Notes
What we learned building the instrument
Writing on what decides whether AI search cites a page: semantic chunking, extraction, and the shape a passage needs to be quoted.
AI reads your page in pieces. Three free skills show which ones
Three open-source Agent Skills show how AI retrievers read a page, what falls past the 26-passage cap, and which Google, Bing and AI rules a page breaks.
Read it
How AI search reads a page
Chunking, embeddings, extraction, the platform changes that move them, and the week's digest.
What Agent Skills Are and Why They Matter for Marketing Teams
A plain-language guide to agent skills: reusable AI procedures, how to review them and why they matter for search and marketing work.
Your Target CPA Is Now a Target
Learn how Google Ads' recent update affects budget-limited campaigns, optimising toward your Target CPA more strictly and what it means for your conversions.
AI Search in the UK: GEO & SEO Strategy for 2026
UK AI search strategy for 2026: strengthen SEO, publish cite-worthy evidence, structure answers for retrieval and measure AI visibility.
What we built, and what it measures
The instruments, their thresholds, and what each number is allowed to claim.
The agent skills directory, read every fortnight
We mirror skills.sh every fortnight: 153 live skills, our own categories, security audits, and a growth trend a live leaderboard cannot show. Free to read.
Four free tools, and what each one measures
What Vectorscope, Brandscope, Voicescope and Patternscope each read, what they refuse to claim, and how to read the result. Free, no account needed.
A glossary built to be quoted
Why our glossary defines 1,070 terms in five languages, how each entry is made and checked, and the rule that keeps hand-written entries out of every queue.
When the answers move
A daily 0–100 reading of how much AI assistants' answers changed since yesterday. What it compares, how to read the scale, and the four things it cannot tell you.
Why ChatGPT Recommends Your SaaS Competitor
A practical US B2B SaaS audit for AI vendor recommendations: test buyer prompts, close proof gaps, strengthen pages, and measure visibility.
Using the tools
Getting a reading, reading it, and what to do about what it says.
What an assistant says your rating is
Three assistants, twelve review platforms, nine questions. Two refused every number, one produced eight — and handed the same 4.6 to four platforms.
What a template sounds like
Why we count repeated shapes, stock phrases and unchecked claims instead of guessing who wrote a draft — and three defects a labelled set found in our thresholds.
What AI Assistants Say About Your Brand — and What to Do With It
How Brandscope reads buyer-facing AI answers about a brand, compares them with first-party evidence and states the limits of the result.
What an AI assistant reads when it reads your page
Six real pages run through the same chunking and embedding a retrieval system uses. The numbers: passages received, text never read, and what actually differed.
The tools
See it on work of your own
Read the passages an assistant would receive from a page, hear how a buyer would phrase the question, or find out what assistants say about a brand. Free, no account.
Questions
Ask something these notes do not answer
Every question asked is public, and answered in the open. The ones asked often become the next note.