Monthly notes
What ten AI engines actually answered.
On the first of every month we ask ten engines the buying questions real businesses are trying to win — no brand names in the questions, the same instrument each time. These are the aggregate findings. No business is named, and every figure below was computed from committed measurements before a word was written around it.
No note has published yet. The first one goes out after the next monthly measurement — and any month with fewer than three businesses measured is skipped entirely, because a sample that small cannot support a true statement. Until then, here is exactly how these numbers get made.
How the measurement works
- Ten engines, every month
- ChatGPT, Perplexity, Claude, Gemini and Grok, plus the Chinese engines Doubao 豆包, Qwen 通义千问, DeepSeek and Kimi. Most measurement stops at the first five; the Chinese four are where Chinese-speaking buyers actually ask.
- The questions never contain the brand name
- We ask what a buyer would type — “best X in Y” — never “what is acme.com?”. A branded question only measures whether an engine can read a website, which any site put up an hour ago passes. It says nothing about whether the business gets recommended.
- The instrument never changes mid-series
- Each data point records the exact spec it was taken under — how many questions, how many engines, how many repeats. Points taken under different specs are never silently compared, and a point measured before a fix is excluded rather than quietly averaged in.
- Figures are computed before the prose is written
- Every number in a note is calculated from committed measurements first; the writing is then constrained to those figures, and a note containing any number that is not in them is rejected rather than published.
- No business is named
- These are aggregate findings across the businesses measured that month. Client names, sites and individual scores never appear here.
No provider can guarantee an AI recommendation — AI answers are non-deterministic and personalized, and the same question can be answered differently on two days. Measurement raises probability and proves movement; it does not promise placement.
Google preferred sources
Make Avowd a preferred sourceGoogle lets you name the sources you want more of. Choose Avowd and these notes carry a “preferred” badge when they turn up in your Top Stories, AI Overviews and AI Mode. To be exact about what this is: it changes what you see, not what anyone else sees. It is not a ranking signal and it does not move us up for people who haven’t chosen us — anyone telling you otherwise has misread the announcement. One click, undone in the same place.