Google's PDF blackout: only indexed PDFs are recovering

Four domains, 40 top PDFs, URL Inspection, SERP check and page analysis: our Search Console data on the PDF blackout of August 2026.

On August 6-7, 2026, PDF clicks on three of four domains whose Search Console data we analyzed dropped by 92%. No relaunch, no changed robots directive, no manual action. The domains span different verticals, including real estate and public transit. Aggregated across all four, clicks fell from roughly 94,000 to roughly 50,000 (-47.0%), but that number obscures the severity: Domain 1 alone accounts for 82% of the PDF clicks and lost only 37%. Domains 2, 3, and 4 lost between 87% and 98%.

Five weeks of analysis followed: URL Inspection on each domain's ten biggest PDF losers (40 URLs), a SERP sample across 60 queries via DataForSEO, and a page analysis of 16 HTML landing pages with Playwright. The central finding: in our data, recovery correlates not with ranking but with the index status before the blackout.

The break: August 6-7, 2026

The breakpoint falls on August 6-7. Stable PDF clicks for weeks, then a cliff with no transition. Google had announced no update between July 15 and that date.

Domain 3 lost 97.6% of its PDF clicks (pre: Jul 6 to Aug 2 vs. post: Aug 10 to Sep 6), Domain 2 lost 91.6%, Domain 4 lost 86.7%. Domain 1 was the outlier at just 37.1% loss because a large share of its PDF inventory remained indexed.

The break hits PDF URLs exclusively. HTML pages on the same domains showed no comparable movement. The affected queries are consistently document intents: templates, forms, timetables, guides. Users searching for a cancellation template or a transit map want a downloadable document. Google removed these documents from the results on three of four domains overnight, across verticals.

Deindexed, not just lower-ranked

We checked each domain's ten biggest PDF losers via the URL Inspection API, 40 URLs in total. 25 carried the status "Crawled -- currently not indexed." Google knows these URLs, crawled them, but decided to stop including them in the index. The last crawl fell between July 27 and August 10. The PDFs were removed from the index without any change to the files themselves.

This is not a ranking loss where a page slips to worse positions. For the affected queries the page simply no longer exists in the index.

The PDF impressions confirm the picture: aggregated, they fell by roughly two-thirds (-65.2%), while HTML impressions remained stable (+1.6%). A pure CTR problem would look different.

The 15 PDFs that retained the status "Submitted and indexed" show no common trait. File size, age, and domain do not obviously correlate with staying in the index.

A caveat on index status: On September 8, 2026, Google confirmed a brief reporting bug in Search Console in which indexed pages temporarily showed as "Crawled -- currently not indexed." Our URL Inspection ran on September 13 via the API (stored index status, not a live test), five days after the glitch. According to SER the bug lasted about an hour and affected the reporting view, not the stored status. A further undocumented reporting error cannot be entirely ruled out.

Why HTML did not absorb the PDF demand

The HTML recovery rate is 11.7%: only a fraction of the lost PDF clicks landed on HTML pages of the same domain. Nearly 85% of the demand simply disappeared.

We analyzed 16 HTML pages that could have served as fallbacks. Not a single one contains the template itself as text. A user searching for "cancellation template" lands on an advice article about notice periods, not on the template. On 12 of 16 pages the download link to the PDF sits below the fold. In 7 of 16 cases Search Console maps a different HTML page to the query than the one that actually links the PDF. Four PDFs on Domain 1 are fillable forms with 6 to 88 form fields, a function that does not exist in HTML.

Even the closest match in our sample reached only 8.9% recovery. Users want the fillable form itself; the page describes it but does not replace it.

Who ranks in the top 10 now

On September 13, 2026, we checked 60 of the affected queries via DataForSEO (desktop and mobile). 85% had a PDF in the top 10 at the time of checking, 65% in the top 3. Of the 60 original PDFs, 27 were in the top 10, 20 of them at position 1. The distribution is highly uneven: all 15 for Domain 1, seven for Domain 3, four for Domain 4, a single one for Domain 2.

Google has not abolished the PDF intent.

The demand persists, but the source has shifted. Where the original PDF is missing, competitor portals (22.2% of top-3 slots), legal-template sites (14.4%), and unrelated PDFs have filled the gap. The original domain held 27.8% of the top-3 positions.

An AI Overview appeared in 60% of the queries. Google had confirmed the dynamic expansion of AI Overviews in late August. 25 of the 60 AI Overviews cited the affected domain, in some cases referencing a PDF that had simultaneously been removed from the organic index.

Recovery from September 4: preliminary and selective

From September 4, the Search Console data shows early recovery signals. The following numbers are preliminary: data as of September 12, dataState "all," not yet finalized.

