A publisher whose search traffic came back after a rebuild had thrown it away
A site rebuild had cost a trade publisher more than half its organic traffic in three months. A crawl-and-logs audit, a fourteen-thousand-URL redirect map, a prune of thin pages and a topic-cluster editorial plan brought it back, and put its answers where AI search could quote them.
The shape of the work
Media & Publishing
16 weeks
Fixed price, phased
Client name withheld under NDA. Engagement details are shown to the extent our agreement permits.
The question we were asked
A specialist trade publisher had rebuilt its site on a new platform and watched organic sessions fall by fifty-eight percent over the following quarter. The rebuild was good; the migration underneath it had never been planned as one.
A specialist trade publisher with a twenty-year archive, a small editorial team and a site that had just been rebuilt on a modern platform. The rebuild itself was well made. The migration had been treated as a design project with no search plan, and the traffic that paid for the editorial team had gone with the old URLs. The engagement was sixteen weeks, front-loaded so the bleeding stopped before the strategy work began.
Search & Content Strategy
The decision, first
- 01
A migration is a search project whether or not anyone plans it as one.
The rebuild's designers were never asked about URLs, so the one deliverable that would have prevented the loss was never on anyone's list to be cut from.
- 02
Removing three thousand pages did more than publishing three hundred would have.
More pages was the problem, not the solution. De-indexing three thousand of them did more for rankings than anything published that quarter.
- 03
Being quoted in an AI answer rewards exactly the discipline good editing already demands.
Being quoted in an AI answer is now a visibility channel in its own right, and it rewards the same discipline as good editing: a clear question, a direct answer, evidence after.
What the numbers couldn't answer
Fourteen thousand old URLs returned not-found, every article carried the same template title, three thousand auto-generated tag pages were indexed and outranking the articles they listed, and the new templates failed Core Web Vitals on mobile. The editorial team kept publishing into a site search engines had stopped trusting.
Four faults, in order of damage. Every old URL returned not-found, so twenty years of inbound links pointed at nothing. Every article carried the same template title, so search results showed the site's name where the headline should have been. Three thousand auto-generated tag pages were indexed and outranking the articles they listed. And the new templates failed the mobile performance thresholds, which held back even the pages that were otherwise fine.
We audited the crawl and the server logs together, built the redirect map the migration had skipped, fixed the templates, pruned the thin pages and regrouped the archive into topic clusters with an editorial plan behind them. Structured data and answer-first article blocks then made the archive quotable by AI search, and a dashboard tied every change to the traffic it moved.
How we worked it through
- 01
Crawled the new site and read six months of server logs side by side
Reading the crawl and the logs together is what found the damage: the crawl showed what was broken, the logs showed which broken pages still mattered.
- 02
Built and shipped a redirect map for fourteen thousand legacy URLs
The redirect map shipped in traffic order because the top eight hundred URLs carried most of the loss, and waiting for the full fourteen thousand would have cost another month.
- 03
Fixed template titles, headings and the mobile performance failures
Template titles and mobile performance were fixed before any content work, since a strategy on a site search engines distrust is wasted effort.
- 04
Pruned three thousand thin tag pages and regrouped the archive into topic clusters
Pruning came before clustering, so the archive being organized was the archive worth organizing.
- 05
Added structured data and answer-first blocks so AI search could cite the archive
Answer-first blocks and structured data were added to the pillars first, where the citations were most likely and the effect easiest to measure.
Phase by phase
Phase 1: Diagnose
Crawl and logs together
Crawled the rebuilt site and read six months of server logs alongside it to rank the damage by the traffic it was costing.
- Audit report
- Prioritised fault list
Phase 2: Stop the loss
Redirects, titles, vitals
Shipped the redirect map in traffic order, fixed the template titles and headings, and cleared the mobile performance failures.
- Redirect map
- Template fixes
- Performance fixes
Phase 3: Reshape the archive
Prune, then cluster
De-indexed the thin tag pages and regrouped the archive into topic clusters with an editorial plan behind them.
- Prune list
- Cluster map
- Editorial calendar
Phase 4: Be quotable
Answers and structure
Added answer-first blocks and structured data to the pillar articles and set up the dashboard that ties each change to its effect.
- Structured data
- Answer-first pillars
- Recovery dashboard
The fault list in order of damage and the sixteen-week plan: diagnose, stop the loss, reshape the archive, be quotable. The first six weeks ran as an incident with a daily index check, the last ten as strategy planned against cluster gaps.
What it changed
−58% → +12%
Organic sessions vs. pre-rebuild
3,200 → 0
Thin pages in the index
0 → 140
Articles cited in AI answers
Organic sessions are compared with the same weeks a year earlier, before the rebuild, so seasonality is accounted for; the figure is the twelfth week after the redirects shipped. Thin pages are counted from Search Console's index coverage. AI-answer citations are counted from a weekly check of the questions each pillar article was written to answer.
Client name withheld under NDA. Figures are approximate, drawn from the engagement’s own reporting.
The redirect map nobody had
Fourteen thousand legacy URLs mapped to their new homes, prioritized by the traffic the logs said they still received.
The migration had changed every URL pattern and redirected none of them. The old sitemap, the analytics export and six months of server logs were joined into one table of fourteen thousand addresses, each with the sessions it used to earn and the crawler hits it was still getting. The map was shipped in traffic order: the top eight hundred URLs carried seventy percent of the lost sessions and went live in the first week, and a rule set covered the long tail, sparing fourteen thousand hand-written lines.
