usdigitalmarketingindex.com SEO
How do you design an SEO recovery strategy for a website with several years of declining visibility?
Updated November 15, 2026
Treat the decline as several possible causes acting over time, and test them one at a time against evidence. A recovery strategy built this way has four parts: a baseline with a dated history of what changed, a breakdown of the loss by page type and query type, a check that separates lost rankings from lost demand and from measurement changes, and a short list of experiments with review dates. A multi-year decline can have more than one cause, and assuming a penalty before checking sends the first months of work in the wrong direction.
Baseline and historical change log
Fix the starting point before interpreting anything.
Export clicks and impressions for the longest period Search Console provides, and pull analytics data for the full period of decline. Then build a timeline of what changed on the site and around it: redesigns, platform migrations, URL changes, content removed or merged, changes of ownership or staff, tracking changes, and the dates of Google’s announced updates.
Check one thing early because it changes the whole plan. Search Console’s Manual Actions report states whether a human reviewer at Google has taken action against the site (Search Console’s Manual Actions report). If the report is empty, there is no manual penalty to lift, and the word penalty should leave the conversation.
All of this presumes a history to examine. A new domain has none, and its plan starts from a different place, laid out in the guide to what an SEO roadmap for the first 90 days of a new domain should include.
Segmenting templates and query classes
A sitewide line hides the pattern. Break the loss down two ways.
- By page type: product or service pages, category pages, articles, location pages, tools. Compare each type’s clicks at the start and end of the period.
- By query class: brand queries, non-brand commercial queries, informational queries.
The combination locates the loss. A decline concentrated in informational queries on article pages points somewhere different from a decline in brand queries, and both differ from a loss spread evenly across every type.
Ranking versus demand versus measurement
Four different things can produce a falling traffic line, and only one of them is a ranking problem.
| What changed | How it shows in the data | How to check |
|---|---|---|
| Rankings | Impressions and average position fall for queries where demand is stable | Compare position over time for a fixed set of queries |
| Demand | Impressions fall while position holds | Compare with search interest for the topic over the same years |
| Measurement | Analytics falls while Search Console clicks hold | Compare the two sources; check the dates of tracking and consent changes |
| Click behavior | Impressions and position hold while clicks fall | Review what else appears on the results page for those queries |
Work through the table for each segment from the previous step. A segment can show more than one cause.
Prioritized experiments and reassessment
Write every explanation down as a hypothesis, with the evidence for it kept apart from the speculation. The register below is an example. Its entries are fictional and show the format.
| Date logged | Hypothesis | Evidence | Speculation | Test | Review date |
|---|---|---|---|---|---|
| Entry date | Article traffic fell after a migration changed article URLs without redirects | Old URLs return 404 in a crawl; decline begins the month of the migration | That lost links are the main cause | Restore redirects for the old URLs with the most external links | Eight weeks after shipping |
| Entry date | Informational demand for the core topic has fallen | Search interest for the topic has declined over five years | That nothing on the site caused the loss | None; record as context and adjust targets | Next quarter |
| Entry date | Category pages lost ground to competitors with fuller content | Position fell for a fixed set of category queries; competing pages carry buying guidance ours lack | That content depth is the reason | Rebuild five category pages; hold five similar pages unchanged | Twelve weeks after shipping |
| Entry date | A broad Google update affected the site | Decline steps coincide with announced update dates | That the update targeted this site | No direct test; assess content against Google's guidance | Next quarter |
On the last row, Google says that “Several times a year, Google makes significant, broad changes to our search algorithms and systems” (Google’s explanation of core updates). Google describes these changes as broad, so the response is an assessment of the content, not a search for a switch.
Rank the tests by the share of the loss the segment represents, the strength of the evidence, and the cost of the test. Run few enough at once that results can be attributed, and keep unchanged pages for comparison where the test allows it.
Set a review date for every row, and at each review move the row to supported, not supported, or still open. A hypothesis without a review date hardens into a belief.
This approach covers decline over years with several causes. An investigation that starts from one specific update date follows a narrower procedure, described in the guide to investigating a traffic decline that began after a Google core update.
A warning for the first report to stakeholders: present the register, not a diagnosis. After the baseline and segmentation you will know where the loss is. You will not yet know why.
Add your comment to this answer