SEO Weekly: Spam Update Closed, JSON-LD Got Less Forgiving

Four different systems moved this week, and they require four different responses. Google completed a global spam update. Its JSON-LD extraction became less tolerant of double escaping. Search Console carried two reporting caveats. Chrome approached a page-lifecycle milestone. None of those facts, by itself, proves that a page gained or lost rankings.

This is the first MAXUOD SEO Weekly, covering August 18–24, 2026. It turns the daily intelligence archive into a short operating brief for Canadian small businesses and the people who maintain their websites. The goal is not to collect headlines. It is to identify what changed, what evidence is still missing, and which owner should act.

Research note: Five dates in this window were reviewed and archived during the operating cycle. The August 19 and 20 records were reconstructed on August 25 from dated public sources. They do not recreate unavailable account state, private dashboards, or social feeds that could not be verified later.

The week in one decision table

Signal	What is verified	What is not verified	Practical response
August 2026 spam update	Google says the global, all-language rollout ran from August 18 to 21.	No new spam policy, affected industry list, or MAXUOD property impact was published.	Annotate the window, then compare stable page and query cohorts before changing content.
JSON-LD extraction	Google said its extractor now applies one HTML-unescape pass.	This was not announced as a ranking factor or rich-result guarantee.	Test the live JSON-LD value and repair the serializer if it depends on double escaping.
Search Console reporting	Google documented a Generative AI impression logging error for August 13–17. A separate Crawl Stats gap was independently observed and later appeared restored.	A reporting gap does not prove lost crawling, indexing, visibility, or traffic.	Keep report availability separate from server activity and business outcomes.
Chrome page lifecycle	Chrome scheduled a broader unload deprecation stage for August 25.	The schedule is not a Google Search ranking update, and planned coverage is not proof of a specific visitor’s browser behaviour.	Replace fragile exit logic and verify form, consent, and analytics delivery.

Google’s spam update is a measurement window, not a diagnosis

What changed

Google completed a global, all-language spam update from August 18 to 21.

Who is affected

Site owners who observe a material change across stable pages or query groups during and after the rollout window.

Evidence

Google’s Search Status Dashboard confirms the dates and scope.

What to do

Annotate the window, wait for stable reporting, then compare like-for-like page and query cohorts.

What this does not prove

That a specific page changed because of the update, or that Google introduced a new spam policy.

Last verified

August 25, 2026.

Google’s Search Status Dashboard records the August 2026 spam update as starting on August 18 and completing on August 21. The incident applies globally and across languages. Google did not attach a new policy definition, target pattern, recovery timetable, or list of affected sectors.

That limits what a responsible site owner can conclude. A ranking change during the same week may be related, unrelated, or partly obscured by reporting noise. The first useful response is to annotate the dates and compare the same important URLs, query families, countries, and device segments before and after the rollout. A single screenshot, a sitewide average, or a community anecdote is not enough.

If a page fell, review whether it still owns a real customer decision, whether its claims are supported, whether another URL now competes for the same intent, and whether a technical release changed access or rendering. If nothing material moved, continue observing. An update name is not a repair instruction.

JSON-LD now needs the delivered value, not the CMS preview

What changed

Google clarified that JSON-LD extraction applies one HTML-unescape pass.

Who is affected

Templates, plugins, or CMS pipelines that depend on repeated cleanup of double-escaped values.

Evidence

Google Search Central’s August 21 notice and the JSON string rules in RFC 8259.

What to do

Parse the live application/ld+json block and repair the serialization owner when a delivered value is wrong.

What this does not prove

That valid markup will create a ranking gain, rich result, AI citation, or enquiry.

Last verified

August 25, 2026.

On August 21, Google Search Central said JSON-LD extraction would perform one pass of HTML unescaping. A value such as a double-escaped ampersand can therefore remain incorrectly encoded instead of being repeatedly cleaned up by the parser.

The correct test unit is the live application/ld+json block. Parse it as JSON, inspect the values Google is expected to consume, and compare them with the visible page facts. If the value is wrong, fix the template, plugin, or serialization layer that produced it. Manually replacing one entity on one page leaves the generator broken.

