Improving our news articles, and what we are doing about it

[Mike]

We are removing thousands of low-quality articles published by our automated AI news system.

These articles included cryptocurrency promotions, stock tips, irrelevant press releases, and other stories that had little or nothing to do with Windows.

Nothing written by forum members will be deleted. Replies and discussions are also safe. We also pre-emptively delete tons of spam before anyone even reads it on these forums using anti-spam moderation systems.

What went wrong​

Our AI news system collects stories from sources across the web. Unfortunately, some of those sources were not traditional news outlets. They were press-release services, financial news sites, promotional blogs, and other publishers that frequently insert terms such as “Microsoft” or “Windows” into otherwise unrelated stories.

The system treated those articles as relevant Windows news when they were not.

This resulted in articles about:

• Cryptocurrency promotions and “the next coin to explode”
• Stock-price predictions and analyst ratings
• Routine press releases about obscure companies receiving Microsoft partner status
• Advertisements for cheap or questionable Windows license keys
• Other stories with little practical value for Windows users

A surprising number of these headlines also appeared in Google News, but it was still our responsibility to prevent them from being published on WindowsForum.

What we have done​

We reviewed all 68,628 automated AI news articles published on the site.

A total of 3,157 articles failed our quality review. Of those, 3,139 have now been removed.

The remaining 18 are still here on purpose: a member had replied to each one. We are not deleting those, however weak the article was — the discussion underneath it is yours.

Every deleted article had zero replies. This was an intentional rule, not a coincidence.

Your posts and discussions are safe​

We are only removing articles created by the automated news system.

We are not deleting member posts, replies, discussions, tutorials, or community contributions. If a member had replied to an automated article, that article would not have been removed.

Your conversations belong to the community. We will not erase them simply to improve a statistic or reduce the number of pages on the site.

Preventing this from happening again​

Removing old articles only addresses the past, so we have also changed how the system works.

We have:

• Blocked 124 websites from being used as news sources
• Added 48 new filters to detect cryptocurrency promotions, stock-market content, irrelevant press releases, and similar material
• Tightened the standards used to decide whether a story is genuinely relevant to WindowsForum
• Expanded our review process for automatically generated content

These filters operate before an article is written, which should prevent most of this material from reaching the site in the future.

WindowsForum also contains a large amount of original, useful content, including tutorials, troubleshooting guides, technical discussions, and community answers that have collectively been viewed millions of times.

Why we are explaining this publicly​

In July, Google issued a penalty against WindowsForum related to what it described as “thin content”—pages that provide little original information or value.

We believe parts of that assessment were unfair, and we have publicly explained our concerns here:

On the Manual Action Against /threads/ — WindowsForum

In July 2026 Google applied a partial manual action — “Thin content with little or no added value” — to every URL under windowsforum.com/threads/. The material most likely to have been judged thin has been indexed, unchanged and unpenalised, since 2008. No example URLs were supplied.

windowsforum.com

However, we also have to be honest about the content that did not meet an acceptable standard.

The irrelevant partner announcements, cryptocurrency promotions, stock stories, and similar articles should never have been published. That was our responsibility, and we have corrected it.

What you may notice​

Some older automated news threads will disappear from the forum. Others may gradually stop appearing in search-engine results.

No content is being permanently destroyed. If an article disappears that you believe was useful or worth preserving, tell us in this thread and we will review it and restore it where appropriate.

We also welcome suggestions for improving our automated news system. Community feedback will help us make it more accurate, relevant, and useful.

We want your honest opinion​

Do you believe WindowsForum publishes content that provides little or no value?

A great deal of development time has gone into the site, its tools, its tutorials, and its automated systems. We do not believe search engines should be able to severely damage an independent website based on vague or inconsistently applied standards.

At the same time, we accept responsibility when our own systems fall short.

Questions, criticism, complaints, and suggestions are welcome below.

 

[Mike]

Having difficulty finding the place? There is a write-up about it here:



We are continuing to document more live activities on what we are doing. We are not blocking any of @whoosh songs or art since its in an off-topic area and has intrinsic value of it being a song or music...

In addition you may see additional verification checks to prevent a lot of spam...

You will see for example a PostGuard system when making a new post (a bit ridiculous for now):

I am looking into other ways to improve the service and waiting for an update from Google about their shadow banning manual action practices.

 

[WindowsForum AI]

Mike, the screenshot makes the temporary PostGuard measures clear. The Cloudflare verification plus explicit certifications should discourage basic automated spam, although the wording can be tightened.
Suggested certification text:

• I certify that this post does not contain unauthorized advertising, self-promotion, or spam and complies with the WindowsForum Terms of Service.
• I certify that I am a human user and am not using an unauthorized autonomous system to submit this post.

A few editorial corrections:

• Change “since its in an off-topic area” to “because they are posted in an off-topic area.”
• Change “You will see for example” to “For example, you may temporarily see…”
• “Manual action” is more precise and less contentious than “shadow banning,” even with the latter struck through.
• Consider explaining that PostGuard is temporary or undergoing adjustment so members understand why it currently feels excessive.

The key distinction is sensible: low-quality automated publishing is being removed, while original member-created songs, artwork, and discussions remain protected.

 

[Mike]

Update: forum-wide content assessment project​

We are now testing a review-only system across the guest-visible forum corpus. It uses a fixed 500-item teacher set assessed through our subscription workflow, together with our existing local XFAI similarity data, to estimate:

• whether a post adds useful value;
• whether it is substantively about Microsoft technology or practical IT;
• thread-level thin-content risk;
• named Google spam-policy risk and forum-spam risk; and
• possible near-duplicate candidates.

The full-corpus pass itself runs locally against already stored embeddings and makes no new provider, model, embedding-generation, or API calls. The output is intended to help staff prioritize human review. It does not automatically delete, edit, merge, noindex, warn, ban, or otherwise penalize a member or their content.

This first corpus-wide output is a development estimate trained from 500 subscription assessments and 0 human labels. It is not ground truth or a confirmed Google-policy finding. Likewise, an XFAI similarity match is only a candidate for comparison, not proof that two threads are duplicates. We will validate the results before making any separate remediation decisions.