What a Google demotion looks like from the inside

bangkok.biz, a business directory we run, lost most of its Google traffic over the May to June 2026 core update. Nothing on the site had changed that week. This is how we told a quality demotion from a technical fault, what we found, and what we changed.

First, what it was not

The first hours went to the technical checklist: robots, canonicals, sitemaps, server errors, CDN misconfiguration, a bad deploy. All clean. The pages were indexed, fetchable and unchanged. A technical cause produces a technical trail in Search Console: crawl errors, "discovered, not indexed", dropped pages. There was none.

The tell for a quality classifier is different. Impressions drop across every page type at once, the pages that still rank keep the same positions, and nothing in the crawl reports moves. That pattern means the site as a whole was reclassified, not that particular pages broke.

What we found

The directory has about seven thousand listings imported from a pipeline, cleaned, and enriched. Over the two weeks of the update rollout every page type fell together. When we looked at the site the way a quality rater would, three things stood out:

• Thousands of listing pages that were structurally identical, each with a generated summary block and a generated review digest on top of the imported facts.
• Many thin category and area pages with only a handful of listings each.
• An offer, on the page, of paid listings with do-follow links. That is a link scheme by Google's definition, whatever the intent.

None of these is a bug. Each is a choice that looked reasonable on its own and added up to a site that reads as generated at scale.

What we changed

The recovery work is unglamorous:

• An index diet. Thin combination pages are kept out of the index; the sitemap lists only pages with enough real content to stand on their own.
• The generated blocks come off the listing pages. What stays is the imported fact set plus data we verify ourselves.
• The link offer became plain listings. Paid tiers buy visibility on the site, not links.
• First-hand data that a scraper cannot produce: verified photos, opening details and owner-supplied descriptions, added listing by listing.

It is ongoing, and we will not claim a recovery until Search Console shows one.

The rules we now run every site by

• No structured data the page cannot prove. If a rating is not in a database with the reviews behind it, it is not on the page.
• No generated blocks at scale. A page exists because it has facts that the template cannot produce.
• No paid links, no do-follow offers, no link exchange between sites we own.
• Every number on a page has a source and a date.
• Diagnose from the daily series, not the monthly one. The shape of the drop tells you the class of the cause before you look at a single page.
• Register the other engines. Bing and the others do not run Google's classifiers, and they carried real traffic while Google was cold.

If your site dropped and the crawl reports are clean, look at what you cannot prove, what you generated, and what you sell in links. That is where we found ours.