A real estate platform asked whether a sudden collapse in its indexed-page count meant it had to move to a new domain. Whole-site clicks had barely moved, which is why nobody had caught it earlier: one page family had fallen 98.4% while two others grew enough to cover the loss. This is how we worked out which pages had gone, what demand they served, and what else quietly depended on them.
A real estate property listing platform came to us with an unsettling question. Thousands of its pages were no longer indexed by Google. The change coincided with the start of Google's 18 August 2026 spam update. Requests to index a small set of pages had not brought them back.
Was something wrong with the whole domain? Would the business need to move to a new one?
A domain move would be a substantial decision. Before considering it, we needed to know which pages had disappeared and what those pages did for the business.
What we found: Google's index had lost roughly 42,700 pages from its earlier peak. The evidence pointed overwhelmingly to one kind of page: the filter landing pages that match searches such as "flats for rent in Vastral" to a selection of properties. Individual property pages and project pages continued to receive growing traffic. The domain-wide diagnosis did not fit the evidence.
Most of the indexed-page loss happened in one day
Google Search Console showed 72,769 indexed pages from 11 to 14 August. The count stood at 67,917 on 17 August, then fell to 30,062 on 18 August.
That is 42,707 fewer indexed pages than the earlier peak. Of that movement, 37,855 happened between 17 and 18 August.
Google's Search Status Dashboard records the start of its global spam update on 18 August. The dates coincide. They do not tell us whether that update caused this particular change, or why Google treated these pages differently.
The size of the fall was clear. Its composition was not. Search Console did not hand us a list of 42,707 removed URLs, so we looked at the site's page families.
The loss was concentrated in search landing pages
This platform had three main kinds of pages. Property pages described individual listings. Project pages described named developments. Filter landing pages brought together the properties matching a search for a place, a property type or a configuration.
The surviving indexed population was largely accounted for by property and project pages. Those families also continued to gain clicks. Other parts of the site were too small to account for the missing 42,707 pages. Meanwhile, filter-page visibility collapsed at the same time, and Search Console reported effectively no indexed filter pages across the 8,497 current filter URLs covered by six sitemap reports.
That is a strong conclusion reached by elimination: the missing population was overwhelmingly the filter layer. It is not a claim that every missing URL was individually inspected.
The traffic comparison made the business effect clearer. We compared two 22-day periods with the same weekday mix, before and after 18 August:
| Page family | 25 July–15 August | 22 August–12 September | Change |
|---|---|---|---|
| Filter landing pages | 2,443 clicks | 40 clicks | −98.4% |
| Individual properties | 5,813 | 7,565 | +30.1% |
| Projects | 8,030 | 8,660 | +7.8% |
| Whole site | 16,896 | 16,827 | −0.4% |
At the whole-site level, clicks barely changed. That was not something the client had reported to us; it became clear in the investigation. Growth in property and project traffic happened to cover almost all the visible loss from the filter pages.
Those lost visits represented specific demand. A page for "flats for rent in Vastral" had received 165 clicks in the longer pre-event period we inspected, then none after 22 August. A page for "bungalows in Hathijan" went from 80 to none; "flats for sale in Thaltej" went from 32 to none. These are searches for a choice of available properties. An individual listing does a different job.
The domain had not stopped appearing in Google. An important route from search demand to the inventory had largely disappeared.
Why could the site create so many search pages?
The filter system could turn combinations of locality, property type, sale or rent, and configuration into page addresses. On 16 September, 12,917 such addresses were declared in the site's sitemaps. Google knew 317,259 URLs across the whole site, against an estimated 45,000 pages corresponding to current content.
That comparison showed the scale of the URL history. It did not tell us how many were old filters, withdrawn listings, redirects or other historical addresses. We could, however, test how the site handled combinations that should not have been search destinations.
On 17 September, we entered an invented locality. The visible page said "404! Page Not Found." Its actual response was HTTP 200, with an instruction allowing indexing, and its title described flats for sale in the invented place. The visitor saw an error; a crawler received a successful, indexable page.