This change concerns extraction correctness. Valid JSON-LD can help a search system understand eligible structured information, but it does not guarantee a rich result, ranking improvement, AI citation, or enquiry. MAXUOD has a separate live-output testing guide for the implementation details.

Two Search Console gaps demonstrate the same discipline

What changed

Google documented a Generative AI impression logging error, while a separate Crawl Stats gap was independently observed.

Who is affected

Teams using those reports to diagnose search visibility, crawler activity, or AI-search exposure.

Evidence

Google’s data-anomalies record supports the first gap; the second remains a bounded multi-user observation.

What to do

Preserve report availability, system activity, search outcomes, and business outcomes as separate records.

What this does not prove

That Google stopped crawling, indexing, showing, or sending visitors to a particular website.

Last verified

August 25, 2026.

Google’s official data-anomalies page says a logging error affected Generative AI impression data from August 13 through 17. During the same broader period, several practitioners reported missing August 15–16 rows in Crawl Stats; the rows later appeared restored in observed properties, but Google did not list a Crawl Stats incident.

These are not interchangeable signals. The first is an official measurement caveat for a named report. The second is a multi-user observation about another report. A missing report row is not server-log evidence. Neither signal establishes what Googlebot requested from a particular server, what was indexed, what an AI answer displayed, or whether a customer visited.

When a report goes blank, preserve the distinction:

1. Report availability: did the interface or export contain the row?
2. Observed system activity: do server logs, URL Inspection, or another first-party record show activity?
3. Search outcome: did stable pages or query cohorts change?
4. Business outcome: did qualified enquiries or sales change?

For a small-business measurement stack, connect this reporting check to the existing Search Console Generative AI guide and the AI visibility tools decision guide. Those pages own the evergreen measurement intent; this weekly report records the dated operating change.

Chrome’s unload deadline belongs to release QA

What changed

Chrome scheduled a broader reduction in default unload support for August 25.

Who is affected

Sites that use exit handlers for analytics, consent, saved state, forms, or other customer-path work.

Evidence

Chrome’s published unload deprecation plan and the behaviour reproduced on the actual site workflow.

What to do

Replace fragile exit logic and verify the complete customer path in a current browser.

What this does not prove

That Chrome changed Google Search rankings or that every browser visit is already affected.

Last verified

August 25, 2026.

Chrome’s published deprecation plan scheduled the next major reduction in default unload support for August 25. This matters when a site depends on exit handlers for saving state, analytics, consent, or form delivery. It can also affect back/forward cache behaviour.

It does not belong in a ranking-update narrative. The useful work is to locate unload listeners, move appropriate state-saving logic to supported lifecycle events, and test the real customer path in a current browser. A tag firing in preview mode does not prove the event reached Analytics, the form endpoint, or the CRM.

What MAXUOD would do this week

For a Canadian SMB, we would keep the response deliberately small:

• Annotate: record the spam-update dates and reporting anomalies beside the affected reports.
• Inspect: validate live JSON-LD on revenue-adjacent templates, not every URL indiscriminately.
• Verify: test one complete consent, form, and analytics path in a current Chrome build.
• Observe: compare stable page/query cohorts after enough post-rollout data exists.

The MAXUOD contribution is the evidence card behind each action: event date, affected surface, page owner, observed result, unknowns, decision, and recheck date. That keeps an official update, a parser defect, a report anomaly, and a conversion failure from being compressed into one vague “SEO issue.”

What we are not changing because of this week

We are not rewriting pages because a spam update completed. We are not adding schema types merely because parsing changed. We are not reporting a blank Crawl Stats row as a crawl outage. We are not calling a browser lifecycle change a ranking factor. And we are not creating four new blog URLs for four events that already have clear intent owners.

The week’s durable lesson is procedural: classify the system before choosing the fix. Rankings, parsers, reporting interfaces, browsers, and conversion endpoints can fail independently. Good SEO operations preserve that separation long enough to find the actual owner.

Platform note: Sources were rechecked on August 25, 2026. Platform schedules, interfaces, and documentation may change. This briefing is educational and does not promise rankings, traffic, rich results, leads, or a particular outcome.