In our preliminary data the index status before the blackout correlates clearly with whether a PDF comes back. Of the 40 most important lost PDFs, 15 had retained the status "Submitted and indexed." 14 of those show return: 10 with at least 50% of their previous impressions, 2 partially (10-50%), 2 with first impressions below 10%. Of the 25 PDFs with "Crawled -- currently not indexed," not a single one has returned.

This is a strong correlation, not proof of Google's mechanism. The returned PDFs all carried the status "Submitted and indexed," and their last crawl fell between September 8 and 12. However, several already had impressions before Google re-crawled them. Whether a fresh crawl helps recovery or merely follows it cannot be determined from the data.

The recovery is unstable: on September 11 the number of visible PDF URLs on Domains 2 and 4 halved again. Domain 1 is recovering most clearly, with 8 of 10 top losers preliminarily back to at least 50% of pre-levels. Domain 2 shows no recovery of the lost PDFs; the uptick there comes from new, seasonal documents.

Where PDFs return, HTML loses. On Domain 1 the HTML clicks for the affected queries dropped by 49.4%, on Domain 4 by 55.1%.

The demand was never meant for the HTML page and migrates back once the PDF reappears.

What the community reports and what Google says

The PDF phenomenon has not been officially confirmed. The Search Status Dashboard lists only the August 2026 Spam Update from August 18-21 (SEL) in the relevant window, described by Google as unrelated to PDFs. No crawling or indexing incident was reported. John Mueller responded to a community report on SER with "I'll take a look," without follow-up.

Barry Schwartz documented the earliest cases on SER from August 25; Savanna Gray (Amsive) had flagged dramatic PDF click drops on LinkedIn, and Lily Ray picked up the post. SER's September report summarized it as "temporarily show fewer PDFs." Kevin Indig's Growth Memo put the window at "served fewer PDFs for 2 weeks."

Not every reported PDF drop had the same cause. Artur Ferreira showed on LinkedIn that his own drop was caused by a self-set X-Robots-Tag: noindex. In late August, US government sites fell out of Google search due to bot blocking. Our break on August 6-7 predates the earliest dated US report of August 12. The cause remains open.

What you should check now

If you manage PDF-heavy sites and have seen traffic drops on PDFs since early August, start by ruling out your own causes: check your PDF HTTP response headers for an unintended X-Robots-Tag: noindex, review robots.txt and CDN configuration for crawl blocks. As Ferreira's case shows, a forgotten noindex header produces the same cliff as a Google-side drop.

Then the checks from our analysis:

1. Run URL Inspection on your most important PDF URLs. The status "Crawled -- currently not indexed" is the strongest indicator in our preliminary data. PDFs with this status have not returned in any of the 25 cases we observed. For PDFs showing "not indexed," a recrawl request in Search Console is a reasonable test but not a proven fix.
2. Check whether your HTML pages contain the template itself. If the user searches for a template and lands on an advice article, the page answers the wrong question. The template belongs on the HTML page as copyable text or an interactive form.
3. Move the download CTA into the visible area. On 12 of 16 pages in our sample the download link sat below the fold.
4. Check the GSC URL mapping. In 7 of 16 cases Search Console mapped a different HTML page to the query than the one linking the PDF.
5. Ensure stable PDF URLs. No version numbers in file names, redirect old versions via 301 to the current URL, and link back to the associated HTML page from within each PDF.

If you run PDFs, watch the index status over the coming weeks, not the rankings. We continue tracking the recovery data and will update this post as soon as the picture becomes clearer.

Sources

1. "Google Confirmed Indexing Reporting Bug From Last Night" -- Search Engine Roundtable, September 8, 2026.
2. Google Search Status Dashboard -- Ranking Updates 2026 -- Google.
3. August 2026 Spam Update -- Google Search Status Dashboard, August 18-21, 2026.
4. "Google Search Showing Fewer PDF Files. But Why?" -- Search Engine Roundtable, August 25, 2026.
5. "Google August 2026 Spam Update Done Rolling Out" -- Search Engine Land, August 2026.
6. September 2026 Google Webmaster Report -- Search Engine Roundtable, September 1, 2026.
7. Growth Intelligence Brief #24 -- Kevin Indig (Growth Memo), September 8, 2026.
8. "IRS Website Drops From Google Search" -- Search Engine Roundtable, August 26, 2026.
9. "Everyone's PDFs Are Disappearing From Google. Mine Did Too -- I Found Out Why" -- Artur Ferreira (LinkedIn), August 2026.
10. "Google Is Dynamically Expanding AI Overviews for Some Queries" -- Search Engine Land, August 28, 2026.
11. Own Search Console data: 4 domains, comparison window Jul 6-Aug 2 (pre) vs. Aug 10-Sep 6 (post), URL Inspection on Sep 13, 2026, dataState "all" for recovery data from Sep 4.