How UN-Helpful Content Sites are penalized by Google?!
Information barrier - communication and feelings
There is a lot of excitement about the impact of the updates. An entire industry feels misunderstood and/or unjustly mistreated.
I feel for the individual website owner who has done their best and now has real problems.
Having been more of a programmer than an SEO for the last 10 years, I can relate to a lot of this because in any large company, similar problems exist between IT departments and other departments that have little expertise in computer technology, programming, etc. The lack of understanding and excitement arises in everyday life when, for example, the new headset does not work immediately because the user is not allowed to install the software/drivers for it themselves. There are usually no explanations as to why the employee is not allowed to do this.
Despite justifications and explanations from the IT departments, the employee is left with a feeling of resentment and that the IT staff are stupid. Communicating guidelines with simplified justifications/explanations would provide clarity here.
Now to Google:
Google's communication on search algo updates is good, with guidelines, announcements and communication, but tends to be paraphrased and contains little justification. This is probably intentional, as even a hint would reveal too much about the technical implementation.
Google sets an information barrier here so that not everyone starts researching like crazy. Historically, Google has also had bad experiences with this and the SEO industry quickly found the next lever.
Introduction to how Google works
Simplicity, therefore the Google search explained in 3 simple steps.
Step 1:
A search result is a list created in order by a numerical value (score) and not sorted from A to Z, Z to A by letter.
Put simply, every list of results is a competition that you have to win!
Step 2:
The numerical value, called a "score", is created from a large number of other numerical values. These other values are evaluations that can be positive or negative and only evaluate one part, e.g. links.
Step 3:
Mathematics ensures that all these numerical values result in a "score" and that the appropriate website is displayed in its position. See the diagram on the side.
What have you learned?
To output a search result, everything is converted into a single score.
Google is math and everything is calculated using numbers.
What we technically know about the updates so far
How does Mashine Learning works?
This rating is created using machine learning, which is simply pattern recognition. Here a Image Example.
Pattern recognition must be trained with positive examples and variants.
This is a B, that is also a B, another B, etc. (see the "sliding b's images ")
The more different variants of an image of B the machine receives during training, the better the recognition of the letter B will be. This is how the machine learns what a B looks like in a picture.
But this also means that "machine learning is only as good as its training data".
Disclaimer: To all software engineers, yes this is oversimplified. Please bear with me.
Once the ML has been trained,
you can ask: "Hey, I have an image of a letter here, what is it?"
Answer ML:
"With a probability of 68% it is a B, it could also be a D 16% of the time or a dot 8% of the time. "
You then have to make decisions based on these results. Is 68% probability good enough? Over 50% should be enough, right?
It always becomes difficult when the probabilities are scattered, e.g. 34% B, 33% D and another 33%: DOT.
What then? How do we decide now?
You can see the difficulty in this simple example of machine learning.
ML (Mashine Learning) in production - problem?!
Recognition only gets better if you have more training data and create a new version of the machine and put the new version into productive use. (By the way, the same applies to AIs like ChatGpt, newer and newer versions are trained, improved and replaced).
A trained machine that is in productive use, does not learn!
("It's clear that the machine is not learning lol." from a Tweet)
Now you know how machine learning works, albeit in a simplified way.
ML(Machine Learning ) for HCU (Helpfull content update) is of course not image recognition, but text.
Explaining natural language processing and entity detection here would be going too far.
Machine Learning for Google HCU is therefore:
- Create training data
- Training takes place
- Machine can recognize patterns
- Decision-making processes take place.
Manual selection and a pre-selection based on data can be assumed.
For example, it would make sense to select pages from a category with good ratings from the Search Quality Rater (Here the link for the Search Quality Rate Guideline. ).
Criteria are sufficiently described in Google's documentation on Helpful Content.
The more you match the pattern of the training data, the higher the probability.
Know Decision processes:
HCU 2023 - simple, hard decision: if probability over x percent then devalue entire site
March 2024 Google has switched to a softer version that evaluates individual pages, devalues 2-3 positions or devalues by 20 - 30 positions. (see documentary).
What measures help with a HCU devaluation?
Play by the rules, here are the rules.
Added value and user interaction
Added value is the magic that Google wants to see. Write something that the AI can't write, e.g. at an event ("it smelled like ...") or a product test ("the surface felt ...")
Be authentic, with your own opinion. Your own experiences and adventures.
A good website needs LOVE.
This sound simple, but with Machine Learning and AI, it is possible to detect the differences.
Category Check
Check whether your text is categorized correctly.
Google Natural Language Detection / API is recommended here.
- Paste the text into the form and submit.
- First, all recognized terms appear with categories (so-called Named Entities).
- The Categories tab disappears a little. Click on it and you will see the recognized categories
LINK https://cloud.google.com/natural-language#demo
Match of
- category on the website
- Recognized category
- Targets of the text
This all must fits together. We use a similar API internally for texts to realize exactly this matching.
Page Quality Rating
The rules for Google Quality Rater are online (https://static.googleusercontent.com/media/guidelines.raterhub.com/en//searchqualityevaluatorguidelines.pdf ).
Work through it yourself and have an external person carry out the page quality rating and then make improvements.
If a Google quality rater looks at the website, the result must be a better rating.
Why does a Quality Rater look at the website?
Logic, a website that has been downgraded as unhelpful and has changed a lot, a new look and quality rating would make sense from Google's perspective, right?
Overall Clarity
Your domain (mytraveltime.domain) - category and texts must fit together, e.g. it is strange if a travel blog suddenly has a product comparison with refrigerators.
Internal links and backlinks must also be considered, e.g. where is a link and in which environment - this cannot be done manually.
This analysis is carried out automatically with Upward and interested parties are welcome to join the waiting list. LINK
Why are online stores not affected?
- There is a clear transactional intention behind this, the user is being helped.
- Category - product and text match.
- Content is often the same or similar anyway because of the product descriptions.
- The visitor can interact and buy a product, leave reviews, use support, etc. .
- Generality of brands
- Many online stores allow reviews of products.
Over 7 months and nothing has happened?
Why can't I get my rankings back?
As hard as it sounds, your website obviously doesn't fulfill the pattern enough for the ML score to dismiss the website.
Depending on the topic, the pattern that needs to be fulfilled can be very demanding. Look at other websites in your field.
As mentioned above, backlinks and internal links are also an issue. We recommend taking a very close look here or using our software to analyze links with machine learning.
Disavowe links. There are some discussion about this topic
Look at links, it looks like a new rating for links has been introduced. Link rating is no longer as simple as the Pagerank or DR concept.
But this is a topic for a complete own article.
Summary of this article
- You have learned a simplified scoring for the results list and how HCU negatively influences the scoring.
- You have learned how machine learning works in general and how Google uses it.
- 6 measures to remedy a HCU devaluation.