- Old sitemap, analytics and logs joined into one table
- Top eight hundred URLs shipped in week one
- Pattern rules for the long tail, hand-checked by sample
Crawl against logs: fourteen thousand legacy URLs by response code (every one a 404 at audit), with the crawler hits each still drew from six months of server logs, the top eight hundred carrying 70% of the lost sessions, and the joined table the redirect map was built from.
Pruned, then clustered
Prune, then cluster: 3,200 thin pages de-indexed in one pass, sample tag pages struck through beside the articles they outranked, and the archive regrouped into twenty-two topic clusters with a pillar, supporting articles and the gaps the editorial calendar fills.
Three thousand thin tag pages removed from the index and the archive regrouped into topic clusters an editor can plan against.
The platform generated a page for every tag ever applied, and three thousand of them (a title and a list of two links) were indexed and outranking the articles they pointed at. They were removed from the index in one pass. The archive itself was then regrouped into twenty-two topic clusters, each with a pillar article and the supporting pieces linked to it, and the editorial calendar was rewritten to fill the clusters' gaps as well as chase the week's news.
- Three thousand auto-generated pages de-indexed in one pass
- Twenty-two clusters, each with a pillar and supporting articles
- The editorial calendar planned against cluster gaps
Answers a machine can quote
Structured data and answer-first article blocks so the archive appears in AI search answers as well as in lists of links.
Search was already answering questions directly, and a publisher that isn't quoted in the answer isn't found at all. Each pillar article gained an answer-first block (the question in the heading, a self-contained answer in the first paragraph, the evidence after), and the article, author and organization were described in structured data. Citations were tracked per article from the week it shipped, which is how the count of one hundred and forty is a measurement and not a claim.
- Question-heading, answer-first paragraph on every pillar
- Article, author and organization described in structured data
- Citations tracked per article from the day it shipped
Answers a machine can quote: a pillar article's question heading, self-contained answer and evidence, the JSON-LD describing its article, author and publisher, and the AI-answer citations tracked each week, 140 articles across the archive.
How the engagement ran
A team of 4 worked on a fixed price basis over 16 weeks, covering Search & content audit, Redirects & templates, Editorial planning. The first six weeks ran as an incident with a daily check against Search Console so each fix could be seen to land; the remaining ten were the content strategy, run with the editor-in-chief in the room so the cluster plan was the editorial team's own.
Sixteen weeks in two halves. The first six were remedial and ran as an incident (the redirect map, the templates, the de-indexing), with a daily check against Search Console so each fix could be seen to land. The remaining ten were the content strategy, run with the editor-in-chief in the room, because a cluster plan the editorial team didn't own would have lasted one news cycle.
Fix by fix
The loss stopped first, then the archive earned its way back
The recovery week by week, one tab for each fix in the order it shipped: sessions against the pre-rebuild baseline, thin pages still in the index and articles cited in AI answers at the end of each window. Switch tabs, or use the arrow keys once one is focused.
Answer-first pillars with structured data
Question in the heading, a self-contained answer, evidence after, and citations counted from a weekly check. Week 12 after the redirects is the reading the study reports.
Study figures: sessions −58% → +12%, thin pages 3,200 → 0, articles cited 0 → 140. Weeks between are illustrative.
From what the crawler saw to the answer that quotes the archive
Every change was ranked by the traffic behind it before it shipped, and measured against the pre-rebuild baseline after. The dashboard at the end is how each fix was seen to land.
- 01 · SourceCrawl + six months of logsThe crawl says what's broken; the server logs say which broken pages still draw crawlers and traffic.
- 02 · JoinOne table, 14,000 URLsOld sitemap, analytics export and logs joined per address, so every redirect ships in traffic order.
- 03 · ReshapeRedirects, prune, clustersPattern rules for the long tail, hand-checked by sample; thin pages pruned before the archive is clustered.
- 04 · DescribeAnswer blocks + structured dataQuestion heading, self-contained answer, evidence after; article, author and organization in structured data.
- 05 · MeasureRecovery dashboardSessions against the same weeks a year earlier; citations from a weekly check of each pillar's questions.
So the recovery doesn’t cost something else
Earning pages kept, links honored, markup honest
Only the thin pages went
De-indexed pages were thin by construction: auto-generated tag pages, a title and a list of two links. The articles they listed stayed in the index, no longer outranked by them, and pruning ran before clustering.
Inbound links land somewhere
Twenty years of links pointed at not-found pages. The top eight hundred URLs were mapped by hand, the long tail by pattern rules checked by sample, and a daily index check confirmed each batch landed.
Markup says what the page says
Structured data describes the article, its author and the organization. The answer an engine quotes is the article's own first paragraph, and citations are counted from a weekly check, not assumed.
Did a redesign take your search traffic with it? Scope your build in 3 minutes.
Scope your buildNearby engagements
Web PlatformsA writing desk where the manuscript never leaves the machine
A desktop notebook built on Electron, where documents live on disk, the editor is a real rich-text surface, and the AI assistant reads only what the writer hands it.
Book Publishing · 14 weeks
AI & AutomationLong text becomes narrated audio, and the job tells you the truth while it runs
A production tool that pulls text out of a source, narrates it, and reports honestly on a job that takes minutes, without pretending to be instant.
Media & Publishing · 12 weeks
Web PlatformsTwo billing models in one product, and neither one pretending the other doesn't exist
A content platform sold both by subscription and as a lifetime deal, where a redeemed code and a monthly plan resolve to the same entitlement, with no parallel systems.
Media & Publishing · 18 weeks
Let's talk
Running a large platform, shaping a first MVP, or getting a product ready for a funding round? Tell us where you are. We'll shape the process around it, and stay with you after launch.
