|
SEO updates you need to know
Sponsor: Semrush
AI visibility in the US & UK: Top brands, top tactics
Our latest AI Visibility Index findings show that the sources and brands that shape decisions in the UK aren’t identical to those in the US. But with the right insights and approach, teams can build for those differences while keeping their brand experience consistent.
Download this flagship study for free and get your blueprint for building consistent discovery. Fuelled by seven months of data across 22 industries and four platforms.
|
Search with Candour podcast
AI search mythbusters
Season 5: Episode 4
Jack is joined by SEOritmo’s André Pitì to bust common AI search myths.
- Why AI visibility metrics are modelled estimates rather than exact numbers.
- How tools infer AI Overview and AI platform volumes.
- Why spikes in direct traffic don’t automatically indicate AI traffic.
- How to investigate direct traffic by URL patterns.
- Segmenting AI-related signals by geography.
And more!
|
|
|
Last week's SEO tips
Strawberry factors forever
Google's "Generative summaries for search results" patent (US11769017B1) lists the following things as measures for choosing which search result documents are used in LLM summarises:
Query-dependent factors and a language measure.
🏆 Ranking position
🖱️ Selection rate for that query (CTR)
🌍 Location/language (locality measure)
Query-independent factors:
🖱️ Selection rate over multiple queries (site_quality_score)
🤝 Trustworthiness ("based on an author thereof, a domain thereof, and/or inbound link(s)")
🥳 Overall popularity
⌛ Freshness ("recency of creation or updating")
|
A page in the life
Your web page fed the AI Overview but didn’t get the link… WHYYY? 😭 One of the reasons is in Google’s US11769017B1 patent, “Generative summaries for search results” ⤵️
How links get attached to AIOs:
🥄 Google’s AI summaries are split into pieces, a piece can be “a sentence, a semantically coherent portion of a sentence, or an N character or N words span”.
🎯 A candidate page for a link is chosen either from the pages that fed the LLM answer or from Google running a new search using that piece as a query
✅ They are turned into embeddings Google runs a pass/fail for if the page is a good enough match for that piece of the AIO
🥉 This doesn’t mean the best match, Google is just taking the first page that hits this threshold
In short: Google is checking for a page that matches its AI output close enough to back up its claims.
|
I want to hold your prompts
If your AI visibility tool made up your prompt list from your own website, you're probably tracking prompts nobody actually types. 🫠
We don't get real prompt data from ChatGPT, Gemini or Perplexity. But People Also Ask and Search Console are both built on real searches, and long, conversational Google queries are a decent stand-in for how people prompt LLMs.
Katarina Dahlin's process:
🌱 Run lots of seed searches in AlsoAsked: category, best-of, use-case, brand, brand + question, "store uk"
🧹 Export, de-dupe and get an LLM to sort each question by intent and relevance.
🔎 In GSC, filter queries with a question regex over 16 months. Low impressions are normal, so don't bin them.
🎯 Don't track all 1,000. Lean towards comparison/best-of and transactional, as they're closest to a sale.
🛠️ Split your list: "how to choose..." questions get answered on your own site. "Best X for Y" questions are won by getting into third-party listicles.
Track fewer prompts, but make sure they're real ones.
Full article: https://alsoasked.com/guides/ai-visibility-find-the-right-queries-to-track-with-alsoasked
|
While my ranking gently weeps
You can improve your rankings and visibility by improving your site quality score. You can improve your site quality score by making people search for you specifically. ⤵️
This could mean making a useful named tool, an original research series or a named product.
Google’s patent “Deriving and using document and site quality signals from search query streams” (US9569504B1) says:
“A user may thus ‘vote’ in favour of the quality of the requested documents (i.e., the documents which the user is deemed to have asked for) through issuance of the search query.”
and
“A document that has been assigned points in act 430 may, therefore, get a ‘boost’ (e.g., be scored/ranked higher) for other search queries that match the content of the document.”
This means a query can contribute to a quality signal when the system identifies it as requesting a particular document; that signal can then influence rankings for other relevant queries.
👏 Create demand 👏
(I also told you this when we uncovered site_quality_score from our Google exploit 😉 )
|
Search together
Audit what Google’s questions imply about your brand, not just which keywords you could target. This is what Arnout Hellemans was telling us at SearchNorwichXL.
When Arnout was investigating Heineken, he found regionally recurring themes around product quality, taste and even confusion about what the brand sells.
You can use AlsoAsked and group your branded PAAs by theme and market, then investigate the ones that keep appearing.
Those can inform clearer product pages, FAQs and brand messaging, but a question appearing in PAA isn’t, by itself, evidence of widespread customer sentiment. Run multiple times, see what keeps surfacing.
Here's the video of Arnout's talk.
|
Refer subscribers and earn rewards!
|
|
|