There was a reasonable product idea behind other empty searches. If the exact combination had no properties, the site could offer nearby options rather than leave a visitor at a dead end. But a useful fallback for a visitor does not make every invented combination a useful, permanent search landing page.
Google's guidance on filter-driven navigation describes how combinations can create a very large URL space and recommends a proper 404 response for combinations with no results. Search Console had separately classified 86,776 site URLs as "soft 404" on 4 September. We did not treat that total as a count of verified filter pages.
These observations gave us a concrete system to fix. They did not reveal Google's internal reason for the 18 August decision, and we cannot claim the spam update specifically targeted this site's filters.
The filter pages had another job
The loss of filter traffic was measured. A second concern emerged when we asked how Google reached individual properties.
None of the 16,136 URLs in the publicly declared sitemaps on 16 September was an individual property page. In 40 individual property URL inspections, all 40 reported a filter page as the referring page; none reported a referring sitemap. That sample shows the route Google had used for those properties. It does not prove no other route exists.
The route also had a narrow doorway. One filter page offered 1,457 results across 49 pages. Its server-delivered page exposed ten ordinary links to results: four properties and six projects. More cards appeared when a visitor used the page, but the page-number controls did not carry web addresses that Google could follow. A page-two address could be loaded directly, yet it identified page one as its official version.
Google's pagination guidance calls for distinct page addresses, links between them and a canonical address for each page. Its link guidance explains why a clickable control without an href is not a dependable discovery route.
This did not mean property pages had vanished from Google. In our 40-page sample, 35 were indexed. A filter page can also be crawled even if it is not indexed. The measured concern was narrower: the demonstrated route into changing inventory depended heavily on a page family that had just lost search visibility. Whether discovery of new listings had actually slowed remained to be measured.
Why asking Google to index the pages again was not enough
Several sensible responses had already been tried or were being tested: reducing the sitemap, marking empty pages noindex, improving content on a few landing pages, requesting indexing and trying a trailing slash to make fresh addresses. Some were reported in our discussion rather than independently verified.
Each addressed part of the symptom. None, by itself, decided which combinations should be enduring search destinations or gave Google a dependable route into the properties behind them.
| Attempt | What it could not settle on its own |
|---|---|
| A smaller sitemap | What to do with URLs Google already knew, or which remaining pages deserved search visibility |
| noindex on empty pages | How the site would limit the creation and discovery of new combinations |
| More titles, FAQs or locality copy | Whether the page and linking rules were sound |
| Indexing requests or new slash variants | Whether the underlying issues had changed |
Google says an indexing request does not guarantee inclusion. A noindex instruction tells Google not to include a page, but Google must still visit it to read that instruction. The important work was to define the pages worth presenting to Google and protect access to the inventory beneath them.
What we recommended doing first
Protect property discovery. Give eligible, live properties dependable routes through linked locality pages and real pagination. Give each page of results its own address. Make links between relevant pages followable. Evaluate a regularly updated property sitemap alongside these routes; a sitemap helps discovery but does not guarantee indexing. Set one official address for each property and a clear rule for sold, rented or withdrawn listings.
Decide which filter pages deserve search visibility. The website can still let visitors try many useful combinations. The pages it offers to Google should be a deliberately chosen, manageable set:
| Kind of combination | Decision |
|---|---|
| Real search demand, distinct purpose, dependable stock | Keep and improve as a landing page |
| Overlaps a stronger destination | Consolidate where appropriate |
| Useful while browsing, but not a search destination | Keep the interaction useful without automatically expanding search URLs |
| Impossible, malformed or genuinely empty | Return an appropriate not-found response; redirect only to a genuinely equivalent destination |
The large number of URLs known to Google also made the listing lifecycle worth investigating. We needed to establish what happened to addresses when inventory was withdrawn, and how much of Google's historical URL population came from each page family. The available GSC totals did not establish those shares.
