|
SEO updates you need to know
Our sponsor: Citation Labs
Shape the Future of Link Building!
🔗⭐ Take the 2025 State of Link Building Survey:
Share your insights to uncover AI usage trends, impact measurement challenges, and ROI strategies.
💡🧠 Get Smarter:
Refine your campaigns and reporting strategies with benchmarks and actionable data.
🤝 Give Back:
For every completed survey, Citation Labs donates $10 to Norfolk Clubhouse.
⏳ Deadline: 1st March 2025
|
Search with Candour podcast
Pagination best practice & Google's secret weapon against spam
Season 4: Episode 8
In this episode of Search with Candour, Jack Chambers-Ward and Mark Williams-Cook discuss best practices for SEO pagination, a Google patent that targets spam tactics and a theory by Tabby Farrar about gaining links via citations in LLMs.
|
|
|
This week's solicited tips:
GSC filters will hide some of your data
Where did the over 77,500 clicks go? When you apply a query filter to Google Search Console data, Google immediately hides a large percentage of clicks for 'privacy' which are not included in totals - not many people seem to know this!
😎 In this example, GSC reports 148,000 total clicks.
✅ If I filter to just clicks that include the brand, I get 39,700
❎ If I filter to just clicks that exclude the brand, I get 30,800
😕 30,800 + 39,700 = 70,500 in total.
🤯 148,000 - 70,500 = 77,500 "missing" clicks
This is important to keep in mind: You can use filtered data to get an idea for a trend, e.g. "branded clicks are going up/down", but Google won't give you the precise data.
Unfortunately, this is not caused by the "1,000 row limit" in GSC, even when connected to Big Query, you will still find queries missing as a result of "Not provided".
In the worst instances, 🔎 Charles Meaden has seen as much as 80% of data missing in the personal finance space💀
|
Classical programming vs machine learning
Do you need deterministic answers, or do you need a set of rules for making approximations? This is the framework of "classical programming" vs machine learning ⤵️
If you have a task that requires deterministic output, such as you want to generate some schema from product data in a database, then classical programming is your tool ⛏️
If you have a task where rules to provide predictions would be helpful, such as mapping redirects, then a machine learning approach may be suitable. 🔨
⚠️ LLMs like ChatGPT are themselves a subset of machine learning, and one of the biggest issues in SEO I see currently is trying to use them for everything and ending up in a mess.
Here's a nice diagram I remade from the book "Deep Learning with Python" by Francois Chollet that describes it well.
|
Don't focus too much on search volume and rankings
Stop worrying so much about monthly search volume and specific keyword rankings.
When I polled people, the #1 reason they wanted this data was for “forecasting” - but it creates a whole set of problems for you:
🔭 Strategic fixation: If you determine you’ll get “50,000 traffic” from “these 62 keywords”, you’ll likely find your team blinkered trying to rank for just those terms, when there is lots more opportunity to achieve business goals outside of this.
📈 Keywords with high visibility volume tend to fall into the head/middle area, which naturally makes them far harder to rank (less clear search intent, more competition from big sites), so you are actually setting yourself a steep curve to climb and ignoring easier wins.
🥇 Ironically, sometimes the best way to rank for big terms is not to directly target them, and gain topical authority on the edges of the topic and move your way to the centre, not start from the outside.
💸 These search terms make the minority of the playing field, so you are betting your budget on a high-difficulty, high-risk investment.
⚔️ Even if you defeat all of the above and you do rank for that 20k per month search term, I see almost nobody forecast for how they are going to keep that ranking. Do you think the site that ranked #1 for years is now going to say “oh well, it was nice ranking for that term” and give up? No, they will also double their efforts in a time you are leading by a knife edge.
All of the above is why we’ve been using AlsoAsked heavily in our content and research plans for years 😇 And here's a video of me explaining it.
|
Search volume shouldn't make your topic choice for you
Choosing not to write about something because a search term as "no volume" is a big mistake from both a user and SEO point of view ⤵️
⚖️ Search intent volume is more important than keyword volume
🏃 Take the example "buy blue running shoes" - any keyword tool would tell you in the UK this has zero monthly searches. When a tool tells you this, it likely means "no data" and there is, in fact, some searches.
👬 When you consider the multiple ways that people can search for the same thing e.g. "buy blue running shoes uk", "uk buy blue running shoes", you can very quickly work into 100s of monthly searches that go below the radar
🤖 Google cares less about exact keywords and more about answering user intent, so a page could potentially rank for many of these queries, making it very worthwhile indeed.
👀 Your competition probably overlooked them too!
This is again one of the reasons we have been using AlsoAsked data for almost a decade at Candour 🤩
|
What we found in the AlsoAsked Google data exploit
We identified Google uses 8 rq_semantic_query_class categories from our Google exploit - "SO WHAT? How is it useful?" ⤵️
☝ Quick recap: Data directly from Google shows alongside many other classifiers, labels, scores and cateogies, there exist 8 classes they put a lot of queries into, which are:
⭐SHORT_FACT e.g: "how much does abiraterone cost in the uk"
⭐OTHER e.g. "what do chefs say about air fryers"
⭐COMPARISON e.g "curtain wall system vs window wall system"
⭐CONSEQUENCE e.g. "what happens to asparagus if you let it grow"
⭐REASON e.g. "why was abilify taken off the market"
⭐DEFINITION e.g. "what is a birthday costume"
⭐INSTRUCTION e.g. "what's the best way to cook an artichoke"
⭐BOOL e.g. "can i become a agile coach with no experience"
So, how is this information useful to us as SEOs?
1️⃣ Firstly, we know for a fact this is how Google is classifying queries. When we are classifying queries as things like "navigational, transactional" etc, this may be a great guess, but it's just another step and abstraction away from how Google is actually working. With actual data, you can look for actual links.
2️⃣ I recently had a friend investigating traffic drops of a website, and as part of the analysis, they ran our classifier (in comments) and found that 100% of their tested 1k queries they lost traffic for were either SHORT_FACT or OTHER, which demonstrates a clear correlation and great factual place to start the analysis.
3️⃣ These categories will be correlated to which SERP features are present. I am currently working on a data piece with Tomek that will show how likely AIOs are to appear for each query class. If you know you're losing traffic to AIOs, you can predict at scale, where they will appear and adjust your content strategy accordingly and allow competitors to waste their time and energy.
You can use our free Refined Query Semantic Class Classifier model that was trained on 4.6M English queries directly from Google.
This is the only model trained on the actual data 😎
|
Refer subscribers and earn rewards!
Top Core Updates referrer leaderboard
A big thank you to our top referrers, who have signed up over 10 people to the Core Updates newsletter, go follow them!
🥇 MJ Cachón Yáñez (LinkedIn / Bluesky)
🥈 Nikki Pilkington (LinkedIn / Bluesky)
🥉 Lidia Infante (LinkedIn / Bluesky)
🏅 Saijo George (LinkedIn / Bluesky)
|
|
|
|