Google''s 2026 Spam Update and the AI-Content Crackdown: What Actually Got Hit

Last reviewed: 28 July 2026. Written by Prateek, founder of TechTipsTool, from live client data across Indian and global sites.

Two things happened this summer that changed how we brief every content project.

First, Google shipped the June 2026 spam update and finished rolling it out in roughly two days — unusually fast for a spam update, and fast enough that most site owners saw the drop before they saw the announcement. Second, through late July, search professionals started reporting a consistent pattern in what got demoted: pages that were plainly machine-written and never edited by a human were being treated the same way Google treats thin, scraped, or doorway content.

That second part matters far more than the update itself. A spam update is an event. The AI-content signal is a policy direction.

This is what we saw, what we changed, and the audit we now run on every content site we manage.

What the June 2026 spam update actually targeted

Spam updates are not core updates. A core update re-weighs quality signals across the whole index. A spam update enforces the spam policies — it goes after specific manipulative behaviours.

From the sites we monitor, the June rollout hit four clusters hardest:

Pattern	What it looks like	Typical damage
Scaled content abuse	Hundreds of near-identical pages generated from a template or a prompt loop	Whole directories deindexed
Site reputation abuse	A trusted domain renting subfolders to third-party commercial content	Section-level demotion
Expired domain abuse	Buying an aged domain and repurposing its authority for unrelated content	Sitewide collapse
Low-value "answer farms"	Pages that restate the query and add nothing	Gradual impression decay

Note what is not on that list: using AI at all. Google has never said AI-assisted writing is against policy, and it repeated that again this year. The policy line is scale without value, not tooling.

The part that surprised people: thin-AI treatment

Through July, reports surfaced that Google appears to be classifying unedited AI output as thin content — not as "spam" with a manual action, but as low-value pages that quietly stop earning impressions.

That distinction is important for diagnosis:

• A manual action shows up in Search Console under Security & Manual Actions. Loud, obvious, appealable.
• Algorithmic thin-content treatment shows up as nothing at all. No message. Your pages simply stop being served, and your impressions curve bends down while your published count goes up.

If your content volume rose in Q2 and your impressions did not, you are probably in the second bucket. That is the most common shape we have diagnosed this month.

Google also debunked a stack of "AI SEO" tactics

Alongside the volatility, Google publicly pushed back on several tactics being sold as AI-search optimisation in 2026. Worth internalising before you pay anyone for them:

• There is no "AI Overviews schema." No markup type makes you appear in AI answers.
• Writing in "LLM-friendly language" is not a ranking factor. Clear writing helps humans and machines equally; there is no separate dialect.
• Stuffing question-shaped headings does not get you cited. Being the best answer does.
• llms.txt is not read by Google. We still publish one — some AI crawlers use it and it costs nothing — but do not sell it as a Google ranking lever.

The through-line: Google keeps saying that optimising for AI search is the same work as optimising for search. Our results agree.

Then July 23–27 happened

On top of the June update, there was a sharp, unranked burst of volatility from Thursday 23 July through the following weekend. No confirmed update. Rankings moved, settled, and partially reverted.

Our rule for unconfirmed volatility has not changed in five years, and it saved two clients from expensive mistakes this month:

Do not react to a ranking move that is less than seven days old. Measure, do not amputate.

Most of what looks like a penalty in the first 72 hours is Google re-testing. If you rewrite your top pages in a panic on day two, you lose the ability to tell whether the recovery was the algorithm or you.

The audit we run on client sites right now

This takes about half a day for a 200-page site. Run it in this order.

1. Segment by intent, not by date

Export every URL from Search Console for the last 16 months. Split into four buckets: money pages (services, pricing), proof pages (case studies, portfolio), informational posts, and everything else. Compare each bucket''s clicks and impressions for June–July 2026 versus March–April 2026.

A drop concentrated in the informational bucket is a content-quality signal. A drop across money pages is usually technical or a core-update quality re-weighing — different fix entirely.

2. Flag the pages you cannot defend

For each informational page, answer one question honestly: what does this page contain that a language model could not produce without our business?

Acceptable answers: our own screenshots, our client numbers, our pricing, a process we actually run, a mistake we actually made, a photo we actually took, an opinion we will defend on a call.

Unacceptable answer: "it covers the topic thoroughly."

Anything with no defensible answer goes on the rewrite-or-remove list.

