories in a single step. A drop confined to mobile points at rendering or interstitials. A drop confined to one country points at hreflang, a geo-redirect, or a market-specific competitor. A drop confined to one template points at a code change to that template.
Line up your own deployments against the dates
Before blaming Google, get the deploy log. Ask engineering for every release in the two weeks either side of the decline, and ask marketing for every CMS change, plugin update, redirect, and site migration in the same window.
In my experience the majority of “algorithm hits” are self-inflicted and dated to a Tuesday afternoon deploy. The usual suspects: a staging robots.txt pushed live, a canonical tag templated to the homepage, a JavaScript framework upgrade that broke server-side rendering, pagination rewritten so page two onward became uncrawlable, or a redirect map that quietly chained three hops and lost the signal.
Check the Wayback Machine for your key templates on either side of the date. Seeing the actual rendered HTML from before the drop resolves arguments that would otherwise run for a week.
Crawl for evidence, not for a score
Now run the crawl — after the timeline, not before. You are looking for specific defects on the specific segment that fell, and you want the evidence attached: the actual title string, the actual URL, the actual H1 text, the actual status code.
An audit that says “42 pages have duplicate titles” is useless in a forensic context. An audit that says “these 42 URLs all render the title ‘Products | Brand’ because the template variable is empty” is a diagnosis. Run a quick scan of roughly 25 representative URLs first to confirm the pattern, then a deep crawl across the affected section — 900+ pages is achievable and often necessary when the defect only appears on deeper templates.
Separate an algorithm hit from everything else
If your dates line up with a confirmed core update and your deploy log is clean, treat it as an algorithm event and change your approach. Core updates are re-assessments of quality and relevance across a whole category; they are not a bug you patch.
The diagnostic questions are uncomfortable and worth asking honestly. Did competitors in your niche move up on the same dates, or did everyone drop? Are the pages that fell your thinnest, most templated, most affiliate-heavy ones? Did you publish a large batch of low-substance content in the preceding months? Thin content loses even when it has links pointed at it, and a core update is often the moment that bill arrives.
Check whether the SERP itself changed
Sometimes nothing about your site changed and you still lost half your clicks. An AI Overview appeared above the fold. A shopping carousel took the top three slots. A local pack got inserted. A competitor bought their way into a shopping unit on your best query.
Compare current SERP features against what you recorded before, query by query for your top 20 terms. If features are the cause, the fix is not technical — it is targeting queries where you can still own the click, and building the kind of brand and entity presence that survives an AI-summarized result. This is also worth measuring directly: brand mention counts across ChatGPT, AI Overviews, Gemini, and Perplexity tell you whether you exist in the layer that increasingly sits above the ten blue links.
Rank the findings by evidence, then fix in order
Finish the forensic SEO audit with a ranked list where every item carries a date, a segment, and a piece of evidence. Anything you cannot attach those three things to goes in a separate “general improvements” pile and does not get worked on this month.
- Reverse anything technical that dates to the decline window — fastest recovery, highest confidence
- Fix crawl and indexation defects on the affected segment
- Address content quality on the specific pages that fell, starting with the highest historical traffic
- Rebuild competitive position where the loss was displacement rather than penalty
Measure recovery as a trend, not a daily reading
Technical reversals can show movement within one to three weeks once the affected pages are recrawled. Content and quality recoveries typically wait for the next broad update, which may be months out. Set that expectation before you start fixing, or the third week will feel like failure.
Track the affected segment separately from the rest of the site and read four-week slopes. Two or three positions of daily movement is normal noise and means nothing. If you want the crawl evidence, the competitor comparison, and the rank trend sitting in one workspace while you work a case like this, SEO Rocket does it at a flat $50 a month — though the timeline discipline above is what actually solves it, whatever tools you point at the problem.