Test before extending the changes. Start with 30–50 filter landing pages that previously earned visits, cover both sale and rental searches, and make sure they have real stock and proper links. Then watch whether Google revisits them, whether they return to search results, whether impressions and clicks recover, and whether properties behind them are discovered. Those are proposed measures, not outcomes we can claim today.
The original fear was about the domain. The investigation found a more specific business loss and a practical order of work. Overall traffic had looked steady because several different parts of the website were moving in opposite directions.
If a search report raises a large question about your website, start with a diagnosis before choosing the fix.
How we checked
The indexed-page series comes from Search Console through 4 September 2026. Click comparisons use equal 22-day windows, 25 July–15 August and 22 August–12 September, with identical weekday composition and uncapped daily totals. We inspected 40 individual property URLs on 15 September and tested the live filter and pagination behaviour on 16–17 September. The missing page family was identified by elimination; Google did not supply an itemised list of removed URLs. The proposed recovery work had not yet been measured as an outcome.
Frequently Asked Questions
- How do you tell a domain-wide penalty from a page-level indexing problem?
- Look at whether the loss is spread evenly or concentrated. A domain-wide problem shows up across every page family at once. Here, individual property pages and project pages kept their indexed population and grew their clicks by 30.1% and 7.8% over the same window in which filter landing pages fell 98.4%. Losses that sit inside one page family point at what that family has in common, not at the domain.
- Why did whole-site traffic barely change when thousands of pages were deindexed?
- Because different parts of the site were moving in opposite directions at the same time. Filter landing pages went from 2,443 clicks to 40 across matched 22-day windows, while property and project pages grew by roughly 2,400 clicks between them. The net was −0.4% at the whole-site level. A site-total view would have shown nothing worth investigating, which is exactly why the loss went unnoticed until the indexed-page count was checked.
- Should a site return a 404 for filter combinations that have no results?
- Google's guidance on faceted navigation recommends a proper not-found response for combinations with no results. The failure mode we found was the opposite: an invented locality returned a page that looked like a 404 to a visitor but responded HTTP 200 with an indexable instruction and a title describing properties in the invented place. The visible error matters far less than the response code and the indexing instruction underneath it.
- Does requesting indexing in Search Console bring deindexed pages back?
- Google states that an indexing request does not guarantee inclusion. In this case requests had already been tried without effect. A request asks Google to look again; it does not change what Google found the first time. If the page and linking rules that led to the removal are unchanged, another look tends to reach the same conclusion.
- Can adding noindex to empty pages fix an oversized URL space?
- Only partly, and slowly. A noindex instruction tells Google not to include a page, but Google has to visit the page to read the instruction, so the URLs still have to be crawled. It also does nothing about the mechanism generating new combinations. Marking pages noindex treats the pages that already exist; it does not decide which combinations should exist as search destinations in the first place.
- Why do filter pages matter beyond the traffic they earn directly?
- They are often the route by which Google discovers the inventory underneath them. In 40 individual property URL inspections here, all 40 reported a filter page as the referring page and none reported a referring sitemap. When that page family loses search visibility, the direct click loss is measurable immediately, and the discovery consequence for new listings shows up later.
- Is a clickable pagination control enough for Google to follow?
- No. Google's link guidance is explicit that a control without an href is not a dependable discovery route. On the filter page we inspected, 1,457 results were spread across 49 pages, the server-delivered page exposed ten ordinary links, and the page-number controls carried no href. A page-two address could be loaded directly but named page one as its canonical, which removes it as a separate destination as well.
- What is the first thing to check when indexed pages fall sharply?
- Which pages disappeared, not how to get them back. Search Console does not hand over a list of removed URLs, so the work is to compare page families against each other over matched date windows: indexed population, clicks, and whether anything else on the site depended on the family that went missing. The recovery plan follows from that answer and is usually different from the one the initial panic suggests.