3. Decide: improve, merge, or delete

Situation	Action
Ranks, earns clicks, but is generic	Improve — add first-hand evidence, keep the URL
Three pages cover the same query	Merge into one, 301 the rest
No impressions in 6 months, no strategic value	Delete and let it 410
Ranks well and already has original substance	Leave it alone

Deleting content still feels wrong to most owners. It is often the highest-ROI hour in the whole audit. A smaller site of pages you can defend outranks a bigger site of pages you cannot.

4. Add evidence, not words

Rewriting does not mean lengthening. On the pages we recovered fastest, we added:

• A real number from a real project ("cut LCP from 4.1s to 1.6s on a 90-page Shopify build")
• One original image or screenshot per major section
• A named author with a real bio and a real face
• An explicit "last reviewed" date, updated only when we actually reviewed it
• Links out to primary sources — Google''s own documentation, not a blog quoting a blog

5. Fix the author and entity layer

Thin-content treatment correlates strongly with anonymous publishing. Every post on a site we manage now carries a byline linked to a real author page, and the author page carries Person schema tied to the organisation. This is not a magic trick; it is the cheapest available signal that a human is accountable for the page.

Safe versus risky AI workflows

We use AI daily. Here is the line we draw internally.

Safe	Risky
Research synthesis and outlining	Prompt → publish, unedited
First-draft sections a human then rewrites with real data	Bulk generation of location or service permutations
Rewriting your own notes into clean prose	Rewriting a competitor''s article into "unique" prose
Generating schema, alt text, and meta from finished content	Auto-publishing on a schedule with no editorial gate
Translation with a native-speaker review	Machine translation at scale, no review

The test is not "did AI touch this." It is "would this page exist, in this form, if the tool did not exist?" If the honest answer is no, and nothing about your business is in it, it is scaled content abuse regardless of quality.

Recovery timeline: what to actually expect

Setting expectations is half the job.

• Weeks 1–2: Audit and cull. Expect impressions to dip slightly as removed pages drop out. This is normal and healthy.
• Weeks 3–6: Rewrites go live in batches. Google recrawls. Some pages recover within one crawl cycle.
• Weeks 6–12: The bulk of algorithmic recovery, if the changes were substantive.
• Next spam or core update: Full re-evaluation. Some sitewide classifications only lift when the system refreshes.

Anyone promising recovery in a week is guessing. Anyone telling you to wait passively is also wrong — the refresh only helps if you gave it something new to evaluate.

What we changed in our own process

Three concrete things, in case they are useful:

1. No post ships without one original artefact. A screenshot, a number, a chart, a photo — something that did not exist before we made it.
2. Publishing cadence dropped, editing time rose. We publish fewer posts and spend the recovered hours on evidence and internal linking.
3. Every post gets a review date on the calendar. A post reviewed in July 2026 and stamped as such outperforms an identical post with no date.

If you want that applied to your site rather than described, that is exactly what our SEO and content service does, and the pricing is public. You can also see the shape of the work in our portfolio.

For the adjacent question — how to be cited inside AI answers rather than just ranked in blue links — read our companion piece on AEO and GEO in 2026.

FAQ

Was the June 2026 spam update a penalty?

Not in the manual-action sense. Spam updates are algorithmic. Unless you see a message in Search Console under Manual Actions, nothing was applied to your site by a human reviewer — your pages were simply re-scored.

Will Google penalise me for using AI to write blog posts?

No. Google''s stated position is that it rewards helpful, original content regardless of how it was produced. What gets demoted is content produced at scale with no added value — which unedited AI output very often is.

How do I know if my site was hit by thin-content treatment?

Compare published-page count against impressions over the last six months in Search Console. If pages went up and impressions went flat or down, and there is no manual action, thin-content treatment is the likely explanation.

Should I delete my old AI-written posts?

Delete only the ones with no traffic and no strategic value. Pages that still earn impressions are worth rewriting with real evidence — you keep the URL history and the existing signals.

How long does recovery take after a spam update?

Individual pages can recover within a crawl cycle of being genuinely improved. Sitewide classifications often need the next update refresh, which historically means one to three months.

Does publishing an llms.txt file help me rank on Google?

No. Google has said it does not use it. Some AI crawlers do, so it is a cheap thing to publish, but it is not a Google ranking factor and should never be sold as one.